realist theory building for social epidemiology building

REALIST THEORY BUILDING FOR SOCIAL EPIDEMIOLOGY BUILDING A THEORETICAL MODEL OF NEIGHBOURHOOD CONTEXT AND THE DEVELOPMENTAL ORIGINS OF HEALTH AND DISEASE USING POSTNATAL DEPRESSION AS A CASE STUDY

John Eastwood

A thesis submitted in fulfilment of the requirements for the degree of Doctor of Philosophy October 2010

The School of Public Health and Community Medicine, The School of Women’s and Children’s Health, Centres for Primary Health Care and Equity University of New South Wales

Declaration

Originality Statement

Copyright Statement

i

Authenticity Statement

ii

Acknowledgements

Acknowledgements

The author would like to acknowledge the contribution of all the Child and Family Health nurses of the previous South Western Sydney Area Health Service (SWSAHS) for their efforts in the collection and maintenance of the IBIS database. Particular acknowledgements to my Supervisors Professor Bin Jalaludin (Director REMS, Division of Population Health, SSWAHS), Dr Lynn Kemp (Centre for Health Equity Training and Research, UNSW), and Dr Adrian Barnett (Institute of Health and Biomedical Innovation, Queensland University of Technology (QUT)) for assistance with methodology, statistical analyses and research advice.

iii

iv

Table of Contents

Table of Contents Originality Statement

i

Copyright Statement

i

Authenticity Statement

ii

Acknowledgements

iii

Table of Contents

v

List of Tables

xv

List of Figures

xxi

List of Abbreviations

xxxi

Definitions

xxxv

Layout of Chapters

xxxvii

Abstract

xxxix

Part A: Introduction and Methodology

1

Chapter 1: Introduction

3

1.1

Introduction

4

1.2

Social Epidemiology

5

1.2.1 A population perspective ..........................................................................6 1.2.2 The social context of behaviour ................................................................7 1.2.3 Contextual multilevel and spatial analysis ................................................7 1.2.4 A developmental and life-course perspective ...........................................8 1.2.5 General susceptibility to disease ..............................................................9 1.2.6 Conceptual frameworks ..........................................................................10 1.3

Perinatal Depression and Context

13

1.4

Aims and objectives of the thesis

14

1.5

Theory Building

15

1.6

Methodological Approach

17

1.7

Delimitations

18

1.8

Conclusion

19

Chapter 2: Critical Realism, Theory Building and Research Design

21

2.1

Introduction

22

2.2

Epidemiology, Theory and Pluralism

22

2.3

Research Paradigm

24

2.3.1 Interpretive and Constructive Research Traditions ................................24 2.3.2 Realist and Empirical Research Traditions .............................................25

v

Table of Contents

2.4

Critical Realism

28

2.4.1 Critical Realist Ontology ......................................................................... 28 2.4.2 Critical Realist Epistemology .................................................................. 31 2.4.3 Critical Realism and Causal Inference ................................................... 33 2.4.4 Critical Realism and Study Design ......................................................... 35 2.5

Theory Building

36

2.5.1 Introduction............................................................................................. 36 2.5.2 Theory Building Approaches .................................................................. 36 2.5.3 Conceptual Frameworks, Theory and Models ........................................ 40 2.5.4 An Explanatory Theory Building Method ................................................ 41 2.6

Research Aims

43

2.7

Emergent Phase Methodology

44

2.7.1 Introduction............................................................................................. 44 2.7.2 Prior Position and Use of Literature........................................................ 44 2.7.3 Phenomenon Detection and Description ................................................ 46 2.7.4 Theory Generation ................................................................................. 47 2.8

Emergent Phase Research Design

48

2.8.1 Introduction............................................................................................. 48 2.8.2 Study Setting .......................................................................................... 48 2.8.3 Concurrent Triangulated Mixed Method Design ..................................... 50 2.8.4 Integration and Triangulation .................................................................. 51 2.8.5 Overview of Literature Reviews .............................................................. 52 2.8.6 Qualitative Methods ................................................................................ 53 2.8.7 Quantitative Methods ............................................................................. 55

vi

2.9

Theory Construction Phase

57

2.10

Ethical Considerations and Approvals

58

2.11

Conclusion

59

Part B: Individual Level Exploratory Analysis

61

Chapter 3: Perinatal Adversity, Life-course Outcomes and Depression

63

3.1

Introduction

64

3.2

Research Aims

65

3.3

Early Years and Life Course Consequences

66

3.4

Perinatal Depression

69

3.5

Impact of Perinatal Depression

74

3.6

Conclusion

77

Table of Contents

Chapter 4: Qualitative Exploration at the Individual Level

79

4.1

Introduction

80

4.2

Aims

81

4.3

Methods

81

4.3.1 Theoretical Approach .............................................................................81 4.3.2 Focus groups ..........................................................................................82 4.3.3 Key Informant Interviews ........................................................................83 4.3.4 Data Analysis..........................................................................................85 4.3.5 Validation, Credibility and Rigour ...........................................................88 4.3.6 Ethical Matters ........................................................................................89 4.4

Phenomenon and Concepts

89

4.4.1 Impact on Foetus and Infant ...................................................................89 4.4.2 Maternal and Family Level Themes .......................................................96 4.5

Situational Analysis

116

4.5.1 Situational Analysis Mapping................................................................116 4.5.2 Social Worlds and Social Arenas .........................................................119 4.5.3 Expectations and Dreams ....................................................................121 4.5.4 Marginalisation and “being alone” ........................................................122 4.5.5 Loss of Power and Control ...................................................................123 4.5.6 Support and Nurturing ..........................................................................124 4.6

Theory Generation

124

4.7

Conclusion

126

Chapter 5: Exploratory Data Analysis of Individual Level Factors

127

5.1

Introduction

128

5.2

Aims

129

5.3

Methods

130

5.3.1 Theoretical Approaches........................................................................130 5.3.2 Sample and Design ..............................................................................130 5.3.3 The Study Variables and Univariate Analysis .......................................134 5.3.4 Bivariate Exploration.............................................................................140 5.3.5 Exploratory Factor Analysis ..................................................................140 5.3.6 Use of Causal models ..........................................................................142 5.3.7 Logistic Regression ..............................................................................144 5.3.8 Ethics Approval.....................................................................................146 5.4

Results

147

5.4.1 Characteristics of the subjects ..............................................................147

vii

Table of Contents

5.4.2 Prevalence of Postpartum Depression ................................................. 150 5.4.3 Bivariate Analysis ................................................................................. 151 5.4.4 Categorical Principal Component Analysis .......................................... 157 5.4.5 Assumptions of Causal Inference ......................................................... 162 5.4.6 Logistic Regression Results ................................................................. 163 5.6

Theory Generation

171

5.6.1 CatPCA Revisited ................................................................................. 171 5.6.2 CFA Exploratory Specification Search (AMOS 17.0) ........................... 175 5.6.3 Emerging Theory .................................................................................. 178 5.7

Conclusion

180

Part C: Group Level Exploratory Analysis

181

Chapter 6: Qualitative Study at the Group Level

183

6.1

Introduction

184

6.2

Aims

185

6.3

Methods

185

6.4


186

6.4.1 Community-Level Social Support Networks ......................................... 186 6.4.2 Social Capital ....................................................................................... 189 6.4.3 “Depressed” Community ...................................................................... 191 6.4.4 Access to Services at the Group-level .................................................. 194 6.4.5 Big Business ......................................................................................... 196 6.4.6 Supportive Social Policy ....................................................................... 197 6.4.7 Ethnic Segregation or Diversity ............................................................ 199 6.5


203

6.5.1 Situational Analysis Mapping ............................................................... 203 6.5.2 Social Worlds and Social Arenas ......................................................... 206 6.5.3 Social support networks, social cohesion and social capital ................ 208 6.5.4 Services planning and delivery and social policy ................................. 209 6.5.5 Global economy, business and media.................................................. 209 6.6

Theory Generation

210

6.7

Conclusion

212

Chapter 7: Exploratory Data Analysis of Group Level Ecological Factors

viii

213

7.1

Introduction

214

7.2

Aims

215

7.3

Methods - Overview

216

Table of Contents

7.3.1 Theoretical Approaches........................................................................216 7.3.2 Sample and Design ..............................................................................219 7.3.3 Missing Data and Zero Cells ................................................................225 7.3.4 Outcome (Dependent) Variables ..........................................................225 7.3.5 Comparison (Independent) Variables ...................................................226 7.3.6 Spatial Visualisation - Mapping ............................................................230 7.3.7 Bivariate and Correlation Analyses ......................................................234 7.3.8 Exploratory Factor Analysis ..................................................................234 7.3.9 Likelihood OLS Linear Regression .......................................................235 7.3.10 Bayesian Hierarchical Spatial Regression............................................237 7.3.11 Bayesian Hierarchical Multilevel Models ..............................................242 7.3.12 Ethical issues........................................................................................247 7.4

Description of Phenomenon

247

7.4.1 Overview...............................................................................................247 7.4.2 Visualisation .........................................................................................249 7.4.3 Factors and Causal Inference ..............................................................268 7.4.4 Likelihood Linear Regression Results ..................................................270 7.4.5 Spatial Linear Regression (Bayesian) ..................................................272 7.4.6 Spatial Multilevel Logistic Regression ..................................................287 7.5

Theory Generation

314

7.6

Conclusion

324

Part D: Theory Construction Discussion and Recommendations

325

Chapter 8: Theory Construction

327

8.1

Introduction

328

8.2

Aims

329

8.3

Methods

330

8.3.1 Theoretical Approaches........................................................................330 8.3.2 Conceptual Frameworks.......................................................................331 8.3.3 Stratified Levels ....................................................................................332 8.3.4 Analytic resolution ................................................................................333 8.3.5 Abductive Reasoning............................................................................335 8.3.6 Comparative Analysis (triangulation) ....................................................336 8.3.7 Retroduction .........................................................................................337 8.3.8 Postulates and Propositions Development ...........................................338 8.3.9 Comparison and Assessment of Theories ............................................340 8.3.10 Modelling ..............................................................................................343

ix

Table of Contents

8.4

Setting the Scene

344

8.4.1 Tentative Conceptual Framework......................................................... 344 8.4.2 Analytical Levels................................................................................... 345 8.4.3 Abductive Theoretical Interpretations ................................................... 346 8.5

Stress Process Theory

347

8.5.1 Introduction........................................................................................... 347 8.5.2 Postulate .............................................................................................. 347 8.5.3 Comparative Analysis ........................................................................... 348 8.5.4 Propositions .......................................................................................... 351 8.6

Social Isolation Theory

353

8.6.1 Introduction........................................................................................... 353 8.6.2 Postulate .............................................................................................. 353 8.6.3 Comparative Analysis ........................................................................... 353 8.6.4 Proposition ........................................................................................... 357 8.7

Social Exclusion Theory

359

8.7.1 Introduction........................................................................................... 359 8.7.2 Comparative Analysis ........................................................................... 359 8.7.3 Proposition ........................................................................................... 364 8.8

Social Services

366


Social Capital Theory

371


Acculturation Theory

381


Global-Economic Level Mechanisms

392

8.11.1 Introduction........................................................................................... 392 8.11.2 Migration............................................................................................... 392 8.11.3 Media and Advertising .......................................................................... 393 8.11.4 Big Business and Power....................................................................... 394 8.12

x

Review of other Theoretical Frameworks

395

Table of Contents

8.12.1 “Black box” Debate ...............................................................................395 8.12.2 Eco-social Theory .................................................................................395 8.12.3 Health Theories of Inequality ................................................................396 8.12.4 Community Psychology Theories .........................................................397 8.12.5 Theoretical Models of Neighbourhood Effects ......................................399 8.12.6 Perinatal Models ...................................................................................401 8.12.7 Qualitative Studies of Postnatal Depression ........................................404 8.13

Comparison of Theories

405

8.13.1 Introduction ...........................................................................................405 8.13.2 Contribution from Analysis....................................................................405 8.13.3 Contribution from other theories ...........................................................405 8.14 8.14

Revised Conceptual Framework ..........................................................407

Conclusion

Chapter 9: The Thesis, Theoretical Framework, Propositions and Models

408 409

The Thesis

410

9.1

Introduction

410

9.2

Theoretical Framework

410

Preamble ..........................................................................................................410 Critical Realist Meta-Theory .............................................................................411 Developmental Origins of Health and Disease.................................................411 Stress and Depression .....................................................................................412 The Triggering Events ......................................................................................412 Contextual “Stressor” Mechanisms ..................................................................413 Contextual Buffering Mechanisms....................................................................414 Cultural Mechanisms ........................................................................................414 Global-Economic Mechanisms .........................................................................415 Historical and Spatial Context ..........................................................................415 Summary ..........................................................................................................416 9.3

Theoretical Propositions

417

9.3.1 Introduction ...........................................................................................417 9.3.2 Maternal Depression and Outcomes ....................................................418 9.3.4 Psychological Stress and Depression ..................................................419 9.3.5 Social Stress and Buffers .....................................................................420 9.3.6 Migrant Social Stress and Buffers ........................................................422 9.3.7 Cultural Mechanisms ............................................................................423 Inference to Best Explanation

425

xi

Table of Contents

Conclusion Chapter 10: Conclusion, Limitations, and Implications

426 427

10.1

Introduction

428

10.2

Conclusion

428

10.3

Limitations and Strengths

433

10.3.1 Samples ............................................................................................... 433 10.3.2 Analytical Approach.............................................................................. 434 10.3.3 Reflection ............................................................................................. 443 10.4

Implications for Further Research

447

10.4.1 Methodological approaches ................................................................. 447 10.4.2 Future Studies ...................................................................................... 453 10.5

Implications for Policy and Practice

456

10.5.1 Health services ..................................................................................... 456 10.5.2 Interagency services ............................................................................ 458 10.5.3 Urban planning ..................................................................................... 459 10.5.4 Implications for Government Policy ...................................................... 460 10.5.6 Implications for Corporate Business ..................................................... 461 10.6

Final Comment

462

Bibliography

463

List of Appendices

495

Ethics Approval Letters

497

Focus Group Consent and Guide

501

Edinburgh Postnatal Depression Scale

505

Ingleburn Baby Information System (IBIS) Survey

507

Individual Level Imputed Data Analysis

509

Univariate and Bivariate Analysis ..................................................................... 509 Logistic Regression EDS >9 (Missing Values Imputed) ................................... 514 Logistic Regression EDS > 12 (Missing Values Imputed) ................................ 518 Group Level Variables Technical Notes

523

Segregation, Integration and Diversity Measures Used ................................... 523 Index of concentration at the extremes (ICE) ................................................... 524 Group Level Correlation and Bivariate Analysis

525

Bivariate Analysis ............................................................................................. 530 Group Level Likelihood Linear Regression

xii

535

Table of Contents

Group-Level Linear Regression EDS >9 ..........................................................535 Group Level Bayesian CAR Bivariate Analysis

549

Final Spatial Regression Model EDS >9

553

Ecological Exploratory Factor Analysis

567

Mixed Method Evaluation

575

xiii

Table of Contents

xiv

List of Tables

List of Tables

Table 1: Determinants of Health and Layers of Reality ..................................................... 30 Table 2: Key Informants Characteristics ........................................................................... 84 Table 3: Frequency of IBIS cases 2002 to 2003 by Health Service Sector ..................... 132 Table 4: Comparison of EDS Missing Data Cases.......................................................... 133 Table 5: Descriptive Statistics for the EDS Variable ....................................................... 135 Table 6: Comparison (Independent) variables studied ................................................... 137 Table 7: Characteristics of Study Sample, 2002 to 2003 ................................................ 148 Table 8: Prevalence of EDS at First Visit ........................................................................ 150 Table 9: Chi Square exploration of EDS possible associations ...................................... 151 Table 10: Bivariate Logistic Regression, EDS>9, Significant Associations..................... 154 Table 11: Bivariate Logistic Regression, EDS > 12, Significant Associations ................. 156 Table 12: CATPCA – Seven Dimensions/Factors ............................................................ 157 Table 13: Initial Logistic Regression of EDS > 9 ............................................................. 163 Table 14: Forward Stepwise (Conditional) Logistic Regression of EDS >9 .................... 164 Table 15: Parsimonious Preliminary Main Effect Model EDS >9 .................................... 165 Table 16: Final Model EDS >9 ........................................................................................ 165 Table 17: Logistic Regression EDS > 12 ........................................................................ 167 Table 18: Forward Stepwise (Conditional) Logistic Regression of EDS >12 .................. 168 Table 19: Parsimonious Preliminary Main Effect Model EDS >12 .................................. 169

xv

List of Tables

Table 20: Final Model EDS > 12 ..................................................................................... 169 Table 21: CPCA – Infant temperament and Maternal Attachment variables removed..... 172 Table 22: Characteristics of Study Suburbs .................................................................... 222 Table 23: Characteristics of Study Suburbs (Cont.) ........................................................ 223 Table 24: Descriptive Statistics for the Group-level Outcome Variables ......................... 225 Table 25: Independent variables used in the group level analysis .................................. 229 Table 26: Factor Analysis Oblique Rotation Output – Pattern Matrix ............................... 268 Table 27: EDS9 Coefficients ........................................................................................... 270 Table 28: EDS > 12 Forward Regression Coefficients.................................................... 270 Table 29: EDS > 12 Backward Regression Coefficients ................................................. 271 Table 30: EDS > 12 Manual Regression Coefficients ..................................................... 271 Table 31: Comparison of DIC for Models with random effects and no covariates........... 272 Table 32: Comparison of DIC for selected EDS >9 Bivariate models with random effects ........................................................................................................................................ 272 Table 33: EDS >9 best fitting Spatial CAR regression models ....................................... 273 Table 34: EDS >12 best fitting Spatial CAR regression models ..................................... 274 Table 35: Coefficient Monte Carlo Error – EDS > 12 Spatial Regression - BYM ............. 276 Table 36: Odds Ratio – EDS > 12 Spatial Regression - BYM .......................................... 276 Table 37 : Sensitivity to Choice of Hyper-priors on Random Effects – EDS > 12 ............ 277 Table 38: EDS > 12 Logistic Regression Individual and Multilevel Models..................... 287 Table 39: Main Effect Multilevel Models with suburb level fixed effects .......................... 289 Table 40: Born in Australia Logistic Regression Individual and Multilevel Models .......... 290

xvi

List of Tables

Table 41: Born In Australia Multilevel Models with suburb level fixed effects ................. 292 Table 42: Not Born in Australia Logistic Regression Individual and Multilevel Models ... 293 Table 43: Not Born In Australia Multilevel Models with suburb level fixed effects ........... 295 Table 44: Coefficient Monte Carlo Error – EDS > 12 Spatial Multilevel Regression with Factor Six ........................................................................................................................ 297 Table 45: Odds Ratio – EDS > 12 Spatial Multilevel Regression with Factor Six ............ 297 Table 46 : Precision and Standard Deviation on Random Effects – EDS > 12 Multilevel with Factor Six ........................................................................................................................ 298 Table 47 : Sensitivity to Choice of Hyper-priors on Random Effects – EDS > 12 Multilevel with Factor Six................................................................................................................. 298 Table 48: Coefficient Monte Carlo Error – EDS > 12 Spatial Multilevel Regression ........ 305 Table 49: Odds Ratio – EDS > 12 Spatial Multilevel Regression ..................................... 306 Table 50 : Precision and Standard Deviation – EDS > 12 Multilevel with No Support ... 306 Table 51 : Sensitivity to Choice of Hyper-priors on Random Effects – EDS > 12 Multilevel with No Support............................................................................................................... 306 Table 52: Factor Correlation Matrix ................................................................................. 314 Table 53: Factor Analysis Oblique Rotation Output – Structure Matrix ............................ 315 Table 54: Factor Analysis Oblique Rotation Output – Pattern Matrix ............................... 316 Table 55: Analytical Levels in Disability Research .......................................................... 332 Table 56: Analytical Levels of Depression and Context .................................................. 345 Table 57: Stress Proposition - Inference to Best Explanation ......................................... 352 Table 58: Isolation Propositions – Inference to Best Explanation .................................... 358 Table 59: Social Exclusion proposition – Inference to Best Explanation.......................... 365 Table 60: Social Service proposition – Inference to Best Explanation ............................. 370

xvii

List of Tables

Table 61: Social Capital proposition – Inference to Best Explanation .............................. 380 Table 62: Social exclusion proposition – Inference to Best Explanation .......................... 391 Table 63: A Comparison of Empowering Processes and Empowered Outcomes across Levels of Analysis ........................................................................................................... 398 Table 64: Social exclusion proposition – Inference to Best Explanation .......................... 425 Table 65: Quality of the Intensive Research Process ..................................................... 436 Table 66: Bivariate Logistic Regression, EDS > 9 Imputed, Significant Associations..... 510 Table 67: Bivariate Logistic Regression, EDS > 12 Imputed, Significant Associations... 512 Table 68: Logistic Regression EDS >9 Imputed Data ..................................................... 514 Table 69: Forward Stepwise (Conditional) Logistic Regression of EDS >9 (Imputed) .... 515 Table 70: Parsimonious Preliminary Main Effect Model EDS >9 (Imputed Dataset) ...... 516 Table 71: Final Model EDS >9 (Imputed Dataset) .......................................................... 516 Table 72: Logistic Regression EDS > 12 (Missing Values Imputed) ............................... 518 Table 73: Forward Stepwise (Conditional) Logistic Regression of EDS >12 .................. 519 Table 74: Parsimonious Preliminary Main Effect Model EDS >12 (Missing Values Imputed) ........................................................................................................................................ 520 Table 75: Final Model EDS >12 (Missing Values Imputed)............................................. 520 Table 76: Ecological Correlation Matrix .......................................................................... 525 Table 77: Bivariate Linear Regression, EDS>9, descending R....................................... 530 Table 78: Bivariate Linear Regression, EDS > 9 Imputed, descending R ....................... 531 Table 79: Bivariate Linear Regression, EDS > 12, descending R ................................... 532 Table 80: Bivariate Linear Regression, EDS > 12 Imputed, descending R ..................... 533 Table 81: EDS9 Coefficients ........................................................................................... 535

xviii

List of Tables

Table 82: ANOVA for Model 9 of the Backward Regression – EDS9% ........................... 536 Table 83: Model Summary and Durbin Watson for Backward Regression - EDS9%...... 536 Table 84: Residuals Statistics ......................................................................................... 537 Table 85: Casewise Diagnostics for (+-3 SD) Backward Regression – EDS9% .............. 537 Table 86: Casewise Diagnostics for (+-2 SD) Backward Regression – EDS9% .............. 538 Table 87: EDS > 12 Coefficients ..................................................................................... 542 Table 88: ANOVA of Model 7 of the Backward Regression – EDS12%........................... 543 Table 89: Model Summary for Backward Regression ..................................................... 543 Table 90: Residuals Statistics backward regression – EDS12% ..................................... 544 Table 91: Casewise Diagnostics at + or – 2 Standard Deviations.................................... 544 Table 92: Bivariate spatial CAR models for percent EDS >9 .......................................... 549 Table 93: Bivariate spatial CAR models for percent EDS >12 ........................................ 550 Table 94: EDS >9 Spatial regression with %No Support and second covariate. ............ 551 Table 95: Coefficient Monte Carlo Errors – EDS > 9 Spatial Regression – BYM.............. 554 Table 96: Odds Ratio – EDS > 9 Spatial Regression – BYM ............................................ 554 Table 97: Sensitivity to Choice of Hyper-priors on Random Effects – EDS > 9 ............... 555 Table 98: Kaiser-Meyer-Olkin and Bartlett’s Tests .......................................................... 567 Table 99: Factor Analysis Communalities ....................................................................... 568 Table 100: Factor Analysis Total Variance Explained ..................................................... 569 Table 101: Factor Analysis Oblique Rotation Output – Structure Matrix .......................... 571 Table 102: Factor Analysis Oblique Rotation Output – Pattern Matrix ............................. 572 Table 103: Factor Correlation Matrix............................................................................... 573

xix

List of Tables

Table 104: Assessment of Integration - 1 ....................................................................... 576 Table 105: Assessment of Integration - 2 ....................................................................... 577

xx

List of Figures

List of Figures

Figure 1: Multilevel structure of repeated measurement of individuals over time across neighbourhoods .................................................................................................................. 8 Figure 2: Multigenerational schema of possible influences of hierarchical and life course exposures on disease risk ................................................................................................... 9 Figure 3: A layered model [framework] of the socio-economic determinants of health ..... 10 Figure 4: Conceptual Framework for Understanding Social Inequalities in Health and Ageing ............................................................................................................................... 11 Figure 5: Clarke’s Situational Matrix ................................................................................. 12 Figure 6: Research map outline ........................................................................................ 16 Figure 7: Positivist or 'successionist' view of causation .................................................... 33 Figure 8: Critical realist view of causation ......................................................................... 34 Figure 9: Continuum of frameworks, theories and model.................................................. 40 Figure 10: Phase of Explanatory Theory Building ............................................................. 42 Figure 11: Proposed Perinatal Conceptual Framework .................................................... 45 Figure 12: South West Sydney as a Deprived Region of Sydney ..................................... 49 Figure 13: Concurrent Mixed Method Design.................................................................... 50 Figure 14: Focused Network of Foetal Stress, Damage and Biological Programming...... 90 Figure 15: Focused Network of poor attachment and nurturing ........................................ 92 Figure 16: Focused Network of Infant Temperament ........................................................ 94 Figure 17: Focused Network of Unplanned Pregnancy..................................................... 96 Figure 18: Focused network of “Support” .......................................................................... 99

xxi

List of Figures

Figure 19: Focused Network of Access .......................................................................... 103 Figure 20: Focused Network of Loss of Control .............................................................. 105 Figure 21: Focused Network of Isolation......................................................................... 107 Figure 22: Focused network of "stress" .......................................................................... 110 Figure 23: Focused network of Class.............................................................................. 113 Figure 24: Focused network of Poverty .......................................................................... 114 Figure 25: Ordered Situational Map: Maternal Depression, Home and Neighbourhood . 118 Figure 26: Social worlds/Arenas Map of Mothers Home and Neighbourhood ................ 119 Figure 27: Emerging Conceptual Framework of Maternal Stress ................................... 125 Figure 28: Conceptual Model of Mothers Mood .............................................................. 125 Figure 29: Distribution of the EDS Variable .................................................................... 135 Figure 30: Shepard (Scree) plot of the first seven dimensions ....................................... 158 Figure 31: Bi Plot of CPCA – Dimension 1 and 2 ............................................................. 159 Figure 32: Bi Plot of CPCA – Dimension 3 and 4 ............................................................. 160 Figure 33: Bi Plot of CPCA – Dimension 5 and 6 ............................................................. 161 Figure 34: DAG for Attachment, Breastfeeding, Smoking ............................................... 162 Figure 35: Deviance versus Predicted Probability EDS >9 ............................................. 166 Figure 36: Cooks Distances versus Predicted Probabilities EDS >9 ............................. 166 Figure 37: Deviance versus Predicted Probability EDS >12 ........................................... 170 Figure 38: Cooks Distances versus Predicted Probabilities EDS >12 ........................... 170 Figure 39: Bi Plot of CPCA – Dimension 1 and 2 with infant related variables removed . 173 Figure 40: Bi Plot of CPCA – Dimension 3 and 4 with infant related variables removed . 174

xxii

List of Figures

Figure 41: Three Factor Specification Search – Two best models................................... 176 Figure 42: Plot of Best Fit for AIC Four Factor Measurement Model Specification Search ........................................................................................................................................ 176 Figure 43: Four Factor Specification Search – Two best models ..................................... 177 Figure 44: Emerging Causal Inference ........................................................................... 179 Figure 45: Focused Network of Social Support ............................................................... 187 Figure 46: Focused Network of Social Capital ................................................................ 190 Figure 47: Focused Network of “depressed” community ................................................. 192 Figure 48: Focused Network of Access........................................................................... 194 Figure 49: Focused Network of Big Business ................................................................. 196 Figure 50: Focused Network of Supportive Social Policy................................................ 198 Figure 51: Focused Network of Supportive Social Policy................................................ 200 Figure 52: Ordered Situational Map: Maternal Depression, Community ......................... 205 Figure 53: Social worlds/Arenas Map of Mothers Home and Neighbourhood ................ 206 Figure 54: Emerging Conceptual Framework of Maternal Stress ................................... 211 Figure 55: Conceptual Model of Mothers Mood .............................................................. 211 Figure 56: South West Sydney Local Government Areas and Suburbs - 2001 Census . 220 Figure 57: Birth place of mothers .................................................................................... 223 Figure 58: South West Sydney 2001 Census Suburbs ................................................... 224 Figure 59: Win Bugs Code for Spatial Smoothing ........................................................... 232 Figure 60: WinBUGS code for spatial regression ........................................................... 239 Figure 61: Process of data and relative risk decomposition, from Haining (2003) .......... 242

xxiii

List of Figures

Figure 62: WinBUGS code for the multilevel model with CAR ........................................ 245 Figure 63: Standardised Morbidity Ratio (a) EDS >9 (b) EDS > 12 ................................ 249 Figure 64: Bayesian Smoothed RR (a) EDS >9 (b) EDS > 12 ........................................ 250 Figure 65: Spatial scan statistic clusters (a) EDS >9 (b) EDS > 12 ................................ 251 Figure 66: Index of Relative Social Disadvantage (a) Score (b) NSW Decile ................. 252 Figure 67: (a) Family Poverty (b) Family Affluence ......................................................... 253 Figure 68: (a) Index of Concentrated Extremes (b) Unemployment Rate ....................... 254 Figure 69: Social Class (a) Low Social Class (b) High Social Class ............................... 255 Figure 70: Population distribution (a) Violent Crime (b) Vacant Dwellings ...................... 256 Figure 71: (a) Population Density (b) One Parent Family ............................................... 257 Figure 72: Dwelling Type (a) Apartments (b) Single Houses .......................................... 258 Figure 73: Home ownership (a) Owned (b) Rented ........................................................ 259 Figure 74: Mothers self-reported rental accommodation ................................................ 259 Figure 75: Different Address (a) 1 year previous (b) 5 year previous.............................. 260 Figure 76: (a) No Regret Leaving (b) No Support ........................................................... 261 Figure 77: Schooling less than Year 8 ............................................................................ 262 Figure 78: (a) Percent Not Volunteer (b) Percent Volunteer ........................................... 263 Figure 79: Entropy (a) Log Entropy Index (b) Entropy Index ........................................... 264 Figure 80: Entropy (a) Simpson Index of Segregation (b) Maly Index of Diversity .......... 264 Figure 81: Nurse Access (a) First home visits (b) Nurse visits per infant in first year ..... 265 Figure 82: (a) Mothers Poor Self-rate health (b) Unplanned Pregnancy ......................... 266 Figure 83: Suburb health behaviours (a) Not breastfeeding (b) Mothers smoking .......... 267

xxiv

List of Figures

Figure 84: WinBUGS Code for EDS >12 Spatial Regression ......................................... 275 Figure 85: Density Plot of the Posterior Distribution – EDS >12 Spatial Regression ....... 277 Figure 86: Trace Plots and sample history plot – EDS > 12 Spatial Regression ............. 278 Figure 87: Autocorrelation Graphs – EDS > 12 Spatial Regression................................. 278 Figure 88: Gelman-Rubin Convergence Plot – EDS >12 Spatial Regression .................. 279 Figure 89: Decomposition of Final Model EDS >12 Bayesian Regression ..................... 280 Figure 90: Map of the Relative Risk of EDS > 12 – Final Model (BYM) ........................... 281 Figure 91: Contribution by covariate “%No Support” – EDS >12 Final Model (BYM) ...... 282 Figure 92: Contribution by covariate Entropy Index – EDS > 12 Final Model (BYM) ....... 283 Figure 93: Contribution by covariate “%No Regret” – EDS >12 Final Model (BYM) ........ 284 Figure 94: Contribution by spatially unexplained components – EDS > 12 (BYM) .......... 285 Figure 95: Comtribution by unstructured unexplained components – EDS >12 (BYM) ... 286 Figure 96: Bayesian CAR Models ................................................................................... 288 Figure 97: Bayesian CAR Models – Australian Born Mothers.......................................... 291 Figure 98: Bayesian CAR Models – Mothers Not Born in Australia ................................. 294 Figure 99: WinBUGS Code for EDS >12 Spatial Multilevel Regression ......................... 296 Figure 100: Density Plot of the Posterior Distribution – EDS >12 Multilevel Regression – Factor Six ........................................................................................................................ 299 Figure 101: Sample Trace and History Plots the Posterior Distribution – EDS >12 Multilevel Regression – Factor Six ................................................................................... 299 Figure 102: Sample Autocorrelation Graphs – EDS > 12 Multilevel Regression ............. 300 Figure 103: Sample Gelman-Rubin Convergence Plots – EDS >12 Multilevel Regression ........................................................................................................................................ 300

xxv

List of Figures

Figure 104: Decomposition of EDS > 12 Multilevel Regression with Factor Six as Suburb level fixed effect .............................................................................................................. 301 Figure 105: Map of the Relative Risk of EDS >12 – Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect............................................... 302 Figure 106: Contribution of Factor Six - EDS >12 Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect............................................... 303 Figure 107: Contribution of spatially unexplained components - EDS >12 – Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect .................. 304 Figure 108: Density Plot of the Posterior Distribution – EDS >12 Multilevel Regression – No Support ........................................................................................................................... 307 Figure 109: Sample Density and History Plot of the Posterior Distribution – EDS >12 Multilevel Regression – No Support ................................................................................. 308 Figure 110: Sample Autocorrelation Graph – EDS > 12 Multilevel Regression ............... 308 Figure 111: Gelman-Rubin Convergence Plot – EDS >12 Multilevel Regression............ 308 Figure 112: Decomposition of EDS > 12 Multilevel Regression with “No Support” as Suburb level fixed effect .................................................................................................. 310 Figure 113: Map of the Relative Risk of EDS >12 – Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect .......................................... 311 Figure 114: Contribution of Factor Six - EDS >12 Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect .......................................... 312 Figure 115: Contribution of spatially unexplained components - EDS >12 – Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect .............. 313 Figure 116: (a) Factor One (b) Factor One-Blue ............................................................. 317 Figure 117: (a) Factor Two Red (b) Factor Two Green ................................................... 319 Figure 118: (a) Factor Three Red (b) Factor Three Green .............................................. 320 Figure 119: (a) Factor Four Red (b) Factor Four Green .................................................. 321

xxvi

List of Figures

Figure 120: (a) Factor Five Red (b) Factor Five Green ................................................... 322 Figure 121: (a) Factor Six Red (b) Factor Six Green....................................................... 323 Figure 122: Graphical representation of critical realist MCO propositions ...................... 339 Figure 123: Tentative Conceptual Framework ................................................................ 344 Figure 124: Modified Stress Process Model ................................................................... 350 Figure 125: Critical Realist Model of the Stress Proposition ........................................... 351 Figure 126: Critical Realist Model of Social Isolation Proposition ................................... 357 Figure 127: Critical Realist Model of Social Exclusion Proposition ................................. 364 Figure 128: Critical Realist Model of Social Service Proposition .................................... 369 Figure 129: Conceptual model of neighbourhood social capital processes .................... 372 Figure 130: Critical Realist Model of Social Capital Proposition ..................................... 379 Figure 131: A framework of acculturation research (Berry 1997) .................................... 382 Figure 132: Critical Realist Model of Acculturation & Ethnic Integration Propositions .... 390 Figure 133: Perinatal health framework .......................................................................... 401 Figure 134:Conceptual Framework of Neighbourhood Influence on Pregnancy Outcomes ........................................................................................................................................ 402 Figure 135: Proposed major pathways of environmental adversity influencing development ........................................................................................................................................ 403 Figure 136: Conceptual Framework of Maternal Depression, Stress and Context ......... 407 Figure 137: MCO Model of Maternal Depression and Outcomes ................................... 418 Figure 138: Moderated Model of Depression and Responsiveness................................ 418 Figure 139: MCO Model of Psychological Stress and Depression.................................. 419

xxvii

List of Figures

Figure 140: Moderated Mediated Model of Expectation, Depression and Responsiveness ........................................................................................................................................ 419 Figure 141: MCO Model of Social Stress, Psychological Stress and Buffers ................. 420 Figure 142: MCO Model of Social Stress and Social Buffers .......................................... 420 Figure 143: Model of Social Isolation, Stress and Depression ........................................ 421 Figure 144: MCO Migrant Mechanisms .......................................................................... 422 Figure 145: MCO Cultural Mechanisms – Expectations of Mothers ................................ 423 Figure 146: MCO Cultural Mechanisms – Mothers Expectations .................................... 423 Figure 147: Model of Social Isolation and “Expectations Lost” ....................................... 424 Figure 148: Distribution of the imputed EDS Variable ..................................................... 509 Figure 149: Descriptive Statistics for the imputed EDS Variable .................................... 509 Figure 150: Deviance versus Predicted Probability EDS >9 Imputed ............................. 517 Figure 151: Cooks Distances versus Predicted Probabilities EDS >9 Imputed ............. 517 Figure 152: Deviance versus Predicted Probability EDS >12 Imputed ........................... 521 Figure 153: Cooks Distances versus Predicted Probabilities EDS >12 Imputed ........... 521 Figure 154: Histogram of EDS9 Regression Residual .................................................... 538 Figure 155: Normal P-P Plot of Regression Standardised Residual ............................... 539 Figure 156: Scatter Plot of Regression Standardised Residual to Standardised Predicted Value – 95% Confidence Intervals Shown ....................................................................... 539 Figure 157: Scatter Plot of Standardised Residual to Standardised Variable Entropy.... 540 Figure 158: Scatter Plot of Cook’s Distance to Centered Leverage Value ...................... 541 Figure 159: Histogram of EDS9M Regression Residual ................................................. 545 Figure 160: Normal P-P Plot of Regression Standardised Residual ............................... 545

xxviii

List of Figures

Figure 161: Scatter Plot of Regression Standardised Residual to Standardised Predicted Value ............................................................................................................................... 546 Figure 162: Scatter Plot of Standardised Residual to Standardised Variable Entropy .... 546 Figure 163: Scatter Plot of Cook’s Distance to Centered Leverage Value ...................... 547 Figure 164: WinBUGS Code for Final BYM Model.......................................................... 553 Figure 165: Density Plot of the Posterior Distribution – EDS >9 Spatial Regression ....... 555 Figure 166: Trace Plots and sample history plot – EDS > 9 Spatial Regression ............. 556 Figure 167: Autocorrelation Graphs – EDS > 9 Spatial Regression................................. 557 Figure 168: Gelman-Rubin Convergence Plot – EDS >9 Spatial Regression .................. 557 Figure 169: Map of the Relative Risk of EDS > 9 – Final Model (BYM) ........................... 559 Figure 170: Contribution by covariate “%No Support” – EDS >9 Final Model (BYM) ...... 560 Figure 171: Contribution by covariate Entropy Index – EDS > 9 Final Model (BYM) ....... 561 Figure 172: Protective contribution by covariate “%Living in Apartment” – EDS > 9 (BYM) ........................................................................................................................................ 562 Figure 173: Contribution by covariate “%Smoking” – EDS >9 Final Model (BYM)........... 563 Figure 174: Contribution by spatially unexplained components – EDS > 9 (BYM) .......... 564 Figure 175: Composition by unstructured unexplained components – EDS >9 (BYM) .... 565 Figure 176: Factor Analysis Scree Plot ........................................................................... 570

xxix

List of Figures

xxx

Abbreviations

List of Abbreviations

ADHD

Attention Deficit Hyperactivity Disorder

AIC

Akaike Information Criteria

BDI

Brief Symptom Inventory

BYM

Besag, York and Mollie

CALD

Culturally and Linguistically Diverse

CAR

Conditional Autoregression

CatPCA

Categorical Principal Component Analysis

CDA

Confirmatory Data Analysis

CI

Confidence Interval

CI

Confidence Interval

Cinahl

Cumulative Index to Nursing and Allied Health Literature

CSR

Corporate Social Responsibility

DAG

Directed Acyclic Graph

DHEA

Dehydoepiandrosterone

DIC

Deviance Information Criteria

DNA

Decoy-Nucleic Acid

DOHaD

Developmental Origins of Health and Disease

DSM

Diagnostic and Statistical Manual for Mental Disorders

DSM

Diagnostic and Statistical Manual of Mental Disorders

E

Expected

EDA

Exploratory Data Analysis

EDS

Edinburgh Depression Scale

EDS12

Edinburgh Depression Scale score greater than 12

xxxi

Abbreviations

EDS9

Edinburgh Depression Scale score greater than 9

EDSA

Exploratory Data Spatial Analysis

EFA

Exploratory Factor Analysis

Embase

Excerpta Medica Database, is a biomedical and pharmacological database produced by Elsevier

EPDS


GP

General [Medical] Practitioner

HES

Household Economic Survey

HPA

Hypothalamus Pituitary Axis

IBE

Inference to the Best Explanation

IBIS

Ingleburn Baby Information System

ICD

International Classification of Diseases

ICE

Index of Concentrated Extremes

IRSD

Index of Relative and Social Disadvantage

IUGR

Intrauterine Growth Retardation

LGA

Local Government Area

LGA

Local Government Area

MAR

Missing at Random

MC

Monte Carlo [errors]

MCAR

missing completely at random

MCMC

Markov Chain Monte Carlo

MCO

Mechanism, Context, Outcome (a critical realist proposition)

MEDLINE

A National Library of Medicine computer data base of current references

xxxii

NESP

Non-English Speaking People

NHMRC

National Health and Medical Research Council

NS

Not significant

Abbreviations

NSW

New South Wales

O

Observed

OBSTET

Obstetric Database

OLS

Ordinary Least Squares Defined

OR

Odds Ratio

OR

Odds Ration

PCA

Principal Component Analysis

pD

Effective number of parameters in the model

PND

Postnatal/Postpartum Depression

PubMed

A free search engine for accessing the MEDLINE

Qual

Qualitative

Quant

Quantitative

ROC

Receiver Operating Characteristics

RR

Relative Risk

SatSCAN

scan statistics software developed by Martin Kulldorff

SE

Standard Error

SEIFA

Socio-Economics Indexes for Areas

SEM

Structural Equation Modelling

SES

Socio Economic Status

SIDS

Sudden Infant Death Syndrome

SMR

Standardised Mortality (or Morbidity) Ratio

SWS

South Western Sydney

SWSAHS

South Western Sydney Area Health Service

TAFE

Technical and Further Education

TEC

Theory of explanatory coherence

WCBA

Women of Child Bearing Age

xxxii i

Abbreviations

xxxiv

Definitions

Definitions

Antenatal

During pregnancy, before birth

Critical Realism

A philosophical view of science and/or theology which asserts that our knowledge of the world refers to theway-things-really-are, but in a partial fashion which will necessarily be revised as that knowledge develops.

Critical Realism

Critical realism is a philosophical view of knowledge. On the one hand it holds that it is possible to acquire knowledge about the external world as it really is, independently of the human mind or subjectivity. That is why it is called realism. On the other hand it rejects the view of naïve realism that the external world is as it is perceived. Recognizing that perception is a function of, and thus fundamentally marked by, the human mind, it holds that one can only acquire knowledge of the external world by critical reflection on perception and its world. That is why it is called critical.

Induction

(Miller 2003)

Inference

(Miller 2003)

Meta analysis

Systematic review of literature that employs statistical methods to combine and summarise the results of several studies

Perinatal

Around the time of delivery, usually refers to 20 weeks from conception to 28 days after delivery

Postnatal/ postpartum

Occurring after birth, usually within 12 months of delivery

Low birth weight

defined as less than 2500 grams at birth

Preterm birth

defined as children born before 37 weeks of gestation

xxx

Definitions

Small for gestational age

defined as less or equal to tenth percentile of birth weight for gestational age

− Neighbourhood

Neighbourhood refers to a person’s immediate residential environment, which is hypothesized to have both material and social characteristics potentially related to health

Social Capital

Social capital or collective efficacy is defined as level of trust and attachment characterizing neighbourhood residents and their capacity for mutually beneficial action

Induction

Discovery of patterns (Onwuegbuzie and Leech 2006)

Deduction

Testing of theories and hypotheses (Onwuegbuzie and Leech 2006)

Abduction

Uncovering and relying on the best of a set of explanations for understanding one’s results (Onwuegbuzie and Leech 2006)

Transfactual conditionals

Transfactual conditionals state what is actually occurring here and now, but because the consequent may be unrealised, the consequent may be nonempirical, or against the empirical fact.(Fleetwood 2010)

Counterfactual

Counterfactual conditionals, state, or enquire about,

conditionals

what might have occurred, had conditions been different (Fleetwood 2010).

xxxvi

Layout

Layout of Chapters Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis


Chapter 3: Perinatal Adversity, Lifecourse Outcomes and Depression

Chapter 6: Qualitative Study at the Group Level Chapter 7: Exploratory Data Analysis of Group Level Ecological Factors

Chapter 4: Qualitative Exploration at the Individual Level Chapter 5: Exploratory Data Analysis of Individual Level Factors


Part D: Theory Construction, Discussion and Recommendations Chapter 8: Theory Construction Chapter 9: The Thesis, Theoretical Framework, Propositions and Models Chapter 10: Conclusion, Limitations and Implications

xxx ii

xxxvii i

Abstract

Abstract Background and Aims: It is increasingly recognised that major adult public health issues, related to development, behaviour and lifestyle have their origins during pregnancy, infancy and early childhood. The aim of this study is to utilise mixed methodology to build a conceptual framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression. Methodology and Methods: I used a critical realist approach to social epidemiology theory building. Emergent methods included: key informant interviews, focus groups, thematic analysis, conceptual mapping, situational analysis, factor analysis, logistic, linear, and Bayesian spatial and multilevel regression studies. Explanatory theory building utilised abductive Inference to

the Best Explanation. Results: Theoretical concepts emerging included: loss of expectation, marginalisation, loss of control, nurturing and support, social support networks, access to services, ethnic migration and the role of global economy, business and media. Multilevel spatial studies suggest that strong ecological social networks increase depression among migrant mothers but not Australian mothers. Discussion The study found accumulating evidence that maternal stress, during and after pregnancy, is a cause of maternal depression and altered developmental trajectory of her infant. Emerging was the centrality of expectation lost as a possible trigger of stress and depression. Global, economic, social and cultural mechanisms were identified that explain maternal stress and depression within family and neighbourhood contexts. The challenge for policy and practice is to support mothers and their partners during the transition to parenthood. The Thesis In the neighbourhood spatial context, in keeping with critical realist ontology, globaleconomic, social and cultural level generative powers trigger and condition maternal psychological and biological level stress mechanisms resulting in the phenomenon of maternal depression and alteration of the infants’ developmental trajectory.

xxxix

xl

Part A: Introduction and Methodology

Part A: Introduction and Methodology Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







1

2


Chapter 1: Introduction Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







3


1.1

Introduction

Social epidemiology is a branch of epidemiology that studies the social distribution and social determinants of states of health. The focus is on specific social phenomenon such as socio-economic stratification, social networks and support, discrimination, work demands and control. Studies of the influence of these contextual factors on health outcomes often analyse individual and community level characteristics separately. This has in part been due to a lack of available data on the contexts of study subjects, lack of easily accessible statistical methods and software for the analysis of complex multi-level data and lack of appropriate theory that explicitly acknowledges the mechanisms by which contexts are related to health outcomes. Recent developments now allow multi-level, temporal and spatial characteristics to be studied simultaneously. It is increasingly recognised that major adult public health issues, related to development, behaviour and lifestyle, such as suicide, depression, violence, crime, school failure, unemployment, injury, unintended pregnancy, substance abuse, and obesity have their origins during pregnancy, infancy and early childhood. Neuro-developmental science has made dramatic advances in assisting us to understand how early influences impact on brain development and subsequent risk or resilience to poor outcomes (Barker 1992; Shonkoff and Meisels 2000). Pregnancy and the initial postpartum months are of special significance for parents in developing a nurturing and responsive relationship with their infant (Barnett and Fowler, 1995). Untreated perinatal depression that interrupts the development of secure and mutually satisfying mother-infant attachment is associated with detrimental effects on cognitive, emotional, social and behavioural development of the infant with both short and long term impacts (Murray and Cooper, 1996). Little is known, however, of the social epidemiology of perinatal depression or its theoretical underpinnings. A recent criticism of social epidemiological studies, and multi-level studies in particular has been a paucity of theory. In this study I will build a theory of the social epidemiology of perinatal depression by exploring the individual and ecological-level causal concepts related to perinatal depression. As theory building involves inductive, abductive, retroductive and deductive processes I will use a critical realist approach which is transdisciplinary, encompassing both quantitative and qualitative traditions, and assumes both ontological and hierarchical stratification of reality (Danermark 2002).

4


The study will thus contribute to methodological approaches to theory building, perinatal and infant social epidemiology and our theoretical understanding of how economic, social, physical and political factors might influence developmental and life-course outcomes.

1.2

Social Epidemiology

Epidemiology is the study of the distribution and determinants of states of health in population (Susser 1973). Oakes and Kaufaman further define social epidemiology as the “branch of epidemiology that considers how social interactions and collective human activities affect health. In other words, social epidemiology is about how a society’s innumerable social arrangements, past and present yield differential exposures and thus differences in health outcomes among the persons who comprise the population” (Oakes and Kaufman 2006, p3). Berkman and Kawachi in the first systematic account of social epidemiology (2000, p6) define social epidemiology as the branch of epidemiology that studies the social distribution and social determinants of states of health. “Defining the field in this way implies that we aim to identify socio-environmental exposures that may be related to a broad range of physical and mental health outcomes … We focus on specific social phenomena such as socioeconomic stratification, social networks and support, discrimination, work demands, and control rather than on specific disease outcomes.”

5


Several guiding concepts in social epidemiology are introduced by Berkman and Kawachi as “useful and sometimes challenging guides that transcend the study of any single exposure.” They are: 1. A population perspective 2. The social context of behaviour 3. Contextual multilevel analysis 4. A developmental and life-course perspective 5. General susceptibility to disease. These guiding concepts underpin this study and are introduced here to situate this study within the discipline of social epidemiology.

1.2.1 A population perspective Rose (1985; 1992) provided a population perspective whereby an individual’s risk of illness cannot be considered in isolation from the disease risk of the population to which she belongs. The views put forward by Rose were controversial (Charlton 1995; Ebrahim and Lau 2001; McCormick 2001; Schwartz and Diez Roux 2001). Rose’s contention that the key to understanding incidence and prevalence lies in “characteristics of populations and not individuals” was considered a startling claim (Charlton 1995). This issue became a central theme in the ‘epidemiology wars’ (Poole and Rothman 1998; McMichael 1999). Poole and Rothman found balance in the concept of eco-epidemiology proposed by Susser (1998) which encompassed the micro, the individual and the macro levels. The crucial implication of Rose’s theory for social epidemiology is that we must incorporate the social context into explanations about why some people stay healthy while others get sick (Berkman and Kawachi 2000). The theoretical matters central to the above debate are also central to the realist philosophy and will be discussed further in Chapter 2: Critical Realism, Theory Building and Research Design.

6


1.2.2 The social context of behaviour Understanding why “poor people behave poorly” (Lynch, Kaplan et al. 1997) requires a shift in understanding – specific behaviours once thought of as falling exclusively within the realm of individual choice occur in a social context. Social environments place constraints on individual choice by: 1. “Shaping norms 2. Enforcing patterns of social control 3. Providing or not providing environmental opportunities to engage in certain behaviours 4. Reducing or producing stress for which certain behaviours may be an effective coping strategy”. (Berkman and Kawachi 2000, p7) Earls and Carlson (2001) contend that “ to grow up in a neighbourhood in which a high proportion of families live in poverty is to experience a context in which demeaning and threatening encounters are qualitatively different from those experienced by a child raised in neighbourhoods in which the majority of families are economically secure.” Earls and Carlson (2001) go on to cite Leeder and Dominello (1999) and Labonte (1999) as drawing attention to the potential of “social capital” as a conceptual and analytical strategy to raise consciousness about social causation in public health research. Social capital and the related concepts of social networks and social cohesion are dealt with in more detail in Chapters 8.

1.2.3 Contextual multilevel and spatial analysis Susser (1998) noted that “risk factor epidemiology in its pure form exploits neither the depth and precision of micro-levels nor the breadth and compass of macro-levels”. Ecological analysis, a central part of both epidemiology and sociology early in the last century, offered an approach to the study of environments, but lost a great deal of respectability because of problems related to the ecological fallacy.1 In recent years innovative statistical multi-level methods have been developed that have given major impetus for examining the role of contexts in explaining health variations (Blakely and Subramanian 2006). Figure 1 illustrates how those same statistical methods can also be utilised to analyse repeated measures of individuals over time across neighbourhoods 1

The drawing of individual inferences from grouped data.

7


with individuals having multiple membership of different neighbourhoods across the time span (Subramanian 2004).

Source: Subramanian(2004, p 1964)

Figure 1: Multilevel structure of repeated measurement of individuals over time across neighbourhoods Early multi-level techniques did not include any specification of possible spatial effects. Individuals may live in different neighbourhoods but be geographically close because they live either side of a neighbourhood boundary. In addition contextual effects such as crime and economic processes do not follow neighbourhood boundaries. Haining (2003, p 365378) proposed the use of multi-level methods that recognize the locations of the neighbourhoods relative to each other. Such an approach is possible using multi-level spatial models that incorporate the use of adjacency matrices and conditional autocorrelation (CAR). Exploratory multilevel and spatial analysis are utilised in Chapters 7 to explore the hierarchal and spatial relationship of possible causative mechanisms.

1.2.4 A developmental and life-course perspective Three hypotheses have been proposed to explain early life influences on the onset of disease in middle and late life (Power and Hertzman 1997). The first is that some exposure in early childhood could influence developmental processes – particularly brain development during periods of great plasticity (Barker 1992). The second hypothesis is one of cumulative disadvantage and is outlined by several medical sociologists (Ross and Wu 1995) and the third is that while early experiences set the stage for adult experiences, it is really only the adult experiences that are directly related to health outcomes. While not supporting one or other of the hypotheses, Berkman and Kawachi (2000) suggest that

8


a developmental and life-course perspective provides a lens through which to examine how social factors may influence adult health.

Source: Ben-Shlomo and Kuh (2002, p 289)

Figure 2: Multigenerational schema of possible influences of hierarchical and life course exposures on disease risk Ben-Shlomo and Kuh (2002, p 289) argue that a life course approach is not limited to individuals within a single generation but “should intertwine biological and social transmission of risk across generations”. They extend the model proposed by Hertzman and others (2001) to link across generations and to include the potential role of household, neighbourhood and national influences acting across time and individuals. The importance of the developmental and life-course perspective is explored further in Chapter 3: Perinatal Adversity, Life-course Outcomes and Depression.

1.2.5 General susceptibility to disease Syme and Berkman (1976) speculated that social factors influence disease processes by creating a vulnerability or susceptibility to disease in general rather than to any specific disorder. The concept was not well grounded biologically until research in social epidemiology was more integrated with research in neuroscience and

9


psychoneuroimmunology and clear biological mechanisms were defined as potential pathways leading from stressful social experiences to poor health. Neuro-endocrinologists have identified classic stress mediators such as cortisol and catecholamines as well as less well understood mediators such as Dehydoepiandrosterone (DHEA), prolactin, and growth hormone, which could influence multiple physiological systems including wide-ranging end organ damage (Berkman and Kawachi 2000). This new understanding provides biological plausibility for the impact of stress on the early developing infant (Welberg and Secki 2001), and for the impact of stress over a life-time on the ageing process (Berkman 1998). The role of stress and related biological pathways are examined further in Chapter 3: Perinatal Adversity, Life-course Outcomes and Depression.

1.2.6 Conceptual frameworks Conceptual frameworks form the basis for the development and communication of theories. There are many conceptual frameworks that have been developed in the fields of population health, social epidemiology, and human ecology. Perhaps one of the most well known is that proposed by Dalgren and Whitehead (1991).

Source: Dahlgren and Whitehead (1991)

Figure 3: A layered model [framework] of the socio-economic determinants of health

10


Kaplan presents a number of other models [frameworks] used to illustrate recent social epidemiologic approaches to understanding the social determinants of health. Figure 4 is taken from Lynch and colleagues (2000). He notes that these [frameworks] have many features in common, the most prominent being an emphasis on layered, multilevel understandings; a multiplicity of pathways; and possibilities for reciprocal influences. He argues that “they act as an important caution against the potential misleading oversimplification that comes from focusing on one level of influence, often the one most proximal to the outcome…” (Kaplan 2004, p 125).

Source: Lynch and colleagues (2000)

Figure 4: Conceptual Framework for Understanding Social Inequalities in Health and Ageing

11


In contrast to the above layered conceptual frameworks, Clark in her work on Situational

Analysis argues that “everything in the situation both constitutes and affects most everything else in the situation in some way … People and things, humans and nonhumans, fields of practice, discourses, disciplinary and other regimes/formations, symbols, controversies, organizations and institutions, each and all can be present and mutually consequential. Here the macro/meso/micro distinctions dissolve in the presence/absence” (Clarke 2005, p 72).

Source: Clarke (2005, p 73)

Figure 5: Clarke’s Situational Matrix

12


1.3

Perinatal Depression and Context

As discussed above there have been a number of theoretical postulates for the possible link between perinatal adversity and developmental origins of health and disease. Baker (1998), and others, have hypothesised that nutritional and other environmental insults during pregnancy and later early life produce lasting organ, metabolic and endocrine abnormalities. The discovery that stress mediators such as cortisol and catecholamines may play a role has thrown a focus on stress as a possible causative pathway. Studies of perinatal depression have consistently found lack of social support and stressful life events to be individual level risk factors and Beach and colleagues (2005) have proposed that depression during pregnancy may impact on foetal development through biological and maternal related behavioural mechanisms. The strong association between stress, social support, and depression calls for the role of these factors in the developmental origins of health and disease to be further examined. Well designed longitudinal studies and randomised intervention studies will ultimately be required. The impact that context might play will require other techniques. The place of social epidemiology is to examine the role that context and specific social phenomenon such as socio-economic stratification, social networks and support, discrimination, work demands and control might play. Hierarchical theoretical models developed using such approaches may later be confirmed using recently developed tools such as multi-level structural equation modelling of clustered longitudinal studies. The recognised paucity of theoretical work in relation to social epidemiology, and the possible role that perinatal depression might play in the developmental origins of health and disease, requires that a theoretical model be developed prior to undertaking confirmatory multi-level studies. Literature on theory building in relation to epidemiology and social epidemiology is scant. This may in part be a result of the predominantly deductive and quantitative nature of the core disciplines of epidemiology and biostatistics.

13


1.4

Aims and objectives of the thesis

This study will thus seek to develop a theoretical model of neighbourhood context and the developmental origins of health and disease using postnatal depression as a case study. The social phenomenon that social epidemiology seeks to understand, and the inductive, abductive and deductive phases of theory building, will require a transdisciplinary and mixed method approach. The aim is to utilise mixed methodology to build a conceptual framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression in South West Sydney. Specific objectives are: 1. To explore and describe the individual level influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney 2. To explore and describe the neighbourhood and community level economic, social, and physical influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney 3. To explain and conceptualise a framework, theory and model describing the mechanisms by which multilevel factors influence developmental and life course outcomes with a focus on perinatal depression.

14


1.5

Theory Building

There was generally a lack of helpful literature on the process of theory building in the fields of epidemiology, and population health. The exception was in the fields of health education and individual level behaviour change (Green and Kreuter 1999; Nutbeam and Harris 1999; DiClemente, Crosby et al. 2002; Glanz, Rimer et al. 2002). This lack of theory was despite the critical commentary of multi-level epidemiology studies, and social epidemiology in general, that had identified a lack of theoretical models informing methodology (Muntaner 1999; Watt 2002; O'Campo 2003; Carpiano 2006; Carpiano and Daley 2006). Multiple theory building methods have been used to build theories that explain the individual and group level determinants of human health and development. I have elected to use four criteria to identify the best theory building method for this study of ecology and maternal depression. They are that the theory building method: 1

“should be dictated by the nature of the theory building being engaged in, and not by the preferred inquiry methodology of the researcher-theorist” (Lynham 2002)

2

be capable of integrating research knowledge across the philosophical paradigms, values and disciplines covered by the study

3

selected must be capable of identifying or generating empirical indicators and hypotheses that can be tested in epidemiological and qualitative studies

4

be selected for their value in relation to the research question and the ability to create new knowledge (Torraco 2002).

There are essentially three main approaches: inductive, abductive and hypotheticaldeductive. These might also be considered as emergent, explanatory and confirmatory phases of a theory building cycle. The theory building approaches of Strauss and Corbin (1990), and Miles and Huberman (1994) are appropriate and useful for the emergent phase of theory building. The hypothetical-deductive approach of Dubin (1969), Merton (1968) and Shoemaker and colleagues (2004) are useful for subsequent theory testing.

15


None of these theory building approaches are sufficient in themselves to meet the criteria outlined above. In particular, the generation of new knowledge requires abductive and reproductive (see definitions) reasoning (Danermark, Ekstrom et al. 2002; Haig 2005). Layder (1993) discusses the gap between theory-testing and theory-building research approaches together with the related gaps between macro and micro levels of analysis and quantitative and qualitative methods. Layder argues that, compared to middle-range theory and grounded theory, “the realist approach attempts to address this complexity by offering a layered or ‘stratified’ model of society” which includes macro (structural, institutional) phenomenon as well as the more micro phenomena of interaction and behaviour” (Layder 1993, p 7). The Research Map (Figure 6) proposed by Layder is directly relevant to the multilevel nature of this study.

Figure 6 has been removed due to Copyright restrictions.

Figure 6: Research map outline The subsequent writings of Haig (2005) and Danermark and colleagues (2002) describe critical realist approaches to mixed method theory building that include abductive analytical processes together with the more commonly used inductive and deductive approaches. These will be discussed in more detail in Chapter 2: Critical Realism, Theory Building and Research Design and Chapter 8: Theory Construction.

16


1.6

Methodological Approach

As stated above, the social phenomenon that social epidemiology seeks to understand, and the inductive and abductive phases of theory building, requires that a transdisciplinary and mixed method approach be used. It is necessary and appropriate, therefore, that I state my philosophical stance with respect to my metaphysical and ontological (study of conceptions of reality and the nature of being) philosophy and epistemological (theory of knowledge) beliefs concerning knowledge. I consider myself to be of a realist ontological orientation and in this study I have adopted the post-positivist epistemology of critical realism. Thus I believe that there is a discoverable reality that is independent of the human mind and subjectivity which is only understood in a partial fashion and is of necessity revised as knowledge develops. I have elected to use emergent and explanatory methodologies to construct the conceptual framework and theory. In this approach I have set aside, as much as possible, my preconceptions and have allowed the concepts and relationships between concepts to emerge throughout the emergent phase. I will use balanced concurrent quantitative (expansive) and qualitative (intensive) approaches to the theory building with early integration of research processes and comparative analysis of quantitative and qualitative findings throughout the study. The emergent phase will be predominantly inductive and abductive. The quantitative methods used will draw on the exploratory analysis approaches advocated by Tukey (1980) and factor analysis methods as proposed by Haig (2005) for abductive reasoning. The qualitative methods will also be inductive and abductive following approaches described by Mile and Huberman (1994), Glaser and Strauss (1967) and Clarke (2005). The theory construction phase will integrate and compare the findings from the quantitative and qualitative studies. Abductive logic will be used to interpret and recontextualised the identified phenomenon within the emerging conceptual frameworks together with retroductive analysis to identify transfactual conditions, structures and mechanisms (Danermark, Ekstrom et al. 2002). The conceptual theories emerging will be compared and contrasted to seek the best explanation. The methodology and methods for these two phases are described in more detail in Chapters 2 and 8 respectively. The confirmatory and hypothetic-deductive phase of theory building is not part of this study.

17


1.7

Delimitations

I have elected to delimit the study in a number of areas. This was necessary in order to limit the size and scope of the study. Delimitation was as follows: 1. The theory building was delimited to emergent, exploratory, inductive, abductive and explanatory approaches to theory building. Confirmatory theory building and related methods were avoided. As a result theory driven data analysis using path analysis and structural modelling was not used. This limited my ability to examine both moderation and mediation effects and to test the proposed models presented in Chapter 8. 2. Multilevel modelling was delimited to Bayesian spatial and non-spatial hierarchical models. Likelihood multilevel models using software such as MLWin and STATA were not used in this study. This decision was taken so as to limit the size of the Chapter 7 and the overall study. The implications of these delimitations are discussed in Chapters 8 and 10. 3. Emergent and Explanatory Theory building was delimited mainly to the critical realist defined “social level” of reality. Consequently biological, psychological, cultural and global economic strata were not examined in any depth. The interviews and focus groups and multilevel studies were thus intentionally delimited.

18


1.8

Conclusion

The purpose of this Chapter has been to introduce the rationale and context to this study. Social epidemiology studies the social distribution and social determinants of health with a focus on social phenomenon. It is increasingly recognised that developmental and health outcomes have their origins in the early years where perinatal adversity is thought to play a role. There is some speculation that maternal depression and stress may play a role in addition to other factors such as nutrition and lifestyle behaviours. Little is known, however, about how social structures and mechanisms might mediate those early processes and critical commentary has drawn attention to the general lack of theory to inform future social epidemiological studies. There is generally a lack of literature on the process of theory building in the fields of epidemiology, and population health. Social science literature is predominantly focused on two general approaches. The first is inductive theory building and the second hypothetico-deductive theory testing. As a critical realist I have elected to utilise an inductive-abductive-deductive approach that is transdisciplinary in nature and utilises mixed methods. The following Chapter will examine in more detail the philosophical, methodological and research design aspects of this study.

19


20

Chapter 2: Critical Realism

Chapter 2: Critical Realism, Theory Building and Research Design Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







21


2.1

Introduction

This chapter summarises the philosophical, methodological and design aspects of the critical realist theory building approach used in this study. The use of a mixed methodology requires that the format of this Chapter differ from that generally used in reporting epidemiological research design and methods. The philosophical section provides a brief overview of approaches to understanding beliefs and knowledge of reality and a case for the ontological and epistemological positions I have chosen to use. The critical realist process of theory building is expanded followed by an overview of the research design. The methods are only briefly described in this Chapter with further detail provided in subsequent chapters as relevant.

2.2

Epidemiology, Theory and Pluralism

A perhaps controversial but nevertheless true statement is that “difference-making

evidence coming from statistics cannot prove causation in any possible means. … evidence about [both] difference-making and mechanisms is needed” (Russo 2009). Although this matter is seldom addressed in epidemiological literature, the need for evidence of mechanisms in addition to difference-making evidence is made clear in the widely accepted use of criteria for causal inference such as those proposed by Hill (1965). Information on causal mechanisms are used at the theory building and theory testing stages. But such assumptions of mechanisms are not statistical in character and belong to the background knowledge of the epidemiologist who is building or testing a causal theory. It can be further argued that causal modelling ought to be conceived as the modelling of possible causal mechanisms. To proceed with the building of a causal theory it is thus necessary to study knowledge and information regarding both “difference-making and mechanistic evidence” (Russo 2009). Difference-making and mechanistic evidence can be taken from either qualitative or quantitative studies. Counterfactual reasoning or comparative case-studies are both examples of qualitative difference-making. Quantitative randomised controlled trials can (dis)confirm plausible or known mechanisms. The important role that qualitative studies play in the generation of evidence of mechanisms is well established but often criticised by epidemiologists (Muntaner and Gomez 2003; Greenland, Gago-Dominguez et al. 2004).

22


The contribution of social and ecological mechanisms to theories of causation has also been well established and forms the basis for modern social epidemiology (Susser 1994a; Susser 1994b; Susser and Susser 1996; Krieger 2001; Kaplan 2004; Susser 2004). Susser (2004) observed that epidemiologists were increasingly considering multiple levels of causation. He called for different approaches to identify the causal mechanisms at each level including cellular, individual and societal levels. Epidemiology is a science, however, that has traditionally studied only observable phenomenon (i.e. empiricism). The challenge with eco- and social epidemiology is that candidate causal mechanisms are often not observable (i.e. social cohesion, social capital, capitalism) although their effects are. As it turns out this is also the case with most psychological mechanisms (Borsboom 2008). Thus for this study of psychological and social level causes of perinatal depression it will be necessary to study both observable and un-observable phenomenon for the purpose of identifying “difference-making and mechanistic evidence”. A quantitative and qualitative study of individual and social level phenomenon would provide both forms of evidence from which a theory could be built or confirmed. The understanding and explanation of un-observable human and social phenomenon requires the reconceptualising of observable phenomenon from a realist philosophical view point using abstract forms of logic (Bhaskar 1979).

23


2.3

Research Paradigm

It is traditional in mixed quantitative and qualitative research to clearly define the researchers’ basic beliefs. Such an approach is seldom taken in traditional epidemiology where it is assumed that all researchers share a “value free” and empirically objective set of beliefs. The epidemiological paradigm debates of the early 90’s (Pearce and CrawfordBrown 1989; Ng 1991; Susser 1991) concerning Popper’s falsification philosophy suggest that the paradigmatic positions of epidemiologists need to also be defined. A research paradigm defines the basic set of beliefs of the researcher. Guba and Lincoln (1994) state that the basic beliefs that define a particular research paradigm may be summarised by the responses given to three fundamental questions: 1. The ontological question i.e. what is the form and nature of reality 2. The epistemological question i.e. what is the basic belief about knowledge (i.e. what can be known) 3. The methodological question i.e. how can the researcher go about finding out whatever s/he believes can be known. To this can be added axiology which is concerned with the role of values in the inquiry (Denzin and Lincoln 2000). There are three well known epistemological belief systems that influence how research questions are asked and answered. They are: Realism (the belief that reality exists independently of observers), empiricism (a belief that knowledge cannot extend beyond experience) and constructivism (the belief that our knowledge is "constructed", and does not necessarily reflect any external realities) (Guba and Lincoln 2005). These are briefly discussed below followed by a more detailed discussion of critical realism.

2.3.1 Interpretive and Constructive Research Traditions Interpretive and constructive research traditions are relevant to this study given that it will seek to explore phenomena that are contextual, historical, socially and individually defined. Interpretive theorists believe that reality is not objective but subjective (Sarantakos 1998) and that there is no objective knowledge independent of thinking. Reality is viewed as socially and societally embedded and existing in the mind (Grbich 2007). Reality is therefore internally experienced and interpreted, and what we know is a construction of the mind.

24


Following this view, the appropriate way to study human behaviours is by historical analysis, ethnography, through critical and sociological theory and hermeneutics. Within this tradition the individual is unique and significant. Consequently this view would suggest that research should be interpretive and approached naturalistically, adopting a qualitative methodology (Bisman 2002). Interpretive and constructive research traditions have predominantly used qualitative research methodologies but recent developments are seeing the use of mixed qualitative and quantitative methodologies with possible clashes of ontology and epistemological perspectives (Creswell and Tashakkori 2007). Kuhn (1962; 1996) made a significant contribution to epistemological debates in the late 20th Century concerning realism and constructivism. He wrote that scientists embracing different paradigms find themselves literally in different worlds. He argued that paradigms were “incommensurable” with one another. The interpretation of this was that the conflict between qualitative and quantitative research was so fundamental that it was impossible to combine them without violating philosophical principles. Mixed methodologists have subsequently argued that the incompatibility approach is “largely discredited” and have argued for the use of a pragmatic approach that accepts external reality and chooses explanations that best produce the desired outcomes (Tashakkori and Teddlie 2003). Similarly critical realism is suitable for mixed method studies as it endorses, or is compatible with, a wide range of research methods including both ethnographic and quantitative approaches (Sayer 2000).

2.3.2 Realist and Empirical Research Traditions For much of the twentieth century the dominant philosophy of science (and epidemiology) was empiricism and related schools of “logical empiricism” or “logical positivism”. Logical empiricism developed a philosophy of science that combined modern mathematical logic with empiricist epistemology and methods used in the natural sciences (Rosenberg 2000). Logical empiricism (aka positivism) philosophy, together with inductivism, refutationism, and Bayesianism, has had a significant influence on scientific thought and the development of epidemiology methodology (Rothman and Greenland 1998). [Logical] positivism assumes the existence of a “real” apprehendable reality driven by immutable natural laws and mechanisms. It then searches for universal mechanistic rules that are not contextually bound and seeks to verify hypotheses (Carpiano and Daley 2006). Raphael (2006) argues, however, that [logical] positivism limits the focus of epidemiological research to the concrete and observable rather than the theoretical and

25


conceptual. Consequently “even when the focus is expanded to include societal structures, there is a reluctance to consider the political and economic forces that drive the creation and maintenance of those structures” (Raphael 2006). Raphael argues further that “as a result the [epidemiology] research tradition eschews examination of complex factors such as the forces that drive the quality social determinants of health”. These shortcomings of positivist social epidemiology are echoed by Oakes and Kaufman (2006) who state that “we must become comfortable with the realisation that some scientific questions will not be answered best by treating observational data as though they arose from controlled experiments”. Yu (2005) states that the “positivist” tradition is misunderstood by its critics and that it is misguided to continue to associate quantitative methodologies, such as epidemiology, with [logical] positivism which has been rejected by philosophers of science over the last fifty years. He notes that logical positivism, which rejects theoretical constructs and causality and emphasises reductionism, is too restrictive to apply to quantitative methodology, which supports the use of latent constructs, causal inferences and the iterative process of understanding the data and developing constructs. Yu argues that logical positivism be considered as part of a research tradition, rather than as a paradigm as proposed by Kuhn (1962), that now includes epistemologies such as post-positivism, critical realism and critical multiplism. In contrast to logical positivism, the post-positivist epistemology assumes that evidence is always imperfect and fallible. In this way it seeks to address some of the criticisms of logical positivism. One of the supporters of post-positivism and refutationism was Karl Popper (Rothman and Greenland 1998, p18). Here reality is still assumed to exist, but in contrast to positivist assumptions, it is only imperfectly and probabilistically apprehendable. Research entails making claims and then refining or abandoning some of them (using both quantitative and qualitative methods) for more strongly warranted claims. Researchers do not verify a hypothesis (as in positivism), but rather indicate a failure to reject one. This paradigm recognises that findings are contextually bound and thus are not generalisable to all cases and situations (Carpiano and Daley 2006). More recently within post-positivism it has been argued that researchers are biased by their education and life experiences and that their observations are thus value laden and fallible. The ability to know reality with certainty is thus viewed as problematic and this has led some to the modified post-positivist epistemology of critical realism (Grbich 2007).

26


Critical realism, while in the post-positivist tradition, goes further arguing that the social world is an “open system” and that causal powers can produce different outcomes depending on the context or spatial-temporal relationships with other objects. Muntaner (1999) in a critique of causal explanations in social epidemiology argued that the problem lay in the scientific method. Muntaner speculated that:

“A plausible reason for the lack of explanations in social epidemiology is the attachment to an empiricist philosophy that searches for empirical generalizations (e.g., using observations to build models) while avoiding conjectures about the underlying social mechanisms that would help us understand how social systems work… Generation of causal explanations in social epidemiology would require abandoning the Humean notion of causality and adopting a realist philosophy that favours generating social theory in addition to observation. The role of theory is precisely to go beyond ‘what we can observe“ (Muntaner 1999, p 124) I contend that critical realism may be an appropriate meta-theory for the generation of causal explanations in social epidemiology and may provide answers to the criticisms put forward by Muntaner (1999), Krieger (2001), O’Campo (2003), Carpiano and Daley (2006) and Raphael (2006). In the following section I will explore critical realist ontology and epistemology.

27


2.4

Critical Realism

Critical realists perceive that reality consists of unobservable elements beyond our empirical realm, but that they are still reachable by scientific enquiry. The basic ontological understanding of critical realism is that reality exists and that it is possible to conceptualise it and make theories that describe it. Critical realism does not claim a total comprehensive understanding of certain phenomenon and considers all knowledge as fallible. Thus a scientific account of a phenomenon is a partial account of certain aspects of reality that are deliberately chosen and subject to change. Critical realists argue that the empirical methods of the natural sciences are not applicable to scientific study of the social world nor is it necessary to adopt the relativist position of postmodernism. In arguing that social reality can be known, even though the social world is unpredictable and complex, critical realism offers a conception of the real that is fundamentally different from the empirical realism of the natural sciences (McGuire 2006). What makes critical realism critical is the fallibility of knowledge claims. Thus knowledge claims should be continually critiqued, challenged, and revised as both culture and practice shape the lens through which we view the world.

2.4.1 Critical Realist Ontology A central aspect of critical realism ontology is the distinction between three ontological domains: the empirical, the actual and the real. The empirical domain comprises of our experiences of what actually happens (i.e. experiences), and the actual is made of things that happened independently of whether we observed them or not (i.e. events). The last ontological domain, the real, is the deepest level of reality and is constituted by mechanisms with “generative power” (Jeppesen 2005). The Empirical domain is when human actors experience or observe happenings or events. Those experiences or events may be in relation to their body, self, or society. Thus the

Empirical is defined as the domain of experience (Sayer 2000, p12). In this study the Empirical reality consists of observations, interviews, survey results, Census data, and my own Empirical experience. The Empirical does not include the unobserved experiences of mothers and practitioners, or the unobserved households, neighbourhoods and social situations which constitute Actuality. The second domain of reality, the Actual is what is, what happens or what is possible, when or if structures, mechanisms and powers of social, human and biological objects are

28


activated (Sayer 2000). The “actual” events, processes or situation may, or may not be seen or perceived by human actors. The Real domain of reality exists independently of our perception or knowledge of it. It consists of objects that are both natural and social. Critical realism posits that reality is stratified and that it consists of structures, powers, and mechanisms. A second critical realist ontological dimension is that reality is stratified. Reality is assumed to consist of hierarchically ordered levels where a lower level creates the conditions for a higher level. The higher level is not, however, determined by the lower level and has its own “generative mechanisms”. The existence of these level specific generative mechanisms is what constitutes or defines a level (Danermark 2002). Mechanisms, objects and events exist at each different stratum have been variably described as including: physical, chemical, biological, psychological, psychosocial, behavioural, social, cultural and economic (Layder 1993; Bhaskar 1998; Sayer 2000; Danermark 2002; Danermark and Gellerstedt 2004; McGuire 2006). Each stratum is separate and distinct and may interact with the layer above or below to produce new mechanisms, objects and events. It is the existence of such level-specific mechanisms that constitute a level. The ability of mechanisms to combine to create something new is called emergence (Bhaskar 1998; Danermark 2002; McGuire 2006). An implication, however, of this stratification is that it is not possible to reduce the causes of what occurs on one level to those on another level (whether lower or higher) (Danermark 2002). Sayer (2000, p 13) observes that “social phenomena are emergent from biological phenomena which are in turn emergent from chemical and physical strata. Thus the social practice of conversing is dependent on one’s physiological state, including the signals sent and received around our brain cells, but conversing is not reducible to those physiological processes”. A significant aspect of critical realist ontology is that real mechanisms can exist as causal tendencies without necessarily being activated. Thus generative mechanisms, at the level of the real, can exist without manifesting themselves as an event that is observed, experienced or measured at the level of actual. A mechanism at one level may be triggered or blocked by other mechanisms at any other level (cf. mediation and moderation). The following table from McGuire (2006) depicts these two ontological dimensions with examples across the body, the self, and society.

29


Empirical (observable)

Actual (objects/events)

Body

Self

Society

Biological

Individual

Social determinants

determinants,

determinants (e.g.

(e.g. poverty),

Diagnosis or illness,

lifestyle), Experience

Disruption of social

Treatment

and meaning of

participation, Health

illness, Coping

and social costs

Normal and

Cognition, Emotion,

Political, economic,

pathological

Development,

social welfare, health

processes,

Behaviour

care systems, Social

Signs and symptoms

behaviours, norms, relations

Real

Biological, physical,

Psychological,

Social, cultural,

(mechanisms)

chemical, genetic

emotional, cognitive,

political, economic,

mechanisms

spiritual

religious mechanisms

mechanisms Source: McGuire (2006, p 118)

Table 1: Determinants of Health and Layers of Reality The concept of mechanisms is central to critical realist ontology. Mechanisms have the power to produce events and are often described as “generative processes” or as having “generative powers”. The mechanisms can exist beneath the empirical surface and are not directly observable. Based on observed phenomenon the task may be to find the “underlying mechanisms” that produce the “phenomenon and to understand the interplay between them and how they shape the outcome”. Alternatively “one can assume the existence of certain mechanisms and try to find how this mechanism is empirically manifested. This can reveal that the mechanism is not empirically manifest because it is not active or there are other counteracting mechanisms” (Danermark 2002, p 59). The task in explanatory research is to find the mechanisms that produce the actual phenomenon and to understand the interplay between them to produce outcomes or events. Some mechanisms may not produce empirically manifest outcomes because it is either not active or there are counteracting mechanisms (Danermark 2002). This leads us to the importance of context with the outcome of a mechanism always dependent on the context within which it is active. “Generative processes” are thus always contextually determined. The identification of generative mechanisms and counteracting mechanisms requires consideration of context. Critical realism argues that events are produced in an “open

30


system” with highly complex contexts. Only when in “closed systems” is it possible to isolate all mechanisms other than those under study. Even in the natural sciences this is difficult. The outcome of a mechanism when activated is therefore always dependent on context. Thus context determines how a mechanism is empirically manifest (Danermark 2002). Related to the above is tendency. The generative mechanisms have regularity but their empirical manifestation is contingent upon context. Therefore this empirical manifestation must be considered as tendencies rather than empirical regularities. This has implications for explanation of observed phenomenon. Danermark (2002) argues that statistical analyses in highly complex situations are not appropriate if the purpose is find a causal explanation but that they may be fruitful if the purpose is to describe aspects of social reality. In this study I will use statistical analysis to both describe phenomenon and to generate explanatory theory (Haig 2005).

2.4.2 Critical Realist Epistemology As noted above critical realists consider that reality exists independently of us (an intransitive dimension) and not as a social construction. There is a debate about how we get access to understanding this reality with social constructionists denying that such access is possible. Realists acknowledge that our knowledge of reality is influenced by social factors but claim that it is nevertheless possible to develop a reliable knowledge about the external world and the mechanisms that cause the complex phenomena we are analysing. Critical realism is mainly concerned with ontology and has a relatively open or permissive stance toward epistemology – the theory of knowledge. Sayer observes that with the influence of idealism and relativism realists have been challenged to answer the epistemological question of these anti-realist positions such as “if the social world is socially constructed and significantly concept-dependent, how can it be treated as independent of the researchers’ knowledge?” (Sayer 2000, p 32) Critical realist epistemology has two dimensions. Bhashar (1975) drew the distinction between intransitive and transitive knowledge. Things that are studied such as physical and social phenomena form the intransitive dimension of science while theories and discourse form part of a transitive dimension. Intransitivity has two dimensions with the “natural world being both existentially and causally intransitive and the social, or humanly

31


constructed world, being existentially intransitive but generally not causally intransitive” (Danermark 2002). The transitive is the realm of ideas, concepts, discourses and theories. It is possible to have fallible knowledge (transitive) of reality “which is a social product much like any other, which is no more independent of its production and those who produce it than motor cars, armchairs, or books” (Bhaskar 1978, p28). Thus when theories change (transitive dimension) it does not mean that what they are about also changes (Sayer 2000, p11). By perceiving reality and knowledge about reality as two dimensions, it avoids what Bhaskar terms “the epistemic fallacy” (Bhaskar 1978). This fallacy relates to confusing “what is” with “our concepts of it” thus conflating the intransitive into the transitive. In her discussion of methodological pluralism, Danermark (2002, p153) critiques the use of quantitative methods to validate qualitative analysis. From a critical realist perspective, Danermark argues that to validate entails epistemic fallacy “i.e. you fail to see that reality has ontological depth; you do not realise that an empirical connection in itself cannot identify the active mechanism or mechanisms, nor does it contribute to any profounder information about the interaction of the forces behind the observed pattern”. Critical realists acknowledge the importance of understanding transitive knowledge or meaning as contained, for example, in lay, academic, media, business and policy discourse. Bhaskar (1979, p59) noted that “meanings cannot be measured, only understood. Hypotheses about them must be expressed in language, and confirmed in dialogue. Language here stands to the conceptual aspect of social science as geometry stands to physics”. Thus the voices of study participants suggest phenomenon and mechanisms but care must be taken to avoid conflating these narratives into the objective reality of maternal experiences. Both critical realism epistemology and ontology make this possible through the way it envisages and deconstructs reality.

32


2.4.3 Critical Realism and Causal Inference Sayer (2000, p 13) notes that “one of the distinctive features of realism is its analysis of causation which rejects the standard Humean ‘successionist” view that it involves regularities among sequences of events”. Sayer further argues that instead of the positivist view of causation displayed at Figure 7, realism views causation as depicted at Figure 8 (Sayer 2000, p 14). Cause

Effect

Regularity Source: Sayer (2000, p 14)

Figure 7: Positivist or 'successionist' view of causation Sayer (1992) argued that:

“Much that has been written on methods of explanation assumes that causation is a matter of regularities in relationships between events, and that without models of regularities we are left with alleged inferior, “ad hoc” narratives. But social science has been singularly unsuccessful in discovering law-like regularities. One of the main achievements of recent realist philosophy has been to show that this is an inevitable consequence of an erroneous view of causation. Realism replaces the regularity model with one in which objects and social relations have causal powers which may or may not produce regularities, and which can be explained independently of them. In view of this, less weight is put on quantitative methods for discovering and assessing regularities and more on methods of establishing the qualitative nature of social objects and relations on which causal mechanisms depend.” As discussed above the realist interpretation makes a distinction between the real and actual with “generative” or causal powers that may, or may not, be activated depending upon other conditions (context, other mechanisms). This is particularly important in the field of social epidemiology where social processes are typically dependent upon the actions of “actors” in the various social and organisational structures. Thus for realists causation is not understood based on a model of regular succession of events. “What causes something to happen has nothing to do with the number of times we observe it happening. Explanation depends instead on identifying causal mechanism

33


and how they work, and discovering if they have been activated and under what conditions” (Sayer 2000, p 14). Effect/event

Mechanism

Structure (with causal powers)

Conditions (other mechanisms)

Source: Sayer (2000, p 15)

Figure 8: Critical realist view of causation Realists argue that for such regularities to occur they must occur in a “closed system” where the object possessing the causal power is stable and the external conditions in are constant (Bhaskar 1975). The social world, however, is an “open system” and the same causal powers can thus produce different outcomes depending on the context or spatiotemporal relationship with other objects and their causal powers which may trigger, block or modify actions. Causal inference is the process of drawing conclusions regarding causation by applying forms of reasoning or logic. Danermark and colleagues (2002, p 79) define inference as “a way of reasoning towards an answer to questions such as: What does this mean? What follows from this? What must exist for this to be possible?” They distinguish between four modes of inference: deduction, induction, abduction and retroduction, defined as follows:

Deduction: To derive logically valid conclusions from given premises. To derive knowledge of individual phenomena from universal laws

Induction: From a number of observations to draw universally valid conclusion about a whole population. To see similarities in a number of observations and draw the conclusion that these similarities also apply to non-studied cases. From observed co-variates to draw conclusions about law-like relations

Abduction: To interpret and recontextualised individual phenomena within a conceptual framework or a set of ideas. To be able to understand something in a

34


new way by observing and interpreting this something in a new conceptual framework

Retroduction: From a description and analysis of concrete phenomena to reconstruct the basic conditions for these phenomena to be what they are. By way of thought operations and counterfactual thinking to argue toward transfactual conditions. Critical realist philosophers have been both critical and accepting of Inductive and Deductive forms of inference (Downward, Finch et al. 2002; Downward and Mearman 2007), but argue for the added use of abstract forms of reasoning such as abduction and retroduction to the process of theory building (Danermark, Ekstrom et al. 2002). The following section will propose an explanatory mode of theory building based on critical realist philosophy and methodology.

2.4.4 Critical Realism and Study Design The epistemological aim of critical realism is to explain the relationship between experiences, events and underlying mechanisms. The emphasis is on explaining the “how and why” of particular phenomenon. To undertake this task it is necessary to use different kinds of reasoning: inductive, deductive, abductive, and retroductive, with an emphasis on linking the abstract to the concrete. Sayer (1992) emphasises the importance of different methods of data collection and analysis. He proposes four types of research: intensive or concrete (empirical and theoretical analysis); generalisation (empirical), abstract (theoretical) and synthesis (interdisciplinary analysis). Sayer (2000) further outlines two different kinds of research design relevant to this study. The “intensive research design” is used in research where we wish to obtain in-depth knowledge of a specific phenomenon for the purpose of causal explanation. “Intensive research design” mainly applies to qualitative methods. “Extensive research” typically uses more quantitative methods that seek to identify regularities and patterns. The “extensive” study typically identifies regularities and has limited explanatory power (i.e. of how and why). Jeppesen (2005) identified the requirement to sometimes supplement the “intensive” and “extensive” designs described by Sayer, with a third “explorative design” aimed at establishing an understanding of the area of investigation according to involved parties. In this study I will use both “intensive” and “extensive” study designs with both “exploratory” and “explanatory” phases of analysis.

35


2.5

Theory Building

2.5.1 Introduction A review of epidemiological, social epidemiological and population health literature found a lack of helpful literature on the process of theory building. The exception was in the fields of health education and individual level behaviour change (Green and Kreuter 1999; Nutbeam and Harris 1999; DiClemente, Crosby et al. 2002; Glanz, Rimer et al. 2002). This lack of theory building literature was despite the critical commentary of multi-level epidemiology studies, and social epidemiology in general, that had identified a lack of theoretical models informing methodology (Muntaner 1999; Watt 2002; O'Campo 2003; Carpiano 2006; Carpiano and Daley 2006). O’Campo (2003, p 9), for example, argued that “Perhaps the most pressing issue standing in the way of progress in multilevel research is lack of theory (i.e., system of hypotheses) on the mechanisms by which neighbourhood environments affect health risks, protective factors, and outcomes. …Moreover, given the lack of strong public health hypotheses regarding neighbourhood effects, qualitative research should be undertaken and used to inform and identify mechanisms by which neighbourhood impact health risks and outcomes”. As noted previously Muntaner was also critical of explanations in social epidemiology and proposed adopting a realist philosophy for generating social theory (Muntaner 1999).

2.5.2 Theory Building Approaches I will limit my discussion of theory building approaches to justifying the theory building approach that I have elected to follow in this study. Aspects of methodology and method will be discussed in subsequent sections and chapters. A review of theory building literature identified two dominant approaches. They are: 1. Emergent theory building which uses predominantly inductive forms of reasoning moving from empirical observation and inquiry toward the development of theoretical concepts. In this approach the researcher enters the research situation with no a priori theory and allows the theory to evolve from the data. Emergent and inductive theory building has a long tradition particularly in anthropology, observational epidemiology, and the natural sciences. The approach utilises both

36


quantitative (Hill 1965; Tukey 1980) and qualitative (Glaser and Strauss 1967; Miles and Huberman 1994) forms of empirical data and is the predominant approach to theory building used in mixed method research. 2. Confirmatory theory testing which uses predominantly deductive forms of reasoning moving from a theoretical concept to empirical testing of hypotheses. In this approach the researcher enters the research situation with an a priori theory and the purpose of the data collection is to “confirm” or “disconfirm” that theory. This approach has a more recent tradition (Popper 1959) cited by Rothman and Greenland (1998) and is the foundation for modern experimental science. The approach is not limited to quantitative data and has application to qualitative and mixed method confirmatory studies. A criticism of both inductive and deductive forms of reasoning is that neither contributes to the development of explanatory theories. Haig (2005, p 304) argues that “it is well known that conclusions of valid deductive arguments preserve the information and knowledge contained in their premises” and that although inductive arguments add new information they are only descriptive in nature. “The Scottish philosopher David Hume .. described a disturbing deficiency of inductivism: An inductive argument carried no logical force; instead, such an argument represented nothing more than an assumption that certain events would follow in the same pattern as they had in the past” (Rothman and Greenland 1998, p 17). A third approach to logical reasoning embraced by critical realists is abduction and the related thought process of retroduction. This type of reasoning adds to knowledge by reasoning from factual premises to explanatory conclusions (Haig 2005). 3. Explanatory theory building uses inductive, abductive, retroductive and deduction as the central forms of reasoning moving from description of the concrete, to the abstract, and back to the concrete (Danermark, Ekstrom et al. 2002). In this approach the researcher begins with descriptive and exploratory examination of the phenomena, events and situations intended for study. This is followed by an analytical process that involves identification of components, abduction and retroduction, comparison of theories and abstractions, and concretisation studies of the theorised mechanisms in different situations (Refer to Box 1). The approach uses both quantitative and qualitative methods (Danermark 2002; Danermark, Ekstrom et al. 2002; Haig 2005; Haig 2005) and is discussed further below.

37


Both Emergent and Confirmatory modes of theory building remain supported by the

Explanatory theory building approach. The Explanatory theory building approach uses deductive logic and confirmatory approaches. Danermark and colleagues (2002) argue that “deductive logic can and should be used in analyses of all scientific argument, regardless of what methodology is behind the results presented”. The concretisation studies, proposed by Danermark and colleagues (2002), are effectively confirmatory studies in different concrete situations which then contribute to the explanatory theory. Similarly Haig (2005, p 372) argues that the hypothetico-deductive method “can play a legitimate role in hypothesis and theory testing [and] should thus be seen as complementary to the broader [abductive] theory of method, not a rival to it.”. Similarly inductive logic is used in the Explanatory theory building approaches. The

Abductive Theory of Method proposed by Haig uses “enumerative induction, or induction by generalization, in order to detect empirical phenomena” (Haig 2005. p 372). Both Danermark et al (2002) and Haig (Haig 2005) promote the use of mixed qualitative and quantitative methods to describe and detect regularities in the phenomena under study.

Emergent theory building methods, including Grounded Theory, Exploratory Data Analysis and Regression are used extensively by critical realists as part of their Explanatory Theory

Building (Yeung 1997; Haig 2005; Haig 2005; Olsen and Morgan 2005; Mingers 2006). For this study I have embraced all three forms of theory building, but will focus on conducting Emergent and Explanatory Theory Building, leaving confirmatory theory testing for future research studies. Below I will integrate these three approaches centered on the development of conceptual frameworks, theory and models. Box 1: The stages in an explanatory research based on critical realism Source: Danermark and colleagues (2002, p 109-110)

Stage 1: description An explanatory social science analysis usually starts in the concrete.

We describe the often

complex and composite event or situation we intend to study. In this we make use of everyday concepts. An important part of this description is the interpretations of the persons involved and their way of describing the current situation. Most events should be described by qualitative as well as by quantitative methods.

Stage 2: analytical resolution In this phase we separate or dissolve the composite and the complex by distinguishing the various components, aspects or dimensions. The concept of scientific analysis usually alludes to just this (analysis = a separating or dissolving examination). It is never possible to study anything in all its different components.

38

Therefore we must in practice confine ourselves to studying certain


components but not others.

Stage 3: abduction/theoretical redescription Here we interpret and redescribe the different components/aspects from hypothetical conceptual frameworks and theories about structure and relations. This stage thus corresponds to what has been described above as abduction and redescription. The original ideas of the objects of study are developed when we place them in new contexts of ideas. Here several different theoretical interpretations and explanations can and should be presented, compared and possibly integrated with one another.

Stage 4: retroduction Here the different methodological strategies described above are employed. The purpose is for each one of the different components/aspects we have decided to focus on, to try to find the answers to questions like: What is fundamentally constitutive for the structures and relations(X), highlighted in stage 3? How is X possible? What properties must exist for X to be what X is? What causal mechanisms are related to X? In the concrete research process we have of course in many cases access to already established concepts supplying satisfactory answers to question of this type. In research practice, stages 3 and 4 are closely related.

Stage 5: comparison between different theories and abstractions In this stage one elaborates and estimates the relative explanatory power of the mechanisms and structures which have been described by means of abduction and retroduction within the frame of stage 3 and 4. (This stage can also be described as part of stage 4.) In some cases one might conclude that one theory – unlike competitive theories – describes the necessary conditions for what is to be explained, and therefore has greater explanatory power (see also Chapter 5). In other cases the theories are rather complementary, as they focus on partly different but nevertheless necessary conditions.

Stage 6: concretization and contextualization Concretization involves examining how different structures and mechanisms manifest themselves in concrete situations.

Here one stresses the importance of studying the manner in which

mechanisms interact with other mechanisms at different levels, under specific conditions. The aim of these studies is twofold: first, to interpret the meanings of these mechanisms as they come into view in a certain context; second, to contribute to explanations of concrete events and processes. In these explanations it is essential to distinguish between the more structural conditions and the accidental circumstances. This stage of the research process is of particular importance in applied science.

39


2.5.3 Conceptual Frameworks, Theory and Models Carpiano and Daley (2006) noted that despite an increased consideration of theory within population health “it can sometimes be difficult to identify explicit theories”. Carpiano and Daley provided a guide and glossary on theory building within a post-positivist epistemology. They note that the terms “conceptual frameworks”, “theories” and “models” are often used interchangeably and together, for example, “theoretical model” or “conceptual model”. They proposed a typology based on that used in policy science that encompassed the different levels of abstraction from the broadest level of conceptualisation (framework) to the more focused model (see Figure 9).

Source: Carpiano and Daley (2006, p 566)

Figure 9: Continuum of frameworks, theories and model The conceptual framework is the point where hypothetico-deductive theory building research methods typically start, while more emergent orientated theory-building research methods use the results of their enquiry to inform the development of the conceptual framework (Lynham 2002). Lynham further describes the conceptual development phase of theory building as requiring that the theorist formulate initial ideas in a way that depicts current, best, most informed understanding and explanation of the phenomenon, issue, or problem in the relevant world context. Citing Dubin (1969) and Kaplan (1964), Lynham proposes that: “.. at a minimum this process will include the development of the key elements of the theory, an initial explanation of their interdependence, and the general limitations and conditions under which the theoretical framework can be expected to operate. The output of this phase is an explicit, informed, conceptual framework that often takes the form of a model and/or metaphor that is developed from the

40


theorist’s knowledge of and experience with the phenomenon, issue, or problem concerned” (Lynham 2002, p 232). Thus “regardless of the sequencing of the conceptual development phase of theory building, the development of an informed conceptual framework is fundamental to all theory-building research” (Lynham 2002, p 232). A Theory is a denser and logically coherent set of relationships that attempts to explain direction, hypotheses and how variables may covary. A Theory makes specific assumptions that will then enable them to diagnose phenomenon, explain processes, and predict outcomes (Carpiano and Daley 2006). One of the most frequently quoted definitions of theory is that “a theory is a set of interrelated concepts, definitions and propositions that present a systematic view of events or situations by specifying relations among variables in order to explain and predict the events or situations” (Glanz, Rimer et al. 2002, p 25). Nutbeam and Harris (1999, p 10) have stated that a fully developed theory would be characterised by the explanation of three major elements: 1. The major factors that influence the phenomena of interest 2. The relationships between those factors 3. The conditions under which these relationships do or do not occur. Finally a model is the narrowest in focus and is used to make specific assumptions about a limited set of parameters and variables. A model may draw on several theories and when presented as a diagram a conceptual model may provide a visualisation of proposed causal linkages (Carpiano and Daley 2006).

2.5.4 An Explanatory Theory Building Method Based on the above discussion I have incorporated emergent and confirmatory theory building approaches within a overarching critical realist explanatory theory building framework. As illustrated in Figure 10 the emergent phase leads to the development of a tentative conceptual framework. The emergent phase as described here includes mixed method inductive, deductive and abductive methods of theory generation such as exploratory data analysis, interviews, conceptual and category coding, situational analysis, exploratory factor analysis and

41


constant comparative analysis (i.e. grounded theory). I will describe the emergent phase methods in more detail below and in their respective chapters. The construction phase (described as theory development and theory appraisal by Haig (2005a)) builds a conceptual framework, theory and model utilising an integration of interdisciplinary (mixed method) research, abstract thinking, comparison of theories, and identification of the best explanation(s). I will describe the Construction Phase further in Chapter 8: Theory Construction. The Confirmatory Phase builds on the Concretization and Contextualisation stage described by Danermark and colleagues. Hypotheses are developed from the theoretical propositions, operationalised, and studied in concrete situations. The Confirmatory Phase will be briefly discussed as proposals for future research in Chapter 10: Conclusion, Limitations and Implications. Explanatory Theory Building

Emergent

Construction

Confirmatory

Phase

Phase

Phase

 Observation

 Integration

 Hypotheses

 Pattern

 Abstract thinking

 Operationalisation

 Concepts

 Theory Comparison

 Observation

 Abstract thinking

 Best explanation

 Confirmation

 Tentative

 Proposition

framework

 Conceptual framework  Modelling

Figure 10: Phase of Explanatory Theory Building

42


2.6

Research Aims

As noted in Chapter 1: Introduction, the over arching aim of the study is to utilise mixed methodology to build a conceptual framework, theory and model describing and explaining the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression in South West Sydney. Specific objectives are: 1. To explore and describe the individual level influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney 2. To explore and describe the neighbourhood and community level economic, social, and physical influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney 3. To explain and conceptualise a framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression. 4. To describe the implications and future confirmatory approaches. Specific analysis questions for Emergent Theory Building Phase are: 1. What is the impact of perinatal events on developmental and life-course outcomes with a focus on perinatal depression? 2. How does maternal depression cause detrimental effects to the foetus and newborn infant? 3. What, and how do, individual and family level factors influence the phenomenon of maternal depression during the perinatal period? 4. What, and how do, group level factors influence the phenomenon of maternal depression during the perinatal period? Further analysis questions can be found in each Chapter as appropriate and relevant. The emergent and exploratory nature of the methodology requires that new questions are asked throughout the theory building process. This is in contrast to the approach taken in deductive or confirmatory phases of theory building.

43


2.7

Emergent Phase Methodology

2.7.1 Introduction A mixed-method methodology will be used to build an emergent conceptual framework, theory and model to describe the mechanisms by which multi-level factors influence developmental and life-course outcomes with a focus on perinatal depression. As previously stated my ontological orientation is in the realist tradition and more specifically that of a critical realist. The epistemological position that I will use in this study is also critical realist. This section will describe the methodological approach for the Emergent Theory Building Phase. The focus of this Phase will be on “Phenomenon Detection” and “Theory Generation”. The integration of these two theory building elements is necessary because several of the methods that will be used in the Emergent Phase are suitable for both purposes.

2.7.2 Prior Position and Use of Literature In emergent and inductive approaches to theory building the researcher aims to enter the research situation with no a priori theoretical position and allows the theory to evolve from the data. In reality this situation seldom exists. In early writings on grounded theory it was emphasised that researchers should set aside, as much as possible, preconceived ideas about the phenomenon of interest. In later years there has been a divergence of approach with some encouraging the use of prior knowledge and cognitive heuristics to help explore the nature of meaning (Jaccard and Jacoby 2010). In this study I avoided, at the beginning, the reading of literature around relevant area(s) of theory development but instead concentrated on reading only related empirical studies and theoretical literature from fields other than perinatal depression. I entered this research study, however, with prior understandings of perinatal and child population health, and recently completed a quantitative study of postnatal depression in South West Sydney. As part of that research I had undertaken an extensive review of epidemiological studies of postnatal depression. That literature review identified theory related to the influence of postnatal depression on developmental and life course outcomes. It also identified individual level risk and protective factors. I had not, however, studied previous literature related to the impact of neighbourhood context on perinatal depression. I subsequently read relevant literature as theoretical concepts emerged from the empirical studies. I treated that literature as additional empirical data to be considered

44


in the generation of theoretical concepts. Theoretical literature was avoided until the

Theory Construction Phase (see Chapter 8). Based on my prior knowledge and beliefs I have constructed the following Perinatal Conceptual Framework (Figure 11) to explore and build theory and models.

Economic, Social, Physical and Political Context

Individual and Family Characteristics

Maternal Response – physical and psychological

Foetal and Infant Response – physical and psychological

Figure 11: Proposed Perinatal Conceptual Framework Four research questions were taken from this conceptual framework. 1. What is the impact of perinatal events on developmental and life-course outcomes with a focus on perinatal depression? 2. How does maternal [stress and] depression cause detrimental effects to the foetus and newborn infant? 3. What, and how do, individual and family level factors influence the phenomenon of maternal depression during the perinatal period? 4. What, and how do, group level factors influence the phenomenon of maternal depression during the perinatal period?

45


2.7.3 Phenomenon Detection and Description Stage one of the critical realist approach to explanatory research described by Danermark and colleagues (Danermark, Ekstrom et al. 2002) involves the description of the “concrete”.

“We describe the often complex and composite event or situation we intend to study. In this we make use of everyday concepts. An important part of this description is the interpretation of the persons involved and their way of describing the current situation. Most events should be described by qualitative as well as by quantitative methods” (Danermark, Ekstrom et al. 2002, p 109) Haig (2005, p 374) argues for the detection of phenomena which commonly take the form of empirical regularities and “comprise a varied ontological bag that includes objects, states, processes, events, and other features that are hard to classify. He argues that phenomena must be distinguished from data which are “idiosyncratic to particular investigative contexts” and which serve as evidence for the phenomena under investigation. “It is for this reason that, when extracting phenomena from data, one often engages in data exploration and reduction by using graphical and statistical methods. … scientists use a variety of methodological strategies [including] controlling for confounding factors (both experimentally and statistically) … engaging in data analytic strategies … and constructively replicating study results”. Haig further notes that these “procedures are used for the detection of phenomena [and] are not used in the construction of explanatory theory” (Haig 2005, p 375). Thus the various forms of quantitative exploratory data analysis including regression studies are suitable for phenomena detection. In relation to qualitative data Haig suggest that “Glaser and Strauss's general plea for checking the data should be taken by grounded theorists as a reminder that they should seek to reliably establish phenomena in multiply-determined ways before they begin to generate grounded theory” (Haig 1995, p 1). Thus in the Emergent Phase I will use multiple mixed methods including: interviews, focus groups, exploratory data analysis, regression, exploratory factor analysis and literature reviews to describe the situation under study and to detect phenomena.

46


2.7.4 Theory Generation The emergent phase of theory building uses predominantly inductive forms of reasoning moving from empirical observation toward the development of theoretical concepts. The mode of reasoning is not, however, only inductive but also involves deductive and abductive analysis. The conceptual and categorical coding of data requires abstract reasoning and the constant comparative method (Glaser and Strauss 1967) uses both inductive and deductive reasoning in a manner that some methodologists equate as abduction. Similarly some modes of exploratory data analysis such as exploratory factor analysis, and its variants, use abductive reasoning to ascribe meaning to identified unobserved dimensions or latent variable (Haig 2005). Three methods proposed by Haig (2005) for the Theory Generation stage of theory building are: Grounded theory method, Exploratory Factor Analysis and Heuristics. Haig observes that Grounded theory method and Exploratory Factor Analysis have data analysis characteristics and can be used for both phenomenon detection and theory generation. Yeung (1997, p 70) noted that “critical realism is still largely a philosophy in search of a method” and proposed the use of iterative abstraction (i.e. abduction and retroduction), grounded theory method, and methodological triangulation as suitable for the complexities of social science research. Analytical resolution is Stage 2 of the explanatory research approach proposed by Danermark and colleagues. It is similar to the coding of concepts and categories in Grounded Theory Method. “In this phase we separate or dissolve the composite and the complex by

distinguishing the various components, aspects or dimensions. The concept of scientific analysis usually alludes to just this (analysis = a separating or dissolving examination). It is never possible to study anything in all its different components. Therefore we must in practice confine ourselves to studying certain components but not others.” (Danermark, Ekstrom et al. 2002, p 109-110) Abduction and retroduction are the third and fourth stages in the approach proposed by Danermark and colleagues and are discussed elsewhere. These forms of reasoning will be used in the Theory Generation Stage to develop tentative conceptual frameworks in both the qualitative and quantitative arms of this study. This will be undertaken using: categorical principal component analysis, exploratory factor

47


analysis, specification scan, coding of concepts and categories, constant comparative analysis, drawing of conceptual networks, and situational analysis to move from the “concrete to the abstract”.

2.8

Emergent Phase Research Design

2.8.1 Introduction The purpose of the Emergent Phase is to build a tentative conceptual model describing the mechanisms by which multi-level characteristics influence developmental and lifecourse outcomes with a focus on perinatal depression. The emphasis is on describing the mechanisms by which multi-level characteristics might influence outcomes. Consequently the first two research questions will be treated lightly utilising literature review and expert panel. Emphasis will be given to exploring analysis questions three and four utilising an emergent mixed-methods design that integrates both quantitative and qualitative empirical research.

2.8.2 Study Setting The setting for this study is in the Local Government Areas of Bankstown, Fairfield, Liverpool, Campbelltown, Camden, Wollondilly and Wingecarribee. The population in SWSAHS is large and continues to rapidly increase. From the 2001 Census of Population and Housing, (ABS 2002; DOH 2003; SWSAHS 2004) SWSAHS was reported to have a population of 769,595 and this is projected to grow to nearly 891,000 by 2006. Fourteen percent of the State’s births are in this area with 10,011 babies born in 2002-2003. The area has a diverse multicultural population with 28.4 percent of the population been born overseas compared with 17.8 percent for the rest of NSW. Twenty percent of babies in SWSAHS are born to women from South East, North East or Southern Asia. Also, one quarter of Sydney’s indigenous population live in South Western Sydney. South Western Sydney is also an area of substantial social disadvantage, and has lower education attainment and lower income levels then other parts of NSW. Based on composite socio-economic indices, about two-thirds of the area is substantially disadvantaged, which is associated with a range of poor health indicators. There is a high rate of public housing, at 10.1 percent, which is nearly twice the NSW rate (SWSAHS 2004).

48


South West Sydney

Source: Baum (Baum 2008, p 18), RD (Relative Deprivation)

Figure 12: South West Sydney as a Deprived Region of Sydney

49


2.8.3 Concurrent Triangulated Mixed Method Design I elected to use a concurrent triangulated mixed method research design. The concurrent triangulation design is one of the most commonly used mixed method designs (Tashakkori and Teddlie 2003, p 229). The data collection usually occurs concurrently during one phase of the research study, in this case the Emergent Phase and findings are integrated during the interpretation phase. In this study I will integrate the research objectives, data collection, findings and analysis in both the Emergent and the Theory Construction

Phases. Integration will thus be a constant process. During the Emergent Phase I will use a comparative approach between qualitative (intensive) and quantitative (extensive) study arms. The emerging findings will be used to inform further exploratory analysis. Literature will be sourced and reviewed as theory emerges. The Construction Phase will build on the earlier integrated analysis using

triangulation, and other theory construction methods. INTEGRATION OF METHOD

QUANT

QUAL

Phase Phase

Construction

Emergent

(Extensive)

(Intensive)

QUAN data collection

QUAL data collection

QUAN data collection

QUAL data collection

QUAN data analysis

QUAL data analysis

TRIANGULATION

Figure 13: Concurrent Mixed Method Design

50


2.8.4 Integration and Triangulation The approach that I have taken has a focus on integration. Integration of methods, data collection and analysis is the hallmark of mixed methods research (Creswell, Fetters et al. 2004; Yin 2006; Woolley 2009). Yin (2006, p 41) argues that without such integration “different methods may sit in parallel, potentially leading to multiple studies, and not the desired ‘mixing’ of methods implicit in mixed methods research”. Yin proposes that integration should occur in relation to: research questions, units of analysis, samples for study, instrumentation and data collection methods, and analytic strategies. The research design adopted for this study will strive to achieve the standards of integration proposed by Yin with integration occurring through use of common research questions, study design, units of analysis, samples for study and analytic strategies during both emergent and construction phases. The degree to which this integration is achieved will be discussed in Chapter 10. Confusion has arisen in the literature concerning the use of the term triangulation in relation to combining quantitative and qualitative approaches (Woolley 2009). Woolley observed that “its use … appears to have resulted in a common misconception … that mutual validation is the goal of mixed methods studies” (Woolley 2009, p 8). Kelle in an earlier critique of triangulation observed that there were three different understandings of the triangulation metaphor, “triangulation as mutual validation, triangulation as the integration of different perspectives on the investigated phenomenon and triangulation in its original trigonometrical meaning” (Kelle 2001, p 1). In a consideration of the micro- and macrolevels of sociological analysis Kelle suggests that an “understanding of triangulation in its original trigonometrical sense may be helpful in gaining a deeper insight into theoretical aspects of method integration” (Kelle 2001, p 1). Critical realism offers a further perspective on triangulation in mixed methods research (Modell 2009). Modell (2009, p 208) argues that critical realism counters many criticisms of triangulation “by re-conceptualizing it as firmly grounded in abductive reasoning. This provides a foundation for maintaining researchers’ sensitivity to context-specific variations in meaning in efforts to derive theory-related explanations”. Triangulation will be used in Chapter 8 in its original trigonometrical sense pulling together micro and macro analysis and as the basis for the abductive and retroductive reasoning processes.

51


2.8.5 Overview of Literature Reviews An integrative review method was used (Dixon-Woods, Agarwal et al. 2005; Whittlemore 2005). Integrative reviews summarise past research and examine relationships between concepts of interest. General conclusions and insights can then be drawn from the broad body of literature. As stated above I entered this research study with prior understandings of perinatal and child population health, and having recently completed a quantitative study of postnatal depression in South West Sydney I had undertaken a review of epidemiological studies of postnatal depression. I had not, however, previously studied qualitative or theory related literature. I was thus able to conduct the qualitative studies with limited knowledge of findings from other studies. I subsequently read relevant literature as theoretical concepts emerged. I treated that literature as additional empirical data to be considered in the generation of theoretical concepts. Theoretical literature was avoided until the Theory Construction Phase (see Chapter 8) where it is extensively used to compare theoretical constructs and to assist the selection of the best explanation. This iterative literature review process in emergent research studies is problematic in relation to writing the dissertation as comprehensive literature reviews are usually conducted at the beginning. The literature review reported in Chapter 3 was undertaken concurrently with the studies reported in Chapters 4 and 5. The Chapter was then revised during the Theory Construction Phase to ensure consistency and logical progression of ideas. Literature reviews related to social and ecological factors was undertaken as part of the Theory Construction Phase and is presented predominantly in Chapter 8. Literature reviews related to the philosophical, theoretical and methodological underpinnings of this study are presented predominantly in Chapter 2 and 9. Literature related to the qualitative and quantitative methods employed are presented in the relevant chapters. The predominant search engines used were Medline, Pub Med, Embase, Eric, Cinahl and Google Scholar. Where possible, systematic reviews, and meta-analyses are used.

52


2.8.6 Qualitative Methods 2.8.6.1 Introduction The qualitative methodology that I have chosen to use follows the theory building approaches described by Miles and Huberman (1994) and Glaser and Strauss (1967). Miles and Huberman describe the building of conceptual frameworks noting that theory building relies on a “few general constructs that subsume a mountain of particulars. Categories … are labels we put on intellectual “bins” containing many discrete events and behaviours… Bins come from theory and experience and (often) from the general objectives of the study (p18, Miles and Huberman 1994; Haig 2005). Miles and Huberman (1994) describe an approach that starts with an organising framework for the codes derived from prior knowledge. In this inductive phase of study I predominantly followed the “grounded” approach first described by Glaser and Strauss (1967) and used minimal “preset” codes. For the second stage of analysis a conceptual mapping approach was used to develop causal networks as described by Miles and Huberman (pp 151-1651994). There are several approaches to analysing the conditions, context or situation that participants experience and that influence causative mechanisms or participants actions. Those approaches include the Conditional/Consequential Matrix, or its earlier variants (Corbin and Strauss 2008 ,p94), Context Charts (Miles and Huberman 1994 ,pp102-105), the Research Map (Layder 1993) or Situational Analysis as recently described by Clarke (Clarke 2005; Clarke and Friese 2007; Clarke 2009). I elected to undertake a third cycle of analysis using the Situational Maps and Social Worlds/Arena Maps described by Clarke (2005). Qualitative research will consist of the literature reviews (discussed above) key informant interviews and three focus groups. The purpose of the qualitative research was to identify and explore community and professional knowledge regarding multi-level influences on perinatal depression. The findings of the Qualitative Research are presented in Chapters 4 and 7.

53


2.8.6.2 Overview of Methods The study setting was the South West Sydney local government areas of Bankstown, Fairfield and Liverpool. Key informant interviews. Subjects for key informant interviews were purposively selected to represent academic, local government, human services (including health), commerce and civil society perspectives. Six key informant interviews were initially planned but others were necessary to explore themes emerging from both the qualitative and quantitative data. Focus groups: Three focus groups with 3-7 participants each were used. Subjects were invited to participate based on their membership of existing “playgroups” in the study setting. Coercion was avoided by using Families NSW funded services to invite the participants. Interview Guides for the initial interviews and focus groups are attached. Open ended questions were used and questions changed as data was analysed. Data Recording was by digital voice recording and note taking. Data was transcribed and voice files secured password protected hard drive. Data analysis of notes, coding and memos was assisted using Atlas.ti Mixed Method research software. Analysis was undertaken in three phases. Open coding was the predominant approach to initial coding followed by conceptual mapping using causal networks (Miles and Huberman 1994). The third phase of analysis used situational analysis as described by Adele Clarke (2005).

54


2.8.7 Quantitative Methods 2.8.7.1 Introduction As discussed above critical realist methodology supports the use of quantitative data to detect phenomena and generate theory (Danermark, Ekstrom et al. 2002; Haig 2005; Haig 2005). Similarly quantitative Exploratory Data Analysis (EDA) is a valid post-positivist and critical realist, methodology (Tukey 1980; Behrens 1997; Haig 2005). Most epidemiological and statistical texts treat EDA as being fundamentally descriptive statistics. But EDA is better defined more as an epistemological attitude to be taken rather than the techniques used (Tukey 1980). In this endeavour Tukey (1980) stated the maxim that “finding the question is often more important than finding the answer” Tukey stressed the need to reorganise the early phase of the research process to explore what is going on and to develop the “idea of a question – something often thought of in terms of the common language as a question – but not at all the kind of question that can have a statistically supported answer”. This tradition has been further characterised by Behrens (1997, p 131-132) as:

1.

“An emphasis on the substantive understanding of data that address the broad question of “what is going on here”?

2. An emphasis on graphical representation of data 3. A focus on tentative model building and hypothesis generation in an interactive process of mode specification, residual analysis, and model respecification 4. Use of robust measures, re-expression, and subset analysis 5. Positions of scepticism, flexibility, and ecumenism regarding which methods to apply”. The goal of EDA is to discover patterns in data. “The role of the data analyst is to listen to the data in as many ways as possible until a plausible “story’ of the data is apparent, even if such a description would not be borne out in subsequent samples” (Behrens 1997). P 132 Inductive processes in quantitative research involve both basic and sophisticated methods of exploration data analysis (EDA). Thus in the Emergent Phase I will use the following quantitative methods: univariate data display, spatial visualisation, cluster analysis, bivariate and multiple regression (including linear, logistic, ecological, Bayesian spatial and multi-level), categorical principal component analysis, exploratory factor analysis, and exploratory CFA specification scan.

55


2.8.7.2 Overview of Methods The study setting is the local government areas of Bankstown, Fairfield, Liverpool, Campbelltown, and Camden. The individual level study population were mothers of infants born in 2002 and 2003. The individual level data was provided by the Sydney South West Area Health Service in a de-identified form from Obstetric and Maternal and Infant administrative information systems. The group level study population is the ABS Census population in 2001 and 2006, subjects in the NSW Crime Statistics and participants in NSW Health surveys. Data Management: The data was examined for missing data and appropriate imputation and deletion methods used. For categorical variables transformation was undertaken (ie dummy variables, centering etc) to enable their use in linear, logistic and multi-level statistical analysis. Group level data sets were constructed using 2001 and 2006 Census, Health Surveys, Crime statistics and aggregated individual level variables. Separate exploratory and confirmatory datasets were created. Individual Level Exploratory Data Analysis: Univariate and Bivariate studies of the exploratory (1998-2003) dataset was undertaken using descriptive statistics, Chi Square, correlations and Bivariate regression. The outcome variables of study are Edinburgh Depression Scale > 9 and > 12. Examination for multicollinearity was undertaken. Categorical Principal Component Analysis and logistic regression were used to further explore associations and latent vectors for consideration in the theory building. Group Level Exploratory Data Analysis: Univariate and Bivariate studies were undertaken using descriptive statistics, Chi Square, correlations and Bivariate regression. The grouplevel outcome variables of study were aggregated EDS >9 and >12. Examination for multicollinearity was undertaken. Factor analysis and linear regression were used to further explore associations and latent vectors for consideration in the theory building. Spatial analysis was undertaken including cluster analysis, and geographical information system (GIS) analysis. Bayesian spatial and multilevel regression studies were undertaken using WinBUGS and posterior RRs included into GIS mapping software. Statistical Software: The statistical software that will be used includes, R 2.5.0, SPSS 15.0 and 18.0, AMOS 18, MS Excel, MS Access, SaTScan 7.0.3, MLWin 2.02, WinBUGS 14 and GeoBUGS 14, MapInfo 7.0, ArcCensus, and ArcGIS-ArcView 9.3.

56


2.9

Theory Construction Phase

The purpose of the Theory Construction Phase is to undertake abductive triangulation of the findings from the mixed method studies conducted in the Emergent Phase and to construct a conceptual framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression. The approach will be to use again the analytical resolution, abduction and retroduction stages described by Danermark and colleagues (see Box 1) together with their stage five: comparison between different theories and abstractions (Danermark, Ekstrom et al. 2002). Theory building approaches used by hypotheticodeductive theorists will also be used. The Methods used in the Theory Construction Phase include: 1. Defining Stratified Levels: The ontological levels described in Section 2.3 will be applied here as a component of the analytical methodology 2. Building a Conceptual Framework: Conceptual frameworks developed during the emergent phase will be revisited with the intention of moving toward a “causal network” as described by Miles and Huberman (1994) 3. Analytic Resolution: The composite and complex are separated and dissolved into their components and dimensions. Analytical resolution will be applied in the Theory Construction phase in order to confine the study to certain aspects of the emerging theory 4. Abductive reasoning: As discussed above abductive reasoning will be used to reinterpret and recontexualise individual phenomena within relevant identified conceptual frameworks 5. Comparative analysis (triangulation): Comparative analysis will be continued in the Theory Construction phase. Here the process will use triangulation in its original trigonometrical sense and avoid any implication of validation (cf. epistemological fallacy) 6. Retroduction: Counterfactual and other abstract thinking will be used to describe the conditions for phenomena to be what they are

57


7. Postulate and Proposition Development: Theoretical statements will be expressed as postulates and propositions leaving the generation of related hypotheses to later studies 8. Comparison and Assessment of Theories: Theoretical propositions will be compared with alternative theories and abstractions seeking the best explanation 9. Modelling: Graphical representation of causal networks and propositions will be used. Directed Acyclic Graphs (DAGs) will be presented where appropriate. Structural equation modelling (SEM) will be left for future confirmatory studies.

2.10 Ethical Considerations and Approvals Ethics approval to conduct this research was sought and obtained from the UNSW Human Research Ethics Committee. Separate applications for the quantitative and qualitative studies were submitted. 1

Perinatal social epidemiology: a study of the economic, social, physical and political context of perinatal and infant health in South West Sydney using perinatal depression as a case study: interview research to inform the building of theory and conceptual model (HREC 08045)

2

Perinatal social epidemiology: a study of the economic, social, physical and political context of perinatal and infant health in South West Sydney using perinatal depression as a case study: Epidemiological Study to inform the building and testing of theoretical and conceptual models (HREC 08246).

Approval letters are at Appendix A.

58


2.11 Conclusion The purpose of this chapter has been to outline overall philosophy, methodology and research design with a focus on the Emergent Phase of theory building (Parts B and C). The Theory Construction Phase is considered in more detail in Part D. The philosophical sections provided rationale and description of the critical realist ontological and epistemological positions I have chosen to use. Subsequent sections examined theory building, methodological and design aspects with a focus on critical realist approaches to explanatory theory building. The methods were only briefly described in this Chapter with further detail provided as relevant in the subsequent chapters. This approach has been taken to assist the reader to link quite different technical methods with their findings. Qualitative and quantitative studies were undertaken concurrently with the intention of ensuring that they each informed the other. The layout of the following Chapters does not reflect the order within which the work was undertaken.

59


60

Part B: Individual Level

Part B: Individual Level Exploratory Analysis Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







61

Part B: Individual Level

62

Chapter 3: Perinatal Adversity

Chapter 3: Perinatal Adversity, Life-course Outcomes and Depression Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







63


3.1

Introduction

The purpose of this chapter is to establish the significance of research in the field of perinatal social epidemiology and specifically the study of perinatal depression and perinatal adversity. The importance of early beginnings will be reviewed with consideration of the “early life hypothesis”, “life course perspective”and emerging field of developmental epidemiology as central tenants of the field of Perinatal and Infant Social Epidemiology. Having established why it is that the early years are important for adult physical, mental and social health I will provide an overview of current perinatal depression empirical studies and explore the mechanisms by which perinatal depression and stress may impact on the infant and unborn foetus. As this phase of the study is emergent in nature, the literature review presented here is intended as a general overview of the field. Literature relevant to the emerging theory will be explored in more depth in Section D as part of the theory construction phase.

64


3.2

Research Aims

The overall aim is to utilise mixed methodology to build a conceptual framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression in South West Sydney Specific objective of the Emergent Phase – Individual Level Studies is: 1. To explore and describe the individual level influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney The specific analysis questions that I will explore in this chapter are: 1. What is the impact of perinatal events on developmental and life-course outcomes? 2. What individual and family level factors influence the phenomenon of maternal depression during the perinatal period? 3. How does maternal depression cause detrimental effects to the foetus and newborn infant?

65


3.3

Early Years and Life Course Consequences

What is the impact of perinatal events on developmental and life-course outcomes? In this section I will explore why it is that the early years are important for adult physical, mental and social health. Thus I will establish the importance of perinatal depression, stress, and ecological context. Neuro-developmental science and population health have made dramatic advances in assisting us to understand how early influences impact on brain development and subsequent risk or resilience to poor outcomes (Barker 1992; Elo and Preston 1992; Barker 1998; Keating and Hertzman 1999; Shonkoff and Meisels 2000; Ben-Shlomo and Kuh 2002). As a consequence it is increasingly being recognised that major adult public health issues, related to development, behaviour and lifestyle, such as suicide, depression, violence, crime, school failure, unemployment, injury, unintended pregnancy, substance abuse, and obesity have their origins during pregnancy, infancy and early childhood. Both the “Early Life Hypothesis” and “Life Course Perspective” address weaknesses in current research theories that attempt to explain social and geographical inequalities in health exclusively on the basis of adult exposures (Brunner 2000). Baker and co-workers (1992) introduced “biological programming” as a concept that emphasises the links between poor development in utero and infancy and later chronic disease. Much of the early work on “biological programming” focused on maternal and foetal nutrition to explain the strong and consistent relationships between birth weight and aspects of health in adult life (Barker 1992; Barker, Hales et al. 1993; Leon, Kupilova et al. 1996; Lithell, McKeigue et al. 1996; Barker 1998). Another area of “biological programming” research has been in relation to the effects of prenatal stress on long term outcomes (Kramer, Goulet et al. 2001; Welberg and Secki 2001; Avishai-Eliner, Brunston et al. 2002; Halfon and Hochstein 2002; Hobel and Culhane 2003; Seckl 2004; Field, Diego et al. 2006; Kajantie 2006; Kapoor, Dunn et al. 2006; Alder, Fink et al. 2007; Gluckman, Hanson et al. 2007; Talge, Neal et al. 2007; Weinstock 2008; Lupien, McEwen et al. 2009). Current infant and child outcomes for which there is some evidence of the role of antenatal stress include: Low birth weight, increased basal glucocorticoid axis activity, unsociable and inconsiderate behaviours, attention deficit hyperactivity, sleep disturbances, drug abuse, and mood and anxiety disorders (Lupien, McEwen et al. 2009).

66


A more recent development has been the work of Gluckman and colleagues (2007, p 19) who discuss the Developmental Origins of Health and Disease (DOHaD) in terms of “adaptive processes that allow genotypic variation to be preserved through transient environmental changes”. Their focus is on explaining the causal mechanism at a genetic and biological level and thus demonstrating that “manipulation of the environment in the period extending from conception to infancy can be associated with permanent changes in physiology and/or structure” (Gluckman, Hanson et al. 2007) p1. Current explanation of the genetic level mechanisms is based on the observation that permanent alteration in gene expression is regulated by epigenetic factors such as DNA methylation and histone methylation or acetylation (Gluckman and Hanson 2006). At the phenotypic level Gluckman and colleagues (2007) propose a developmental plasticity mechanism whereby the developmental trajectory is reset by the foetal response to an adverse intrauterine environment in expectation of poor postnatal conditions. This process of phenotypic plasticity allows a range of phenotypes to be expressed from one genotype in response to environmental factors operating during the phase of developmental plasticity. The Life Course Perspective proposes the accumulation of risk throughout an individual’s lifetime from the social and environmental influences on biological processes. The Life Course Perspective (Kuh and Ben-Shlomo 1997) accepts that there may be of critical periods and “biological programming” but emphasises the accumulation of risk resulting from exposure to adverse environments and illnesses during childhood, adolescence and adulthood. This view sees the influences of the intervening years between early life and the onset of disease to be essential in a full account of health determinants (Brunner 2000). There is, for example, evidence that emotional deprivation in childhood is linked to poor educational attainment and behavioural problems, such as hyperactivity and other conduct disorders, which may be precursors of a lifetime of material and emotional insecurity (Brunner 2000). Brunner cites Fonagy (1996) whose studies of the attachment patterns of parents and their children suggest early caregiver experiences may contribute to the intergenerational transmission of physical and psychological vulnerability. Similarly breast feeding has been identified as having important developmental and lifecourse consequences for infants. Anderson and others (1999) report on a metaanalysis of observed differences in cognitive development between breast-fed and formulae fed infants. Twenty studies met their initial inclusion criteria. They found that significantly higher levels of cognitive function were seen in breast-fed infants than in formula fed infants at 6-23 months of age and that the cognitive development benefits increased with duration of breastfeeding. Higher cognitive function is known to be

67


positively and significantly correlated with educational achievement, job performance, occupational achievement and income. It is also inversely related to delinquency rates (Anderson, Johnstone et al. 1999). Of relevance to this Study is the observed association of perinatal depression with reduced breast-feeding rates and the association of breastfeeding with maternal-infant attachment. The above discussion has referred to the role that social level structures and mechanisms may play in relation to “biological programming”, DOHaD, and Life Course origins of adult disease. Numerous scientific studies have identified that socio-economic conditions in childhood have an influence on health in adult life (Notkola, Punsar et al. 1985; BenShlomo and Davey Smith 1991; Elo and Preston 1992; Lundberg 1993; Davey Smith, Hart et al. 1997; Kuh and Ben-Shlomo 1997; Lynch, Kaplan et al. 1997; Barker 1998; National Research Council and Institute of Medicine 2000; Ben-Shlomo and Kuh 2002; Gilman, Kawachi et al. 2002; Galobardes, Lynch et al. 2004). A significant contribution has been made by Galobardes and others (2004) who undertook a systematic review of individual-level studies examining childhood SES and adult overall and cause-specific mortality. They identified 29 studies. All were prospective or cohort designs with the exception of two case-control studies (Ekbom, Hsieh et al. 1996; Preston, Hill et al. 1998) and one cross-sectional survey of reported sibling mortality (Bobak, Murphy et al. 2003). The authors also reviewed migrant, ecologic and disease trend studies in an effort to “triangulate” the evidence on the potential for early life SES conditions to influence adult health. From the evidence the authors concluded that there was a “reasonably” consistent picture of the contribution of childhood SES circumstances in shaping adult disease risk. In the following section I will review perinatal depression with a view to subsequently explaining how perinatal depression might impact on biological programming, DOHaD, and life course outcomes.

68


3.4

Perinatal Depression

What individual and family level factors influence the phenomenon of maternal depression during the perinatal period? It has been well established across ethnic and geographically distributed groups that the lifetime rate of major depressive disorder among women is double that of men. Depression among women clusters within the childbearing years with up to 70 percent of pregnant women reporting depressive symptoms and 10-16 percent fulfilling diagnostic criteria (O'Hara, Neunaber et al. 1984; Affonso, Lovett et al. 1990; Beck 1995; Weissman and Olfson 1995). The majority of estimates of prevalence are between 10-20 percent. O’Hara and others in a meta-analysis estimate a prevalence of 13 percent (O'Hara and Swain 1996). Differences between studies may be accounted for by method of assessment used and the period under study. While the overall prevalence of postpartum depression is similar to other life periods, there is some evidence that there is an increased risk of depression occurring in the early postpartum period. There may be a threefold increase in the first 5-6 weeks (Cox, Murray et al. 1993; Fergusson, Horwood et al. 1996). There have been a number of Australian published studies of postpartum depression that report on prevalence. An Australian study of 258 primiparous women found high Edinburgh Postnatal Depression Scale (EPDS) scores (using the recommended cut-off above 12.5) in 8.9 percent of women at three months and 6.4 percent at six months (Boyce, Parker et al. 1991). The Brief Symptom Inventory (BDI) estimates of depression (using scores above 10.5) were higher, with 13.5 percent of women being identified as depressed at three months and 11.4 percent at six months. Boyce and others (1993) undertook a validation of the EPDS in a group of 103 primiparous and multiparous Australian women who were assessed at around 12 weeks postpartum using DSM-III-R criteria. Major depression was diagnosed in 8.7 percent. There were significantly different mean EPDS scores for depressed (mean=18) and non-depressed (mean=5) women, although there were no differences in age, marital status, or parity (Boyce, Stubbs et al. 1993). The EPDS was used to screen 235 primiparous and multiparous Australian women in the first few postpartum days, at six weeks (N=222) and six months (N=192). High EPDS scores were recorded for 22 (9%) women while still in hospital (most likely measuring the

69


maternity blues), 21 women (9%) at six weeks and 19 women (10%) at six months postpartum (Astbury, Brown et al. 1994; Stamp and Crowther 1994). Astbury and others (1994) undertook a population-based study in Victoria. It involved a postal survey of all women who gave birth during one week in 1989. Questionnaires were mailed to 1,193 primiparous and multiparous women at eight months postpartum, and 799 questionnaires were returned with full responses available for 771 women (71.4% of the sample). Younger women, single parents and women born overseas of non-English speaking background were under-represented. Postpartum depression was assessed using the recommended EPDS cut-off level and the point prevalence was 15.4 percent at eight to nine months postpartum. A prospective study of women planning to give birth in one of four participating New South Wales hospitals were assessed at 8 weeks using the EPDS (Johnstone, Boyce et al. 2001). Using an EPDS cut-off score of >12 to identify cases, 13.1 percent were defined as having postpartum depression at 8 weeks postpartum. Extensive studies have been undertaken of individual level risk factors for postnatal depression. Fewer individual-level studies of antenatal depression have been undertaken, but those that have, indicate that the implicated factors are different. To date there have been no multi-level studies and few studies explore the area, ecological and contextual factors that may impact on the prevalence of antenatal or postnatal depression in a community. In most previous studies personal, marital, relationship and family psychopathology factors are strong risk factors. The purpose of this section is to examine the published literature on perinatal depression to identify factors that might be relevant for building a conceptual framework that describes the mechanisms by which multi-level characteristics influence developmental and life-course outcomes. Depression during pregnancy - Systematic Reviews There have been few studies of the epidemiology of antenatal depression and no systematic reviews of risk factors. A systematic review of antenatal prevalence was published in 2004 (Bennett, Einarson et al. 2004). O’Hara et al found that the rate of depression during pregnancy was not significantly lower than the rate during puerperium (O'Hara, Neunaber et al. 1984). In a later study O’Hara (1995, pp 110-135) sought to identify the psychological and environmental risk factors for

70


depression during pregnancy. Depressed childbearing women tended to have lower SES, histories of past depression and alcoholism, less support from their partners, more negative life events, poorer social adjustment, and more dysfunctional self-control attitudes. In addition, several indices of lower SES (e.g. lower personal and spouse education, unmarried, being seen as clinic patient) and history of personal and family psychopathology were significantly associated with depressive symptomatology during pregnancy. O’Hara noted that all the classic risk factors [of postnatal depression] were found to be associated with depression during pregnancy. O’Hara also noted that several previous studies had found low SES to be associated with higher levels of depression during pregnancy (Zajicek 1981; Cox, Connor et al. 1982; Gotlib, Whiffen et al. 1989). He concluded that all the findings suggested that social stress may play an important role in accounting for both clinical depression and high levels of negative affect (ie. symptomatology). Bennett et al (2004) undertook a systematic review to estimate the prevalence of depression during pregnancy by trimester, as detected by validated screening instruments. There was a considerable heterogeneity observed in the results. The rate of depression observed in the first trimester (7%) of pregnancy was similar to that observed in the general female population, whereas rates during the second (12.8%) and third trimesters (12.0%) were nearly double the rate in the normal population. For women of low SES the meta-analysis rates for the second and third trimesters were 47 percent and 39 percent respectively when obtained by self-report and 28 percent and 25 percent when determined by structured [diagnostic] interviews. Based on the available evidence Bennett et al believe that low SES women are likely to experience substantially higher rates of depression than are women of higher SES. A French Canadian study sought to analyse the relationship between stressful life conditions, social support and depression symptoms among low SES pregnant women compared with higher SES pregnant women. They found that the rate of self-reported depressive symptomatology was 47 percent in low SES women compared to 20 percent in high income women. Analysis demonstrated that chronic stressors such as financial and housing problems, negative life events and inadequate social support were all linked to high rates of depressive symptomatology in pregnancy.(Seguin, Potvin et al. 1999).

71


Postnatal depression - systematic reviews Five systematic reviews have studied factors associated with postpartum depression (Beck 1996; O'Hara and Swain 1996; Wilson, Reid et al. 1996; NHMRC 2000; Beck 2001). Wilson and others (1996) undertook a systematic review (118 articles) of associations between psychosocial risk factors and adverse postpartum family outcomes including postpartum depression, child abuse, assault of women by their partner, marital dysfunction and physical illness. They stated that there was “good evidence” of an association between postpartum depression with poor marital adjustment or satisfaction, recent stressful life events and antepartum depression. There was “fair” evidence of an association with lack of social support, current and past abuse of the mother by her partner or past psychiatric disturbance in the mother. Socio demographic factors were not studied. A meta-analysis of 44 studies (Beck 1996) examined the relationship between postpartum depression and prenatal depression, history of previous depression, social support, life stress, child care stress, maternity blues, marital satisfaction and prenatal anxiety. Moderate to strong association was confirmed for all eight variables studied. The mean correlation coefficient (r) effect size indicator range for each predictor variable was: prenatal depression (.49 to .51), child care stress (.48 to .49), life stress (.36 to .40), social support (.37 to .39), prenatal anxiety (.30 to .36), maternity blues (.35 to .37), marital satisfaction (.29 to .37) and history of previous depression (.27 to .29). Socio demographic variables were not studied. O’Hara and Swain (1996) assessed the rates and risk of postpartum depression in a metaanalysis. Their study confirmed the association with life events, marital relationships, social support, and previous psychopathology. They also found a weak association with obstetric complications. They studied the following demographic variables: age, education, marital status, length of relationship, income, occupation, working through pregnancy, number of children and parity. Only income was significant across studies. The NHMRC systematic review (NHMRC 2000) grouped risk factors as follows: 1. Confirmed risk factors, with agreement from approximately 75 percent of reported studies, include a personal history of depression, depression during pregnancy, difficulties in the marital relationship, lack of support, and stressful life events

72


2. Probable risk factors, with agreement from 40 percent to 60 percent of published studies include a family history of psychopathology, single parenthood, severe “maternity blues”, personality characteristics, negative cognitive style, birth experiences and obstetric complications, partner’s level of depression, infant health, temperament and behaviour problems, genetic vulnerability and neurotransmitters 3. Possible risk factors, with very little and equivocal evidence available at present, include thyroid dysfunction, hormonal changes, early discharge from hospital, preterm delivery, breastfeeding, poor relationship with parents, bereavement, maternal age, parity, cultural issues, Aboriginality, living in rural and remote areas, difficulty adjusting to parenthood, childhood sexual abuse and physical illness 4. Possible protective factors include optimism and self-esteem, having a good marital relationship, increased availability of social support, and adequate preparation for the physical and psychosocial changes of parenthood. In her 2001 meta-analysis of 84 published studies Beck identified 13 significant predictors of postnatal depression. The mean effect size indicator ranges for each risk factor were as follows: prenatal depression (.44 to .46), self esteem (.45 to. 47), childcare stress (.45 to .46), prenatal anxiety (.41 to .45), life stress (.38 to .40), social support (.36 to .41), marital relationship (.38 to .39), history of previous depression (.38 to .39), infant temperament (.33 to .34), maternity blues (.25 to .31), marital status (.21 to .35), socioeconomic status (.19 to .22), and unplanned/unwanted pregnancy (.14 to .17). The results confirmed findings of her earlier meta-analysis and in addition identified four new predictors: self-esteem, marital status, socioeconomic status, and unplanned/unwanted pregnancy (Beck 2001).

73


3.5

Impact of Perinatal Depression

How does maternal depression cause detrimental effects to the foetus and newborn infant?” As noted above stress may play a role in the causative pathway to detrimental effects on the foetus and newborn infant. Beach and colleagues (2005) propose that the effects of depression during pregnancy on foetal development can be transmitted in any of 3 ways. 1. Foetal sequelae of antenatal depression are mediated by neurobiological substrates of depression, (e.g. glucocorticoids and/or proinflammatory cytokines), that can cross the placenta into the foetal circulation. 2. The foetus is indirectly affected via neuroendocrine mechanisms (e.g. placental hypersecretion of corticotrophin-releasing factor), whereby prenatal depression modulates the physiological maintenance of pregnancy and leads to preterm delivery. 3. The impact of antenatal depression on foetal development could be mediated indirectly by its impact on maternal behaviour (e.g. depressed pregnant women are more likely to experience heightened stress, decreased social support, poor weight gain, and are more likely to use tobacco, drugs and alcohol) Beach et al (2005) note that the pathways of foetal exposure in most cases remain obscure. They cite, however, several studies that have documented the association of depression during pregnancy with deleterious obstetrical outcomes such as preterm delivery, and intrauterine growth retardation (IUGR) (Steer, Scholl et al. 1992; Perkin, Bland et al. 1993; Orr and Miller 1995; Hedegaard, Henriksen et al. 1996; Hoffman and Hatch 2000). Children exposed to maternal stress, anxiety, or depression during pregnancy may also suffer adverse neuro-developmental sequelae that affect various aspects of cognitive and psychosocial function. For example, newborns of depressed mothers perform poorly on neurological examinations, exhibiting poorer motor skills, activity, coordination, and resilience (Lyons-Ruth, Wolfe et al. 2000). Infants of mothers experiencing antenatal stress and depression have also been found to have altered infant temperament (Davis, Glynn et al. 2007; Huizink 2008; McGrath, Records et al. 2008; Henrichs, Schenk et al. 2009). Alternative explanations for the findings exist including altered maternal perception of infant temperament (McGrath, Records et al. 2008).

74


Evidence also indicates that the neuro-developmental effects of antenatal depression persist beyond infancy (Stott 1973; O'Connor, Heron et al. 2002). Talge and colleagues (2007) in a review of prospective studies of maternal stress during pregnancy, found that children were more likely to have emotional or cognitive problems including an increased risk of ADHD, anxiety and language delay. These findings were independent of the effects of postnatal depression and anxiety. The authors drew on findings from animal studies to propose that the effects may be mediated by the activity of stress-responsive hypothalamic-pituitary-adrenal (HPA) axis and the hormonal end product cortisol is involved in the effects on both mother and offspring (Talge, Neal et al. 2007). An extensive literature seeks to explain these effects of antenatal stress and depression on the foetus and infant (Avishai-Eliner, Brunston et al. 2002; Seckl 2004; Field, Diego et al. 2006; Kapoor, Dunn et al. 2006; Young 2006; Wright 2007; Kinney, Munir et al. 2008; Lupien, McEwen et al. 2009; O'Donnell, O'Connor et al. 2009; Rice, Harold et al. 2009) The effect of maternal depression after delivery, on child development, has been more extensively studied. A meta-analysis of 46 observational studies of depressed mothers demonstrated a moderate association of maternal depression with negative (i.e. hostile, coercive) parenting behaviours and disengaged parenting behaviours (Lovejoy, Graczyk et al. 2000). That review found that the effects were strongest for studies of disadvantaged women and mothers of infants. Depressed mothers have also been found to be less likely to put their children in car seats, give vitamins to their children, talk with their children, or play with their children. There have been extensive studies that have shown that children of mothers who were depressed during infancy are more likely to have behavioural and psychological problems (Downey and Coyne 1990; Gelford and Teti 1990; Cummings and Davies 1994). Postnatal maternal depression has also been shown to be associated with impairment of cognitive development and secure attachment (Cogill, Caplan et al. 1986; Murray and Cooper 1996; Murray, Fiori-Cowley et al. 1996). A meta-analysis of the relationship between maternal mental health and infant attachment, encompassing 35 studies and over 200 mother-infant pairs, indicated that maternal stress and depression are associated with a greater prevalence of insecure infant attachment (Atkinson, Paglia et al. 2002). Martins and Gaffan (2000) propose that attachment may be one, of several, pathways by which maternal depression causes later childhood problems. They suggest that the other possible pathways may include common genetic inheritance and adverse social

75


circumstances during early life. In relation to the proposition that attachment as a causative mechanism, Martins and Gaffan note that it has been proposed that the depression prevents the mother from “interacting in an optimally sensitive and psychologically available manner with her baby. This is thought to disrupt the development of secure attachment between mother and infant which in turn sets the scene for later behavioural and emotional problems for the child” (Martins and Gaffan 2000, p 737). A meta-analysis of 19 studies undertaken by Beck (1995) found that postnatal depression had a moderate to large effect on maternal-infant interaction. Martins and Gaffan argued that only one of the studies reviewed by Beck strictly measured attachment. In their metaanalysis of seven studies they used a more strict definition of attachment and found that infants of depressed mothers “showed significantly reduced likelihood of secure attachment and marginally raised likelihood of avoidant and disorganised attachment (Martins and Gaffan 2000, p 737). Sohr-Preston and Scaramella (2006, p 70) note that the most consistent theme emerging from literature is that ”mothers experiencing postpartum depression are less able to meet their infants emotional and physical needs”. Their review, however, found inconsistent literature linking postpartum depression and children’s cognitive development. They concluded that “children most at risk for cognitive and language deficits are likely [to be those] of mothers exhibiting depressed mood throughout their early childhood years” (Sohr-Preston and Scaramella 2006, p 70). The evidence presented in this section, and the previous section, strongly suggests that perinatal depression (prenatal and postnatal) represents a child’s earliest adverse life event and carries profound implications for future life-course outcomes.

76


3.6

Conclusion

The purpose of this Chapter was to establish the significance of research in the field of perinatal social epidemiology and specifically the study of perinatal depression and perinatal adversity. The importance of early beginnings has been reviewed with consideration of the “early life hypothesis”, DOHaD and the “life course perspective”The literature review was informed concurrently by the qualitative and quantitative exploratory studies. Not all the phenomenon emerging from those studies have been reviewed in this Chapter. The key phenomenon reviewed included: foetal stress, biological programming, attachment, and stress. Others including social support, social position and marginalisation will be discussed in Chapter 8: Theory Construction.

77


78

Chapter 4: Individual Qualitative

Chapter 4: Qualitative Exploration at the Individual Level Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







79


4.1

Introduction

The purpose of this Chapter is to describe the qualitative exploration of individual-level factors that might be associated with the phenomenon of perinatal depression, perinatal adversity and their possible impact on developmental and health outcomes for the foetus and infant. The study was undertaken concurrently with both the literature review and quantitative studies which both informed, and were informed by the findings in this study. There has been little qualitative research undertaken in relation to perinatal outcomes and social context, relative to the high number of quantitative studies of individual level risk factors. Similarly there are few examples of qualitative research preceding the development of epidemiological studies of depression and context. Recent studies highlight the rich insights that can be gained from qualitative studies of psychosocial experiences of neighbourhoods and place (Airey 2003; Pretty, Chipuer et al. 2003; Attree 2004; O'Campo, Salmon et al. 2009). The main focus of the overall study is to examine the role of neighbourhood context in relation to the developmental origins of health and disease. It is not the purpose of the study reported, here to examine in detail phenomenon at the Individual Level other than to identify the possible mechanisms and pathways by which the neighbourhood and other ecological (Group Level) factors may operate. The qualitative findings are thus presented in two separate Chapters. This Chapter will focus on the “Individual Level” which for the purposes of this study includes the infant, mother and family members. Ecological findings are reported in Chapter 7.

80


4.2

Aims

The aims of this chapter are to: 1. Undertake a qualitative study of the individual and family level factors associated with perinatal depression and the developmental origins of health and disease 2. Use the findings to inform the emerging theoretical framework. The specific analysis questions are: 1. Do events during pregnancy and infancy influence developmental and life outcomes for the infant? 2. How are those influences mediated through the mother, father and family? 3. Are there influences that might protect the mother and infant? 4. Does maternal depression during pregnancy or after have a role in mediating these influences on the foetus and infant? 5. Why do some women get depressed during their pregnancy or shortly after the delivery of their baby? 6. What individual and family level factors influence the phenomenon of maternal depression during the perinatal period?

4.3

Methods

4.3.1 Theoretical Approach In Chapter Two I discussed the philosophical and methodological approaches used in this study. As stated earlier, the epistemology of both the Emergent and Explanatory Phases of this study is critical realist, and the ontology is also critical realist. The qualitative research described in this chapter is part of the mixed method research design and contributes data to the theory building. The qualitative study was conducted concurrently with both the literature review and quantitative studies. This concurrent approach was used to maximise integration of data collection and analysis.

81


4.3.2 Focus groups 4.3.2.1 Sample selection Qualitative research aims to gain a deep understanding of phenomenon by exploring the experiences of groups of people. Hence sampling methods are purposeful rather than random samples. Sampling aims to obtain the broadest possible range of knowledge, views and perspective on the matters under study. Guba and Lincoln (1989) argue that this is the preferred strategy of qualitative enquiry. Ideally the sample should be selected “theoretically”. This approach involves selecting subsequent participants based on analysis of the initial interviews. This is often not possible within available timeframes and resources. The focus groups were naturally occurring groups consisting of participants of existing “playgroups”. The “playgroups” were purposefully selected from: a) a community with dense housing, low socio-economic circumstances and high numbers of overseas born mothers, b) a community with predominantly single dwellings, mixed ethnicity and average socio-economic circumstances, and c) a community with predominantly single dwellings, low socio-economic circumstances and high numbers of overseas born mothers. The participants were mothers, fathers, or other carers of newborn infants, who might be expected to have insight into the impact of family and contextual environments on their own sense of wellbeing. The individual members of the focus groups were not purposively selected. 4.3.2.2 Attendance The focus groups were conducted at the normal times and place of an established mothers group. Subjects were invited to participate based on their membership of existing “playgroups” in the chosen study setting. Coercion was avoided by using Families NSW funded Child and Family Nursing services to invite the participants. Group members knew each other and anecdotal feedback indicated that participants appreciated the opportunity to discuss postnatal depression. A medical practitioner was available to discuss personal concerns and answer questions. Handouts on postnatal depression were given to participants. 4.3.2.3 Focus group procedure The groups were facilitated by a research assistant who was a medical practitioner. This avoided coercion and kept researcher distance. The focus groups ran for approximately one hour. The initial facilitators’ discussion guide is at Appendix B. Notes were taken

82


during the focus group including general information on the setting and participants. Digital recording was used. The issues covered included: 

Why some women get depressed



Whether there are more depressed mothers in some suburbs



Why there might be more or less depression in some suburbs



What the characteristics of those suburbs might be



Whether there are things at a city, state and national level that might increase or decrease a mothers depression.

The discussion guide for the third group was amended based on analysis of findings from first two focus groups, key informant interviews and themes emerging from the quantitative exploratory data analysis and exploratory spatial analysis. That discussion guide explored: 

Whether communities with high numbers of ethnic groups were likely to have higher rates of depression.

4.3.3 Key Informant Interviews When I initially entered the field I interviewed and sought advice from a number of international social epidemiology experts. Their contribution included comment and advice on the potential role of globalisation and cultural factors. Selection of participants for interviews was by purposeful sampling. Potential participants were identified based on: 1. Gender 2. Local government area where they worked 3. Experience with population subgroups, ie Aboriginal, CALD, “high risk” mothers. 4. Professional or industry background 5. The need to explore concepts emerging from the data analysis.

83


Code

Gender

Area of work

GA

M

TC VN

F M

Canterbury and Inner West Liverpool South West Sydney

Sh

F

South West Sydney

LC

F

Fairfield

JH JM

F F

South West Sydney Macarthur

Experience with group Practitioner

Professional Background Public Health

Clinical Advocacy

Nursing Medical Academic Health

Non English Speaking Non English Speaking Practitioner Practitioner

NGO NGO Medical

Table 2: Key Informants Characteristics 4.3.3.1 Approach Key informant interviews began early in the study process. Initially the interviews were open ended exploratory. The interviews informed the early inductive process and selection of empirical and theoretical literature to review. Written notes and memos were taken from the early interviews. Subsequently interview guides were developed and interviews were digitally taped. The Interviews were arranged by administrative staff thus avoiding coercion. The Interviews lasted approximately one hour and were undertaken in the subjects own office. 4.3.3.2 Interview Guides Interview Guides for the initial interviews were as for the focus groups and are attached at Appendix B. Open ended questions were used. The questions asked changed as analysis was undertaken and the conceptual theory emerged. Initial interviews were intended to inform both the emergent theory development and the review of relevant empirical and theoretical literature. Initial questions were: 

Whether events during pregnancy and infancy influence developmental and life outcomes for the infant?



How those influences are mediated?



What community and society influences might harm or protect the foetus and infant?



Information on theories and empirical evidence related to community and society influences on the mother, foetus and infant

84




Whether depression during pregnancy or after has a role in mediating adverse influences on the foetus and infant?



Why some women get depressed?



Whether there are more depressed mothers in some suburbs?



Why there might be more or less depression in some suburbs?



What the characteristics of those suburbs might be?



Whether there are things at a city, state and national level that might increase or decrease a mothers depression?

4.3.4 Data Analysis 4.3.4.1 General Thematic analysis was undertaken using coding, sieving, grouping and interpretation of data (Boyatzis 1998). The data analysis sought to identify and describe emergent concepts. The focus was on possible causes of high rates of depression in certain communities. Comparison of study suburbs was not a purpose of the qualitative study. Quantitative comparison of suburbs will be undertaken in Chapter 7. Data recording was digital voice recording and note taking. Data was transcribed and key word notes were taken as key concepts or ideas emerged during the interview. Digital recording was secured on a password protected partition of a network server. Transcripts for each focus group and key informant interview were prepared from digital tapes and checked for accuracy against the digital recording and notes taken during the discussions. The data coding, memos and analysis were undertaken by me using the Atlas ti software package. 4.3.4.2 Coding Codes are tags or labels for assigning units of meaning to the descriptive or inferential information collected (Miles and Huberman 1994, p 58). Miles and Hubermann (1994) describe an approach that starts with an organising framework for the codes derived from prior knowledge. In this inductive phase of study I predominantly followed the “grounded” approach first described by Glaser and Strauss (1967) and used minimal “preset” codes. Open coding was the predominant approach taken to initial coding (Charmaz 2006; Saldana 2009). The interview transcripts were coded line-by-line and paragraph-byparagraph. Each “incident” was coded into as many categories that it might fit during the

85


early stages of concept and category generation to enable maximum emergence of patterns and relationships (Glaser 1978). Preliminary data analysis was an ongoing process undertaken every time data was collected. This involved checking and tracking the data to see what was emerging and identifying areas that may require further follow-up and active questioning. The above process of engagement with the qualitative text was expanded as part of the mixed method design to include concurrent analysis of matters emerging from the exploratory analysis of quantitative data and from the literature review. The direction of the literature review was thus also informed by both the qualitative and quantitative exploratory analysis. For the second cycle of coding I used focused coding where the most frequent or significant Initial Codes were used to develop categories that were more selective and conceptual (Charmaz 2006, p 57). The coding process was considered finalised at the point when there was theoretical saturation. This was the point when no new concepts emerged from reviewing the data from the focus groups and interviews. 4.3.4.3 Thematic Analysis and Conceptual Mapping There are a number of approaches to thematic analysis including block and tile approaches and conceptual mapping. The advantage of the block and file approach is that it keeps fairly large chunks of data intact but the disadvantage is that it can be difficult to work with such large amounts of data. The conceptual mapping approach has the advantage of keeping a brief summary of the emerging issues (Grbich 2007, p 32). A conceptual mapping approach was used to develop causal networks as described by Miles and Huberman (1994, p 151-165). In this approach the causal network is a display of the most important concepts and processes and the relationships among them. The plot of these relationships is directional, rather than solely correlational where it is assumed that some factors and processes exert an influence on others. To be useful the causal networks are associated with text describing the meaning of the connections among the factors (Miles and Huberman 1994, p 153). The identification of “important” emerging concepts for focused analysis was based on a) density of coding to that concept or closely related concepts, b) density of

86


interrelationships and connections among codes, and occasionally c) cognitive reasoning. Some of the concepts judged as important were only mentioned by one person but reasoning indicated that they should be subject to focused analysis. In this later case, a focused network diagram might not have been drawn. Decisions regarding the nature of the theoretical links between codes was based on a) information provided by key informants and parents, b) prior knowledge of empirical literature, c) supporting information arising from concurrent literature reviews and quantitative studies, and d) cognitive reasoning. The most common relationship used was “associated with” where the relationship had no clear direction. The codes “is cause of” and “protects from” indicate a possible direction of the association. The term “is cause of” is a statement of tentative causal association arising from the data but subject to subsequent analysis. 4.3.4.4 Situational Analysis There are several approaches to analysing the conditions, context or situation that participants experience and that influence causative mechanisms or participants actions. Those approaches include the Conditional/Consequential Matrix, or its earlier variants (Corbin and Strauss 2008, p 94), Context Charts (Miles and Huberman 1994, p 102-105), the Research Map (Layder 1993) or Situational Analysis as recently described by Clarke (Clarke 2005; Clarke and Friese 2007; Clarke 2009). I elected to undertake a third cycle of analysis using the Situational Maps and Social Worlds/Arena Maps described by Clarke (2005). The situational maps lay out the major human, nonhuman, discursive and other elements in the research situation and “provoke” an analysis of the relations among them. The building of the situational maps utilised codes and analysis from the thematic and conceptual mapping together with personal knowledge of the field. Placement of the codes in the situation map was based on cognitive reasoning and experience. The social worlds/arenas maps graphically lay out the collective actors and the arenas of “commitment and discourse within which they are engaged in ongoing negotiation”. Thus “mesolevel” interpretations were undertaken of the situations influencing the experiences of mothers (Clarke 2009, p 210). The development of the social worlds/arenas maps was undertaken using logical placement of elements in association with each other. The placement was based on local knowledge, experience and reasoning.

87


4.3.5 Validation, Credibility and Rigour The reliability of qualitative research findings is sometimes questioned. Several recent publications have sought to provide guidance on quality within qualitative research (Tobin and Begley 2004; Kitto, Chesters et al. 2008).

Memo Writing and Coding Memos are the researcher’s field notes. I used them used to record what was heard, seen, experienced or thought while collecting and reflecting on the data (Groenewald 2004). Memo writing was also undertaken during the data analysis stage to record significant thoughts and interpretations. Memos were primarily conceptual in intent. They sought to tie together different pieces of data and to show linkages of data to the emerging general concept (Miles and Huberman 1994, p 72)

Integration The overall study design uses data from multiple sources including: published empirical research and scholarly writings of opinion, interviews, focus groups, and quantitative analysis of survey data. Different methods are used to analyse the data and thus a comprehensive understanding of the phenomenon can be achieved. This approach ameliorates the potential bias of using one method (Kitto, Chesters et al. 2008). The approach that I have taken, to both the integration of findings and theory development, is discussed further in Chapter 8.

Reflexivity Reflexivity requires demonstration of the researchers’ socio-cultural position and how their value systems might affect the research. I declared my philosophical position, prior knowledge, and value base with respect to social justice and child rights. I sought to “bracket” my prior understanding and thus allow understanding of the phenomenon to emerge through the voices of the informants (Creswell 1998). As noted above for the initial coding I made minimal use of prior codes and made maximum use of open coding.

Evaluative rigour Evaluative rigour refers to ensuring that the ethical and political aspects of research are addressed. In addition to proper ethics approval it also entails consulting with relevant community leaders in the design and conduct of the research. For this purpose I used my supervisors and professional colleagues from different agencies. Presentation of early results was also made to key Child and Family practitioners and leaders.

88


4.3.6 Ethical Matters Coercion and disempowerment of focus group participants was avoided by managing the process through the auspices of “third-party” play group organisers. Participation was voluntary, the purpose was made clear, questions were invited, benefits explained and signatures of both the researcher and participant obtained after there was agreement to participate. The Focus group participants were deliberately not drawn from currently stigmatised communities or neighbourhoods. The privacy and anonymity of the participants was maintained and the data will be destroyed after 7 years. Discussion with mothers regarding the adverse effects of depression on the foetus and infant could have caused distress. The information regarding the study also included information that had previously been distributed as part of the Beyond Blue and Integrated Perinatal Care initiatives.

4.4


The qualitative research results are reported thematically below. Findings were used iteratively to inform the literature review which was presented in the previous Chapter. Confidentiality is preserved for comments and quotes by only identifying them as mothers group or experts. Photographs taken in the study communities are presented in figures to illustrate some of the themes and issues that emerged during mothers focus group discussions and key informant interviews.

4.4.1 Impact on Foetus and Infant Much of what is known about the foetal and infant response to perinatal depression is empirically derived from biomedial studies. Literature relating to the foetal and infant origins of health and disease were reviewed in Section 3.3. The purpose of the expert interviews related to foetal and infant level concepts, was to inform the review of literature, enable concurrent comparison and to assist in the selection of infant and maternal level variables for exploratory data analysis (Chapter 5).

89


4.4.1.1 Foetal stress, damage and biological programming The concept of biological programming had been selected based on prior knowledge and my belief that early adversity had an impact on developmental and health outcomes. Stress~

Tobacco Smoking~

Alcohol Consumption

Toxins consumed

Drug Use~

Nutrition of Mother

is cause of is cause of

is cause of is cause ofis cause of

Cortisol levels~ is cause of

Foetal Stress~

is cause of

is cause of

is cause of

Foetal Damage

is cause of


"Biological programming"

is cause of

is cause of

is cause of

Poor developmental and life outcomes

Figure 14: Focused Network of Foetal Stress, Damage and Biological Programming

Stress and cortisol There was a strong view among the experts that foetal or intrauterine damage occurred as a result of foetal stress and related raised cortisol levels and that this was then translated into impacts on foetal brain development and physical health. A number of those interviewed cited animal experiments and believed that there was compelling evidence for the impact of foetal stress and damage on life outcomes. It was further stated that “stress is what produced the very high cortisol” and that “high cortisol levels impact on a range of DNA transcription by activating [some] genes and down regulating other genes.” Thus there was a suggested causal link between maternal stress, foetal stress and raised foetal cortisol levels.

Gene environment interaction One informant also discussed emerging research on the gene environment interactions citing evidence that “the [hormone] receptor population is also set genetically in response to whatever our intrauterine environment is”. This adaptive response to the intrauterine environment is described as a predictive adaptive response whereby “the developing organism senses its metabolic milieu and adjusts its physiological homeostatic set points

90


according to the environment it forecasts will exist after birth” (Gluckman, Hanson et al. 2007, p 7).

Maternal education and behaviours In addition to the impact of foetal stress there was a consistent view among those interviewed that foetal damage was also affected by maternal nutrition and environmental toxins such as tobacco smoke and alcohol. The view was that nutrition and stress operated independently but with some possible overlap. Gestational diabetes and iron deficiency were given as examples of how poor maternal nutrition might impact on the foetus. The role of nutrition was linked to maternal education and social or family support. As discussed later there was seen to be a link between “nurturing” the mother, her nutrition, and the health of the foetus. One expert cited work in a developing country where interventions that included education of girls and maternal nutrition resulted in increased birth weight. The overall theme emerging was that foetal damage had life long consequences and that that damage was principally mediated through foetal stress, nutrition, and chemical toxins. Some respondents believed that there was empirical evidence that might explain how those factors biologically damaged the foetus. There was also the view that the factors damaging the foetus were themselves caused by maternal stress, nutrition, education, lifestyle behaviours, occupation and exposure to environmental toxins.

Developmental origins of Health and Disease The foetal and developmental origin of adult disease was the common concept discussed by all key informants. Two informants drew a link between “foetal stress and damage” and “biological programming”. This theoretical concept of “biological programming”, as proposed by Baker (1992; 1993; 1998), suggests that nutritional and other environmental insults occur at critical periods of early development producing last organ impairments and metabolic and endocrine abnormalities. Section 3.3 summarises literature on our current understanding of how early foetal influences might impact on brain development, organ and endocrine systems (Barker 1992; Elo and Preston 1992; Barker 1998; Keating and Hertzman 1999; Shonkoff and Meisels 2000; Ben-Shlomo and Kuh 2002; Gluckman, Hanson et al. 2007).

91


4.4.1.2 Attachment and Nurturing Attachment and nurturing were not initially selected as prior concepts but emerged very early in both the expert interviews and literature review. The importance of parent-infant attachment is well established in contemporary developmental psychology as a “developmental pathway of major significance throughout the child’s life course” (Swain, Lorberbaum et al. 2007, p 264). Postnatal depression has also been shown to impact negatively on maternal-infant attachment (Martins and Gaffan 2000). Subsequent examination of the study EDA data set, identified three relevant variables for analysis which were highly correlated with increased EPDS scores.

Figure 15: Focused Network of poor attachment and nurturing

Mothers attachment Most of the informants and mothers’ groups stressed the link between postnatal depression and poor attachment and nurturing of the infant. Attachment was seen to have an important role in terms of brain development and impacts on the developing child. Several experts spoke of the negative impact of maternal depression on attachment and bonding. One respondent described the impairment of mothers’ interaction with the baby and the lack of appropriate affect. Consequently the infant would not receive appropriate

92


nurturing care such as getting fed on time, eye contact, “attachment and that kind of stuff that’s so important”. Another expert spoke of depression as a veil that prevents a mother from relating to the infant, understanding the baby’s cues, and connecting to the baby as an individual and as a person. The lack of eye contact was seen as important by another contributor together with “lack of appropriate affect, less interaction and less intellectual stimulation”. The long term consequences of poor attachment were said to be poor relationship, communication, cognitive and behavioural outcomes, child abuse, violence, and developmental delay.

Infant attachment A further consequence of poor maternal attachment was that “you will have a restless baby and insecure attachment”. The “restless baby” and insecure infant attachment to the mother would thus affect the infants’ temperament. The negative impact of altered infant temperament on postnatal depression was commented on by several mothers in relation. This sets up a complex situation where infant temperament may be a cause of postnatal depression and postnatal depression may itself contribute to altered infant temperament.

Family dysfunction Family dysfunction and “chaos” was also described as interfering with attachment. In this situation the baby gets “lost sight of in the whole thing”. The chaos in the family, and other things going on, affect how people relate to the infant. In this situation the physical care of the infant may be satisfactory but the need for psychological interaction “can be overlooked”.

Attachment starts during pregnancy Attachment was seen by one respondent as starting during pregnancy with a link between the events during 9 months of “incubation” and subsequent attachment. For this respondent, attachment was strongly linked to “increasingly intrinsically valuing the subsequent birth of the child”.

Breastfeeding and nurturing One expert spoke of the relationship between maternal depression and breast feeding. They believed that depressed mothers would be less likely to breast feed and that consequently there would be an impact on the developing brain. They related this not only to the “process and content of breastfeeding” but also to the impact of bonding, nurturing and “caring kind of effect” on the brain. They felt that there would also be a neglect and lack of bonding because of mothers’ depression. The respondent stated that there was “a

93


lot of data on breast feeding [and] cognitive and behavioural outcomes and possible links to chronic disease in adult life”.

Nurturing Bonding, attachment and the idea of “nurturing” were often discussed together in the literature. The term nurturing was used by one respondent in the context of breast feeding, and lack of bonding. The respondents seemed to draw a distinction between the psychological effects of attachment and bonding and the more physical aspects of the phenomenon called nurturing. Nurturing is seen to be also impaired by postnatal depression but respondents placed greater emphasis on the detrimental impact of poor attachment on infant development. Nurturing of mothers by society and family is a category that I have discussed in relation to the concept of “social support”. 4.4.1.3 Infant Temperament Infant temperament has been defined as the infant’s behavioural style and is how they behave in relationship to the environment and caregiving that they receive (Thomas and Chess 1977) cited by (McGrath, Records et al. 2008). Restless baby

is cause of

Infants affect their mother

is cause of

is cause of

Infant Behaviour

is cause of

Infant Temperament

is cause of

is cause of

Sleep deprivation

is cause of

Crying Baby~ is cause of

is cause of

is cause of

Feel bad mother

is cause of

is cause of

is cause of

High expectations

Stress~ is cause of

Postnatal Depression~

Figure 16: Focused Network of Infant Temperament The role that infant temperament might play in postnatal depression emerged early from the expert and mothers groups interviews. The subsequent exploration of relevant

94


literature is summarised in Section 3.5 and five variables measuring aspects of infant temperament were included in the EDA (Chapter 5). Mothers spoke of the difficulties faced when babies were not sleeping, crying or behaving in ways that they did not expect. One mother spoke of being so depressed and upset at everything because “I can’t get help with whatever he is doing, or he is crying”. Another spoke of the reality of lack of sleep, “not being able to control was he is going to do”, not being able to make the baby stop crying and consequently “wanting to cry”.

Link to expectations Mothers discussed infant behaviour together with comments about unreal expectation. Both mothers and experts felt that society and mothers expected to take home the perfect baby and everything was “going to be fine”. When discussing why mothers may get depressed one mother talked of the rude shock when motherhood was not as portrayed in the “Huggies adverts”. The reality was that her baby would not stop crying. She spoke of the frustration of not knowing what to do after checking the “three things they told me to check”. She had the expectation that when she got home she would know what to do and “it will be all good”. But the reality was that “it doesn’t work like that”. The comments on unreal expectations were associated with discussion about “lack of control” which I discuss later.

Link to attachment As discussed above, consequence of poor maternal attachment was that “you will have a restless baby and insecure attachment”. This then sets up a complex situation where infant temperament may be a cause of postnatal depression (and poor attachment) and then may further contribute to altered infant temperament.

95


4.4.2 Maternal and Family Level Themes 4.4.2.1 Unplanned Pregnancy and Sole Parent The phenomenon of unplanned pregnancy in relation to maternal depression was thought to be significant, particularly if the pregnancy was unwanted or there was ambivalence toward the foetus and infant. Also important was the implication that the mother might be unsupported. Sole parenthood and unplanned pregnancy were also found to be significant in the logistic regressions reported in Chapter 5.

Figure 17: Focused Network of Unplanned Pregnancy

Disadvantage Those interviewed spoke of a link of unplanned pregnancy with disadvantage, maternal education, violence and behaviours such as tobacco smoking and higher alcohol consumption. “She may be pleased or not pleased at all that they are pregnant. In poorer areas, [there are] generally higher rates of unwanted pregnancies, less use of appropriate contraception, [and] that is likely to contribute to depression”.

Family and cultural acceptance

96


Several respondents considered that unplanned pregnancy would be associated with access to support from family members. In some situations there are strong links between unplanned pregnancy and cultural acceptance. It was felt that in some families and cultures there would be a lot of support and the family belief would be “oh my god, you are pregnant, this is god's gift to the family”. In this situation the mother would be supported. But in other families this is not the case. Support would be conditional on whether the pregnancy was expected and planned. Cultural issues were considered to be important with some parents not giving support to their pregnant daughters if she is not married. Informants felt that the degree of support given may depend on the nature of the communities that mothers lived in. It was also considered possible that unsupportive cultural beliefs and attitudes may also exist within some mainstream cultures. In other situations the unplanned pregnancy was considered to be linked back to the lack of support given to women in that family or community. The lack of support from a partner or parents was considered important. Consequently there was a strong link between unplanned pregnancy, sole parenthood and the level of support available in communities. In relation to this some experts talked of higher rates of unplanned pregnancy and sole parenthood in depressed communities. Some felt that the pregnancies had occurred because the women were “just looking for a relationship”. Women in this situation were described as “having nothing to look forward to” and that when they got pregnant they had no plans as to how they would raise the child.

Altered partner relationships Mothers in the focus groups also spoke of the impact of an unplanned pregnancy on their relationship with their partner and on their mental preparedness. In this situation the partner may not be prepared and although they are with them, they are not supportive. In some cases the relationship changes. Mothers were clear that “the mother is always seen as the one that bears it all”.

Social exclusion Experts made the link of unplanned pregnancy with depressed communities paralleled with low levels of education and an intergenerational cycle of poor educational attainment and early pregnancy. In these communities it was considered that children weren’t sent to school, had less education and consequently became pregnant while they were still young. As young parents, they were often in the same situation as their parents had been.

97


With a lack of social skills learnt from school, they would have difficulty meeting friends and may be left unsupported to raise the infant on their own. Subsequent education and employment would be compromised. This view was echoed by mothers who felt it would be hard for sole parents who didn’t have an education to find a job or get to TAFE. They felt that these mothers would not have the “social skills to make friends with nice people, or even have someone to look after your baby while you go to TAFE and try and educate yourself”. Unplanned pregnancy, low income and poor education were said to be associated with staying home and not being able to network with friends. One mother noted that: “If you have no one there, you are stuck. You have to stay at home with them. If you don't have the money to go out for lunch with friends. Like that is a big thing for me, I have to be able to go out and see my friends have lunch with them or what ever. I have to get out of the house at least once every day. If you don't have the money to go and do that then I guess you will feel a bit snowed under.”

There may often also be a stigma associated with being unmarried and pregnant. That stigma may be related to the social and cultural background of the mother and/or community she lives in. In these situations the optimal caring of the baby is compromised by the community’s beliefs. Thus unplanned pregnancy was seen as the cause of both antenatal and postnatal depression. It was considered to be more common in “depressed communities” where there would also be low maternal education, increased tobacco smoking and low income. A stigma might be attached to having an unplanned pregnancy in some circumstances that could be culturally or ethnically related. Importantly where a mother was not married different cultures may respond in different ways.

98


4.4.2.2 Support for Mothers Care and support of mothers was a strong theme emerging from both experts and mothers groups which is consistent with the findings of systematic reviews of perinatal depression literature discussed in Sections 3.4. The exploration data analysis in Chapter 5 also identified the importance of social support, emotional support and practical support in the Categorical Principal Components Analysis, Exploratory Specification Search and the Logistic Regression Studies.

Figure 18: Focused network of “Support” Support has a number of sub-categories including family support, partner support, emotional support, and support from services through play groups, visiting nurses or midwives and phone calls. Support networks were mentioned by both experts and mothers and usually included family members and friends. A strong view of experts was that support of mothers during and after pregnancy protected them from stress and depression and consequently also from alcohol consumption and smoking. This was expressed by one expert as “well through networks and supports, reduced stress, and having additional support, reduction of stress is likely to result in less alcohol consumption, less smoking and less depression”.

99


Link to nurturing The importance of caring for mothers during pregnancy was seen as being linked to the nurturing of both mothers and babies. It was noted that “if the mother is properly cared for and well acknowledged, nutritionally the baby is healthier…. So how well the mother is looked after, not only emotionally, physically, environmentally and nutritionally is how well the foetus develops during pregnancy.” There is thus a strong link between support of mothers, “nurturing mothers” and then the nurturing of the unborn infant.

Families The care and support of mothers differs among families and may be related to whether she is married. There is therefore a link between support for mothers and the above concept of unplanned pregnancy and sole parent. For some families pregnancy was a gift and there is a recognition that priority be given to supporting the mothers

“The mother needs support that is the main thing. … Within some families there is a lot of support and some families the belief is like ‘oh my god, you are pregnant, this is god's gift to the family' and its very much appreciated and the mother is supported.“ But other families have different limitations and different support mechanisms. In other families the response may be different if the pregnancy is not planned or not within a marriage. One expert noted that “a lot of young mothers who get pregnant don’t have that support from their partners or their parents. So it depends what level of support there is.”

Support of partners The mothers spoke of the importance of their partner being there for them and also wondered what it must be like if the mother was a single parent. One mother expressed this as:

“but even then I find it, looking at me, I go then, thank God Matt's with me, so he can give me the hug, during he night when I am so depressed you are really upset about everything because I can't get help with whatever he is doing or he is crying or whatever. Then I think a single parent, I go how do you do it?”

100


Support of “mum-type” person One expert spoke of looking for support from a “mum-type” person –

“she hasn't got a social network, if she doesn't have parental support, like one of the things we look for is who is the mum-type person who is looking after you in pregnancy. If she hasn't got that support and that mum-type person who is looking after her in pregnancy”. Another expert expressed the same idea as – “does she have sources of support, informal and formal? Has she got a mother or maternal guide mentor or support from a wise woman close by?”

Support Networks and Mothers Groups Mothers were interviewed at a mothers group. Consequently those mothers had experience of support from a mothers group. They placed strong importance on the importance of the groups for establishing and maintaining support networks. One mother noted that support networks may include family, friends and neighbours - “I think just having support networks whether it is something like this or whether it is your family, friends, your next door neighbour, again even if you have all those things it is no guarantee that you won't suffer from PND”. Another important source of support are parent groups similar to those used for the focus groups. The parent groups provide a useful support network. One mother stressed their importance and linked this to filling a gap when there is a lack of support available from family - “…more groups. This is fantastic. I didn't have a group like this when I had my first, and didn't have support. You are crying for ... family and that sort of thing. Having these networks has helped me a lot more.”

Access to support from services Access to mothers groups run by the nurses was thus seen as an important source of support. There is an obvious link here between support and access to services from child and family nurses and other family support services. Access to phone calls from service providers was seen to be another important source of support. Mothers made several comments regarding the importance for calls from child and family nurses and midwives. The importance of phone contact from health staff was stressed by several mothers – “even just a phone call will be nice every so often”. Just from the community nurses, just to check up, how are you going? Do you need me to come in, rather than waiting for us to

101


call them? If they could call us. Just a little show in the right direction”. Another mother stated “I think any support you get. Just a phone call that lasts 5 minutes. That is great. It just shows that somebody is there and taking that interest in what is happening with you not only what is happening with the baby”. Phone support from the midwives was also important to the mothers - “I had my midwife ring me couple of times, that was nice as well. She rang up until he was about 2 months old and then stopped calling me. That was good. Because you don't actually want to ring someone, and say what do I do with this. Ring maybe once a week or once every couple of weeks up until a couple of months old. Then you can have all these questions to ask them, they might just catch you before you snap off into that depression”.

Religious Support The role of religious organisation was briefly explored by some experts. One expert felt that there was evidence that people in certain religious or spiritual groups had more support and consequently better health outcomes. Conversely a mother felt that religion can be a barrier to getting support – “sometimes we can’t mix with other people, you have to move with your own religion [or] culture. … or religion not allows you to talk about private things”.

102


4.4.2.3 Access to Services Access to services and amenities was a well grounded theme at both the individual and group levels. An individual level variable is a characteristic of that individual or in this case their family or household. Individual level access variables might include access to information, car and phones. Living in a community with few amenities or services was considered a group level phenomenon and is discussed later in Chapter 6. Access to child care Access to baby checks

Access to shopping centres

Access to books

is property of is property of is property of is part of

Access to services

Isolation

protects from protects from

is property of

Accessis cause of contradicts

Services not talking to each other

transport

is part of

is property of

is associated with Courage to make

is associated with

Link between comminication isolation and network

Location of services in CBD

contradicts

call for help

is property of is property of

Geographical isolation

is property of contradicts is associated with is associated with is property of

Distance from resources

Information

Access to medical care Service access Knowledge of services

Amenities

Coordination of services

difficulties accessing services

Figure 19: Focused Network of Access

Information Some mothers may lack information and it was felt by mothers that it was the governments role to give mothers access to that information. A mother in a mothers group felt that “people aren't aware of what is available. So may be government needs to let any mother know that type of treatment is available. I don't know of any type of treatment yet. We are not aware of it. And whether we can access it or not”. Another mother felt that access to books and the internet was important. The role of mothers groups and friends for sharing

103


information was raised by one mother –“talking to other people you know who are going through the same thing … but when you are in a group and actually get talking”. The importance of information was supported by an expert who when discussing the importance of linking to services noted that “not all of them have the knowledge to be linked”.

Transport Transport is a common theme at both individual and group level. Not having a car is an access related variable and in the bivariate EDA lack of car access was associated with postnatal depressive symptoms. One of the mothers linked transport problems with isolation “I think transport problem is significant, it is hard to get out and about, ---- I suppose there is isolation”. This view was supported by four experts who spoke of the importance of having cars for “getting out to other people”. Public transport was seen as a possible barrier – “If it is just an effort if you have baby and a toddler and have to pack up a pram and wait an hour for a bus to go somewhere. Then you can only do limited shopping, you have to carry it home”. Another expert felt that there was an expectation that people have cars. She had similar concerns regarding the effort to pack up “babies and kids” and take them out.

Phone Access to a phone was considered important to mothers and that phone may be at a neighbour’s phone. One mother stated – “or you speak to someone you know, who has just rung them. That is what I have done. I haven't rung myself yet but I keep asking a woman who does”. Another mother recalled “… a lovely lady at three in the morning, after my son had this massive projectile vomit at about 2 weeks. Something out of a horror movie, I thought this is not right. Yes, just to be able to talk to her at that time knowing that somebody was there. … Even when I was crying to them on the phone they were so supportive and sympathetic.”

104


4.4.2.4 Mothers “Loss of Control” A number of key informants and mothers themselves spoke of “loss of control”. This “loss of control” was felt to contribute to a mothers development of depressive symptoms. Three main sub-categories were identified related to personal relationships, birth of a new baby and conflicting advice. There is also an association between “loss of control of your life” and “expectations lost”. “Expectations lost” emerged as an important theme from later analysis. Stress~ is a cause of

is a cause of

is a cause of

Sleep deprivation

is a cause of

Expectations Lost

is a cause of

is a cause of

is a cause of

Crying Baby~is part of

is a cause of

is a cause of

is a cause of is a cause of

Infants affect their mother

Loss of control of your life

Restless baby

is a cause of

is a cause of protects from is associated with is part of

High expectations Expectations

is associated with

Partner Problems

is associated with is a cause of protects from

Significant new responsibility

is associated with is associated with

is a cause of

New Baby

is a causeisofassociated with

is associated with is a cause of

is associated with

Relationships is associated with Unstable

is associated with

Mother terrified Conflicting advice

Relationship with is associated with partner Support

Figure 20: Focused Network of Loss of Control

Control in relationships For the experts this “loss of control” related to whether the mother was supported or “unsupported” in her personal relationship. One expert expressed it as “Not purely but mostly feeling safe healthy supported and connected”. Another expert related it to unsafe personal relationships – “If a woman is feeling out of control so she is in an unsafe, unstable or violent relationship, what are the internal chemicals that are going on that impact negatively?”

105


Birth of a new baby For mothers the birth of new baby brought with it a loss of control often resulting in change in their relationship with their partner. The loss of control was expressed as a complete change in their sense of personal self control – “once I had the baby I didn't have that control and that affected me the most”. Mothers spoke of a sense of letting themselves go and not dealing with their own personal needs. Consequently “things started to build up”. One mother spoke of “complete loss of control of your life”. She reflected that it is not possible to prepare emotionally for the loss of control when she had previously been in control of her life and able to plan ahead. With the birth of the baby “now you can forget it”.

Conflicting advice Advice given to mothers by “western” trained workers sometimes conflicted with advice that mothers received from their own communities. One key informant identified this as cause of stress for mothers. She spoke of the “individual strengths” needed by mothers to stand up for what they wanted for themselves and their infants. Speaking of her personal experience with mothers she identified increased stress at home when mothers didn’t have the “power to say that this is what I want for my child”. In these situations she felt that mothers didn’t know who to listen to and they didn’t have the “individual strengths to stand for what they want or they don't know how to form their own judgement [about] what is best for their child”.

106


4.4.2.5 Isolation Isolation was a concept spoken about by experts and mothers. Isolation and having “no support” are closely related. “No support” was significant in the logistic regression studies reported in Chapter 5 and 7. Isolation

protects from

Social Support~

is cause of

Don't know people


is associated with

Can visit their mother

is cause of

protects from

is cause of

is cause of

is cause of

contradicts protects from protects from protects fromis cause of

Pregnancy

Home visiting

Don't know your neighbours

Antenatal groups

No Support.~

transport Belonging to community

is cause of

Access

Figure 21: Focused Network of Isolation Related to this was the concept of marginalisation which is discussed further in Chapter 6. One expert felt that “marginalisation probably is a big one.” By marginalisation he described the mother as not being “accepted as part of main stream”. Marginalisation was described as being possible in more than one level - within family, community, or country. He felt that marginalisation operated through stress. Thus he proposed that there was a link between marginalisation, stress and depressive symptoms“. A focused network of marginalisation is illustrated in Chapter 6.

“Not belonging” “Belonging to the group” emerged as an idea that might be important to consider in relation to social and ethnic groups. Thus mothers who were not part of the main group could become marginalised and this could impact on their mental wellbeing. In this sense mothers would feel like “they didn’t belong” and if they were marginalised from the main group “they would not do well”. This same expert felt that the stress caused by marginalisation would result in raised cortisol levels. Although “marginalisation” was not used by other experts or mothers, there was a consistent use of “isolation” to describe a

107


similar phenomenon. Mothers spoke of isolation when a mother was not of the same language group – “if they don’t know your language they are just isolated”.

Poverty, Single Parent and Access Marginalisation could also occur by being isolated through the impact of poverty, single parenthood and lack of transport. An expert when speaking of the impact of poverty described the impact as something other than survival. “Not survival it’s something else. It’s just marginalised being so disadvantaged”. Isolation was spoken of by several experts and mothers as relating to family relationships and in particular lack of support. “Again I am going to hype back to relationships. Is she isolated? Is she vulnerable? Is this a single pregnancy?” One of the mothers related isolation to transport problems and the difficulties of getting out and about. This importance of transport was highlighted by two experts who both talked of the isolation that occurs when there is only one car in the family and that car is being used by the partner.

Community Safety Community safety emerged as sub-category for isolation. Community safety is a feature of “depressed communities” and is discussed later in Chapter 6. Three experts spoke of the isolation that can occur as a result of not feeling safe in a community. If mothers do not feel safe enough to put the baby in the pram and walk to the park then they will not be able to get out of their house and will feel more isolated. In this situation mothers may have their doors locked and will not feel safe to sit on the front veranda or to let their children play in the yard. Mothers were also concerned about community safety and linked it to isolation – “If you want to go the park for example and there are some violent people then you choose not to go to the park. Then you are isolated.”

New neighbourhoods The isolation experienced by mothers may be experienced by those not living in poverty or marginalised by society. One expert spoke of the financial pressures experienced by many families and the isolation experienced by living in an area where they do not feel welcome or in a new area away from their parents or usual “social base”. Some of these mothers may be career women who have had high expectations who suddenly, with the new baby, have “things crashing on top of them and not going the way they want. They don’t have social support, they don’t know the people in the neighbourhood and things can isolate them even further”.

108


Linkages Isolation would seem to be the result of a number of processes including marginalisation from the group for social, language or cultural reasons, social exclusion as a result of poverty and sole parenthood, and isolation as a result of lack of transport. Isolation and marginalisation were seen by experts and mothers to be potential causes of depressive symptoms and one expert suggested that marginalised mothers would have raised cortisol levels.

109


4.4.2.6 Stress Previous studies of perinatal depression have consistently found stressful life events to be risk factors and few studies have found no association (see Section 3.6). It was not surprising therefore, that the phenomenon of stress was mentioned by most experts and was strongly implicit in responses from the mothers groups.

Figure 22: Focused network of "stress" The term stress was only mentioned once by a mother when speaking of postnatal depression she said that “you have more stresses in your life, financial, relationships, loss

of social income”. But I considered that stress was often being talked about when mothers were talking of “losing control”, “crying over the phone”, “something out of a horror movie”, and “I am going to be graded”. Stress was considered by experts to be a cause of both antenatal and postnatal depression. Two experts also considered that stress would have a detrimental impact on the foetus with one expert suggesting that it may also affect foetal growth. This view is supported by the studies by Glover and O’Connor (2002) and Huizink, and colleagues

110


(2002) which suggest that maternal stress during pregnancy is a predictor of both foetal growth restriction and negative temperament (irritability and negative emotionality) and anxiety in children. Some of the predictors identified by experts and mothers included: lack of support, poverty, isolation, living in “depressed communities”, conflicting advice, domestic violence, drug problems, fractured relationships, arranged marriage, pregnancy, traumatic experience, loss of family members, working and travel to work. As discussed elsewhere support was felt to reduce stress.

Adversity and Life events Adversity from what ever source causes stress and was linked to depression. Both experts and mothers spoke of a wide range of causes of adversity including cost of living, crowding, violence, crime, and drug problems. One expert linked this “chronic” form of adversity to lack of control of environment, financial and violent situations and suggested that the adverse outcomes were “mediated via all the stress effects inside them” Stressful life-events clearly overlap with adversity but may be of shorter duration. One expert spoke of life events that would bring on postnatal depression and suggested that mothers should be asked “what has happened in the pregnancy?” Clearly pregnancy itself is a stressful event which when coupled with other events such as violence, and fractured relationships would pose significant stress to the pregnant woman.

Marginalisation, Isolation and Loss of Control As discussed above there is a link between stress than the earlier themes of marginalisation, isolation and loss of control. Stress may be the result of these processes and through stress the psychological and physiological responses may be mediated.

Impact on foetus and infant One expert saw that the link was not only through the psychological responses of the mother but also physiological. Consequently there was a clear implication that stress would have a physiological impact on the foetus and infant. Others saw the link as being psychological with an impact not only on depressive symptoms but also life-style behaviours such as alcohol, drugs, smoking and nutrition.

111


Support as protective There was a constant theme that support somehow reduced stress and would therefore be protective of the psychological and physiological consequences of stress. One expert put it this way “well through networks and supports, reduce stress, and having additional support reduction of stress is likely to result in less alcohol consumption less likely that smoking and less likely to be depression”. Coping resources are also important, described as “not having enough assets of social support or living in an area without social inclusion, so there are not many opportunities to get together with many people”. The suggestion is that getting together with people provides support and will be protective.

Psychological resilience The response of mothers to stress might be also be affected by her resilience. That resilience and the chances of depression were linked by one expert to previous childhood experience. They suggested that the level of resilience and “how much they can cope with stress” is because of their childhood experience and consequently they may have a long period of depression.

Impact on child development The life-course consequences of maternal depression and stress working together was recognised by one expert who felt that “depression is obviously a huge influence on kids development” and “that stress probably works in with depression, or if not depression poor mental health or poor coping strategies”. This impact on child development was also seen by that expert to be reversible through postnatal interventions directed at mother-child interaction.

Linkages Stress was seen to be closely linked to both adverse and protective factors that may impact on maternal depression. Importantly stress was seen to be the possible mechanism by which both maternal physiological and psychological responses were influenced. Stress would thus be central to a theoretical model of neighbourhood context and the developmental origins of health and disease as associated with maternal depression.

112


4.4.2.7 Class Class, social position and social hierarchy were discussed by several key experts in the context of stress and marginalisation. Social position has also been discussed in the literature in relation to social class and occupational class. One expert saw stress mediating itself through social position. They saw stress as a fairly common blunt outcome for “a whole bunch of different ways of getting there”. Cortisol levels~ is a cause of

Class is associated with

is associated with

is associated with

Hope

is associated with

Absolute bottom of the pile

is associated with

is a cause of

is associated with is associated with

Gap between rich and poor

Choice to live in is associated Survival with that community Expectations Lost Unemployment

Occupation

Lack of hope

Figure 23: Focused network of Class Through the data I got the idea that social position might give some resiliency to stress and adversity. One expert analysed this in terms of people’s perception of their existence and optimism for the future. They stated that “there is lovely work showing that if you are optimistic about life, even if you are deluding yourself you live better, you have less disease, than the realist who is depressed about life”. Another expert was clear that social hierarchy overrode other risk factors. They linked the social phenomenon to hormonal or endocrine changes and suggested that “where you sit in the hierarchy determines your net cortisol secretion rate”. But the same expert was also clear that the social position was also mediated through other means such as housing, threat of violence, financial stability, access to books, family support and positive words and language.

113


4.4.2.8 Being Broke and Poverty Both experts and mothers spoke of financial hardship as a cause of postnatal depression. Experts and mothers felt that living in “depressed” and disadvantaged communities was associated with increased depressive symptoms. This is discussed further in Chapter 6. Financial stress at the individual family level was also considered important. This view is consistent with findings of the systematic review of postnatal depression undertaken by O’Hara and Swain (1996) who found that income was significant across all the studies they analysed. In this study I have also found that family financial situation is significant correlated with EDS >9 and EDS > 12 (see Chapter 5). Vulnerable Social isolation is cause of

No car~

is associated with

is cause of

Isolation Lack of phones

is cause of

is cause of

is associated with is cause of

is cause of is property of is cause of

Stress~

is cause of is property of


Poverty~ is cause of

is property of

Unplanned Pregnancy

"Depressed" Community~

is associated with is cause of protects from is cause of


is cause of

is cause of

Unemployment

is cause of

Mothers Education

financial support

Well priced food

Fathers Education

Figure 24: Focused network of Poverty Mothers were very clear that postnatal depression could be the result of “being broke”, “economic factors” and “their income”. Mothers spoke of not being able to provide as much for their baby, “things like nappies and formula and that type of thing”. Mothers spoke of the struggle to cover basic living needs after the baby was born. Their partners would look to “pick up extra overtime” to cover mothers being off work. The financial stress of losing an income meant that many would struggle to cover the mortgage, rent,

114


food and “basic living”. The struggle to cope was clearly a cause of stress for the mothers. When asked if there was anything that Government could do to assist, one mother suggested “give us a house cleaner once a week – someone to clean my shower”. They did not want to be “bludging off government” but strongly felt that family assistance should be for everyone. The stress for many mothers was related to living beyond their means at the time the baby was born. The importance of financial pressures was also recognised by experts interviewed. Some spoke in terms of “absolute poverty”, having adequate shelter, and “survival”. Comments on financial difficulties were often included with other adverse influences such as “toxic environments”, and violence. Experts also often commented on education levels of parents – “the whole nuclear family is flawed unless you have an extraordinary amount of intellectual and financial reserve”. Sole parents were considered to be particularly vulnerable often having no transport, less money, living in disadvantaged suburbs and not being able to “buy happiness”. Financial resources were considered to “play a huge part in anyone’s system”. If mothers don’t have money they don’t know where the next meal is coming from, are struggling to pay the rent, fill the car with petrol and “juggling to get to places”. Lack of financial resources was described by one expert as being something other than just “survival” – “It’s just marginalised being so disadvantaged or living on the poverty line”. She described women who had nothing to look forward to “so why wouldn’t they be depressed”.

115


4.5


In the previous section I described the concepts that emerged from the qualitative data a

priori. In this section I will use two kinds of situational analysis, situational maps and social worlds/arenas maps, to further “open up” the data and to interrogate in a fresh manner (Clarke 2005). These two approaches are analytic exercises that contributed to the description of phenomenon and abductively commenced the generation of theory. The analysis worked with the coded data and diagramming described in the previous section. As with the previous section the situational analysis drew on findings from the literature review and quantitative analysis. This adds to the richness of the data available for analysis.

4.5.1 Situational Analysis Mapping The messy working maps described by Clarke were used extensively throughout the analysis but are not presented here. The more structured map that (Figure 25) presents the data in terms of the situational elements such as individual and collective human actors, silent actors, discursive constructions, political elements, spatial elements and major issues and debates. Figure 26 focuses this analysis on the mothers close home and neighbourhood arenas. The contribution to her situation of wider arenas is expanded in Chapter 6. The situational analysis mapping and analysis of relations did not contribute significantly to the analysis. Most phenomena and their relationships had previously been identified but as a result of the analysis the following concepts were strengthened and drawn together: 1. Expectations and dreams 2. Marginalisation and being alone 3. Loss or absence of power and control 4. Support and nurturing. The analysis highlighted the importance of other arenas, processes and structural elements that are discussed in Chapter 6. Findings of the analysis were then used to inform the building of the Social worlds/Arenas Map and together both analyses contributed to this Chapters overall findings.

116


INDIVIDUAL HUMAN ELEMENTS/ACTORS

NONHUMAN ELEMENTS/ACTANTS

Unborn infant (foetus), Infant Mother Mothers’ Partner Family members, sister, grandmother Mothers friends Midwife and child and family nurse Neighbours Birth trauma

Huggies advert Television Cars, bus, train Phones Shops and Malls Call centres providing advice Information, books, web sites Physical safety of neighbourhood

COLLECTIVE HUMAN ELEMENTS/ACTORS

IMPLICATED/SILENT ACTORS/ACTANTS

Social networks Groups of mothers, Play groups Hospitals, Obstetric Services Child and Family Nursing service Non Government Organisation helpers Church visiting Mosque visiting

Employers Government, Centre link Housing, Community Services Council Land owners Television producers Advertisers Producers of products for infant care and mothers

DISCURSIVE CONSTRUCTIONS OFINDIVIDUAL AND/OR COLLECTIVE HUMAN ACTORS

DISCURSIVE CONSTRUCTION OF NONHUMAN ACTANTS

Family dysfunction Family help Fear Foetal stress, biological programming, life course Hope, purpose Cleanliness Getting graded by the nurse Feel bad mother Giving up career Happiness Breastfeeding is good “Broke” “Battlers” “ “Bottom of pile” “Blues” Isolation Loss of control of your life, Loss of identity Mental health of mother Nine month incubation and bonding nurture “Its just the hormones” Psychological resilience Survival Social hierarchy or Class, poverty Drug centre of Sydney Stress Expectation of mothers and society Conflicting advice Predictive adaptive response of foerus Unhealthy life-style behaviours Partner violence and drugs Not welcome

Estates with new houses are good Apartment living is not good “toxic community” Unreal advertising of motherhood Crowding of some homes Depressed neighbourhoods Estates of bliss Hot houses Cold houses Mouldy houses Damaged houses Lack of sunlight

117


POLITICAL/ECOMONIC ELEMENTS

SOCIOCULTURAL/SYMBOLIC ELEMENTS

Safety from violence of community Marginalisation of poor Marginalisation of ethnic groups Region is relatively poor Differences between rich and poor

Religious groups have different norms Racism Ethnic segregation

TEMPORAL ELEMENTS

SPATIAL ELEMENTS

Families NSW Programme started 2000 Integrated perinatal care started in SWS Innovative programmes started in SWS Traumatic pregnancy Previous losses

Distanced to amenities Variation in distribution of race and religion Movement of amenities to big centres New Estates with Nice Big Homes Apartments House Neighbourhood Suburb

MAJOR ISSUES/DEBATES (USUALLY CONTESTED)

RELATED DISCOURCES (HISTORICAL, NARRATIVE, AND/OR VISUAL)

Teen pregnancy is bad Pregnancy out of marriage is bad Unaffordable housing Homelessness Mums need to work to survive

Teen pregnancy is bad Pregnancy out of marriage is bad Outsiders vulnerable Every one will own a home Every family will own a car A nice house looks like this A bad house looks like that.

OTHER KINDS OF ELEMENTS Stress causing increased cortisol Biological Programming Gene environment interaction Attachment – mother Attachment - infant Nurturing of mothers Nurturing of infants Infant temperament Family dysfunction Loss of expectations Loss of control Unplanned pregnancy Parent support

Figure 25: Ordered Situational Map: Maternal Depression, Home and Neighbourhood

118


4.5.2 Social Worlds and Social Arenas

Figure 26: Social worlds/Arenas Map of Mothers Home and Neighbourhood What is this map telling us? Firstly there is the domestic or home social arena drawn as the largest dotted square. The dotted line means that the home is permeable to multiple actions. The arena is made up of a few actors: mother, baby, father and others such as mothers’ sister, mother, midwife, visiting nurse, and friends. The home is the domain where mother and baby spend most of the day. It is here that mothers receive, or don’t receive their closest emotional support. It is also here that control is lost and expectations are most manifest. Loss of control and failure of expectations may be manifest with a crying, unsettled and sick infant, unsupportive partner and increased financial stress. The expectations are that the house, baby and relations will be perfect. Those expectations are raised through family, friends, services, radio, magazines and television. But it is here that expectations are sometimes lost and mothers may feel alone, unsupported and even marginalised. The physical environment adds to stress with houses that are hot, cold, mouldy and isolated from amenities.

119


A second arena is the local neighbourhood. In this case it is depicted as the immediate close environments. This includes neighbours, mothers groups, buses, GPs, and nurses. Other aspects of the neighbourhood arena are explored in more depth in Chapter 6. Important also are the attitudes of the local neighbourhood to sole parents and teenage mothers. These attitudes may be influences by the cultural and religious composition of the local neighbourhood or wide community. Family is very important to most mothers. Here mother is close even when she is physically distant. The family arena can span generations and continents. With it comes culture, religion, emotional support, nurturing and sometimes financial help. But family can also bring expectations and attitudes to sole or teenage parenthood leaving the mother unsupported and alone. Support networks are a well recognised determinant of maternal depression. This was supported through the focus groups and interviews with experts. The support network arena contains: school friends, sisters and brothers, neighbours, new friends, and mothers groups. Also necessary are cars, and telephones. With the support networks come connections, nurturing of mothers, and not being alone. There may, however, also be expectations of parenting practices and pressures to conform. Aspects of support networks overlap with the neighbourhood arena, government arena, family arena and services arena. The Services Arena includes a wide range of public and private services and utilities that reach into the lives of mothers and their families. There is close connection with midwives, nurses, GPs, family support workers and mothers groups. The services support the provision of nurturing environs and connections. They also, however, influence and make expectations of mothers and families. The arena for government support and policy is only lightly touched on in the map. It is clear that this domain reaches into the close lives of the family members through provision of income support, maternity and paternity leave policies. The nature of these policies reaches into the lives of the mothers who commented on the means testing of current policies pleading that all parents with babies need support. The financial stress that impacts on families appears to have its origins in higher levels of government policy. This is explored further in Chapter 6. The government arena also places expectations on families which are manifest in policy requirements and expectations of parents.

120


4.5.3 Expectations and Dreams The discursive elements of the situational mapping highlighted the importance of expectations. Absent was discussion of the dreams that mothers might have. Yet implicit in comments by mothers were their dreams for the perfect baby, home and future. As discussed above both mothers and experts felt that society and mothers expected to (Beck 1993) take home the perfect baby and that everything was “going to be fine”. The reality is often different to that portrayed in media such as the “Huggies adverts”. The shattering of dreams and expectations was described in a range of different ways. For mothers with few resources their hopes for the future may be dashed by the impact of the new baby on both financial and emotional resources. This may be especially true for a young woman who had plans for her life. The reality is different as expressed by on expert.

“[Parenting] takes a great deal of emotional energy and to do it well, to be available for children and provide all their needs. Parents that are distracted by adverse environmental factors such as finance or domestic violence don't have that energy” For some disadvantaged parents there may, however, be no expectations or dreams. One expert put it this way

“I am not sure whether it is hope or whether there is expectation…. Expectation is the better way to put it. These people expect life to get better where as the other disadvantaged people who have been here, may be 2nd or 3rd generation think this is it.” Mothers “Loss of control” was discussed previously. Mothers implied that they had been “in control” of the lives (and also their husbands) and “expected” to be in control after the baby was born. But this was not always the case and several mothers when speaking of the “loss of control” implied that this was not their expectation of how motherhood was going to be. With the birth of the baby, planning ahead was no longer possible. Society has expectations of what a “good mother” will be like as discussed in relation to the “Huggies Advert”. One expert felt that some mothers “don’t think they are doing a good job”. Mothers agreed with this expectation. Some spoke of expectations raised by helping services such as midwives. One mother talked of the stress associated with the nurse visit.

121


“Now I have to get myself ready, make sure I don't look like I have …., so they don't think I am a bad mother” Discussion about the home visit also focused on expectations related to “you are not bonding with your child and you are expected to”, and “I am going to be graded, … please don't put me in the high risk”. One expert spoke of “career women losing it”. Several experts agreed with the mothers who had spoken of the shattering of expectations of what motherhood would be like. One expert put it like this

“Career women that have been working, who have had high expectations when they have been working, high achievers in their career, suddenly with a baby, things are crashing in on top of them and it is not going the way they want”.

4.5.4 Marginalisation and “being alone” The situational analysis and social worlds/arenas maps confirmed and strengthened the importance that marginalisation and isolation may play. The earlier analysis had identified the category of marginalisation and isolation and linked it to “not belonging”, poverty, access, language, culture, single parenthood, community safety, and living in new neighbourhoods. Examination of the social worlds map drew attention to the situation that may be experienced by most mothers. Buried in the voices of the mothers were stories of “being alone” with the crying infant, with an absent partner, mother or other support friend. Sometimes mothers spoke of not being able to get out seeing their friends. They spoke of how important the phone call from the nurse was. The descriptions were powerful and suggested that this situation is experienced by many mothers. The link between “being alone”, “depression” and partner support is clearly apparent in one mother’s story

“I go then, thank God Matt's with me. So he can give me the hug, during he night when I am so depressed. That mental help as well. Just Matt going. ‘It is ok. I am home. I can relieve you. Go have a shower’ “ Her related comment was “I think a single parent, … I don't know how they do it?”

122


4.5.5 Loss of Power and Control The initial analysis identified mothers “loss of control” as contributing to the development of depressive symptoms. That analysis identified three sub-categories related to personal relationships, birth of a new baby, and conflicting advice. Also evident in the discursive elements of the situational analysis was also the lack, or absence, of power that mothers have. This lack of power may extend to families with children and those who work to support them. One expert commented that:

“there are some women who have social expectations which are well above what the society can let them have.” When discussing mothers in ethnic communities one expert spoke of the lack of power that mothers may have in relation to themselves and their baby.

“and lots of them they don't have the individual strengths to stand for what they want or they don't know how to form their own judgement - what is best for their child or they even have the power. But lots of them … they don't have the power to say that is what I want for my child”. The “loss of power” is also implicit in comments about career women who were high achievers with high expectations when they were working

“suddenly with a baby and things are crashing in on top of them and it is not going the way they want. They don't have the social support [and] they don't know the people in the neighbourhood, … I guess they lock themselves in to a mental state where they need some help” The loss of control may therefore be also associated with a loss, or absence of power. I will explore the possible role of power in Chapter 6.

123


4.5.6 Support and Nurturing Support of mothers was identified in the earlier analysis as a strong emerging protective theme. That analysis drew on the findings of not only the interviews and focus groups but also systematic reviews of perinatal depression literature and the exploratory data analysis in Chapter 5. The social world/arena maps highlight the potential breadth of this support within neighbourhood and home environments. The term nurturing was previously used to describe the care of the infant. As discussed earlier (section 4.4.2.2), the concept of nurturing can be broadened to encompass the care and protection of the mother and family unit. The situational analysis map was silent on a discursive element related to nurturing families. I will revisit the concept of nurturing in Chapters 6 and 8.

4.6

Theory Generation

The situational analysis was the third stage of data analysis in this chapter. The focus was on possible situations experienced by mothers with newborn infants. The theoretical concepts of: expectations and dreams; marginalisation and “being alone”; loss of power and control; and support and nurturing were briefly explored. The previously identified concepts of stress and infant temperament remain significant for the process of theory building. Stress in particular is a strong contender as the proximal cause of depression. The following conceptual network illustrates how the concepts identified in this Chapter might combine to create the conditions necessary for perinatal depression to occur or be maintained.

124


is cause of

Isolation

protects from is cause of

is cause of

protects from

Stress~ is cause of

is cause of

Support

protects from

Marginalisation

Loss of control of your life


is cause of Crying Baby~ is cause of

protects from

is cause of

is cause of

Expectations Lost

Infant Temperament

Figure 27: Emerging Conceptual Framework of Maternal Stress

Figure 28: Conceptual Model of Mothers Mood The above models will be taken forward to Chapter 6 where I will explore group level factors, and to Chapter 8 where I will undertake further theory development. The focus of this study was not on theory building at the foetal and infant level. I will, therefore, leave further analysis of the findings in Section 4.4.1 until Chapter 8 where I will attempt to integrate those findings into a mid-level Theory of Maternal Depression, Stress and

Context.

125


4.7

Conclusion

The purpose of this Chapter was to describe the findings of the qualitative exploration of individual level factors that might be associated with the phenomenon of perinatal depression, perinatal adversity and their possible impact on the developmental and health outcomes for the foetus and infant. The qualitative study was undertaken concurrently with the literature review and exploration of quantitative data. The key phenomenon identified in the focused coding were: foetal stress and damage, biological programming, poor attachment, nurturing, unplanned pregnancy, support for mothers (individual level), access to services and amenities (individual level), mothers loss of control, stress, social position and marginalisation of individuals. Situational analysis was conducted as a third coding stage. The theoretical concepts of: expectations and dreams; marginalisation and “being alone”; loss of power and control; and support and nurturing were briefly explored and used to develop an emerging conceptual framework of maternal stress. This model together with those developed at the foetal and infant level will be taken forward to Chapter 8 where I will undertake further theory development. The following Chapter will report on the exploration of quantitative data at the individual level. The phenomenon of postnatal depression will be described and early theory generation will be undertaken using quantitative modelling methods.

126

Chapter 5: Individual Quantitative

Chapter 5: Exploratory Data Analysis of Individual Level Factors Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







127


5.1

Introduction

The purpose of this Chapter is to describe the quantitative exploration of factors that might be associated with perinatal depression, perinatal adversity and their possible impact on developmental outcomes of the infant. As previously mentioned the study was undertaken concurrently with other analyses which both informed and were informed by the findings in this Chapter. As an emergent and critical realist study the methods of exploratory data analysis (EDA) and exploratory factor analysis (EFA) used differed from that commonly used in epidemiological deductive and confirmatory analytical studies. As previously cited, Tukey (1980, p 23) stated the maxim that “finding the question is often more important than

finding the answer”. In this same EDA tradition Behrens (1997, p 132) stated that “the role of the data analyst is to listen to the data in as many ways as possible until a plausible ‘story’ of the data is apparent, even if such a description would not be borne out in subsequent samples’ There is a substantial body of previous risk factor studies of postnatal depression. One of the reasons for undertaking the analysis reported in this Chapter was to explore the data that would later be used in spatial and multilevel exploratory analysis (Chapter 7). The other purpose was to conduct the study on a sub-set of available South West Sydney data, with a view to undertaking subsequent confirmatory analysis on the larger data-set. As stated in the introduction to Chapter 4, it is not the purpose of this Study to examine in detail, phenomenon at the Individual Level other than to identify the possible mechanisms and pathways by which the “Group Level” factors operate. In the final section of this Chapter I will summarise the Emerging Concepts at the maternal and family level.

128


5.2

Aims

The aims of this chapter are to: 1. Undertake a quantitative study of the individual and family level factors associated with postnatal depression 2. Use the findings to inform emerging theory. The specific objectives are to: 1. Explore the South West Sydney Ingleburn Baby Information System dataset, using basic and advanced exploratory data analysis methods 2. Identify individual-level factors that are associated with increased postnatal depressive symptoms as measured using the Edinburgh Postnatal Depression Scale 3. Identify groupings of variables that may be associated with latent vectors 4. Use causal directed acrylic graphs (DAGs) to clarify assumptions about the causal structure of models explored 5. Explore a number of nested models to determine the most parsimonious set of individual predictors. 6. Describe conceptual models arising from the data for use as part of the theory construction. The specific analysis questions are: 1. Are there associations in this South West Sydney data that have not been found previously? 2. What factors might be linked to “group level” factors and what parsimonious set of factors can be identified for analysis in a multi-level regression? 3. How are the factors related to each other and, therefore, what latent, unmeasured factors might be important?

129


5.3

Methods

5.3.1 Theoretical Approaches In a mixed method study it is expected that the analyst is explicit regarding his/her theoretical approach. This aspect of rigour should also be applied to quantitative studies. In Chapter Two I discussed the philosophical and methodological approaches used in this study. As noted critical realist methodology supports the use of quantitative data to detect phenomena and generate theory (Danermark, Ekstrom et al. 2002; Haig 2005; Haig 2005). Similarly quantitative EDA is a valid post-positivist, and critical realist, methodology (Tukey 1980; Behrens 1997; Haig 2005). As characterised by Behrens (1997, p 132) EDA will be used here to “focus on tentative model building and hypothesis generation” using both basic and sophisticated EDA methods including: univariate data display, correlation matrices, bivariate and multiple regression categorical principal component analysis, exploratory factor analysis, and exploratory CFA specification search.

5.3.2 Sample and Design 5.3.2.1 Study Design The study is a quantitative population-based cross-sectional study of mothers of infants born in South Western Sydney Area Health Service (SWSAHS) from 2002-2003. The study is a sub-sample from a larger dataset collected from 1998 to 2006. Data from 2004 to 2006 is suitable for a subsequent confirmatory study. The exploratory data analysis will include descriptive analysis of data, factor analysis and principal component analysis, and logistic regression. Prior knowledge of causal inference will be utilised during the analysis to ensure that the variables “conditioned” in the regression analysis are appropriate. 5.3.2.2 Study Setting The setting is in the New South Wales Local Government Areas (LGAs) of Bankstown, Fairfield, Liverpool, Campbelltown, Camden, Wollondilly and Wingecarribee. Community nursing services are provided in the five sectors of Bankstown, Fairfield, Liverpool, Macarthur and Wingecarribee. The Macarthur sector includes the LGAs of Campbelltown, Camden, and Wollondilly. The population in SWSAHS is large and continues to rapidly increase. From the 2001 Census of Population and Housing, (ABS 2002; DOH 2003; SWSAHS 2004) SWSAHS was reported to have a population of 769,595 and this is

130


projected to grow to nearly 891,000 by 2006. Fourteen percent of the State’s births are in this area with 10,011 babies born in 2002-2003. The area has a diverse multicultural population with 28.4 percent of the population being born overseas compared with 17.8 percent for the rest of NSW. Twenty percent of babies in SWSAHS are born to women from South East, North East or Southern Asia. Also, one quarter of Sydney’s indigenous population live in South Western Sydney. South Western Sydney is also an area of substantial social disadvantage, and has lower education attainment and lower income levels than other parts of NSW. Based on composite socio-economic indices, about two-thirds of the area is substantially disadvantaged, which is associated with a range of poor health indicators. There is a high rate of public housing, at 10.1 percent, which is nearly twice the NSW rate (SWSAHS 2004). These socio-economic factors further complicate the health status patterns within the SWSAHS Area. There are independent associations between socio-economic disadvantage and poor health status, somewhat counter balanced by better health status for some conditions amongst people from non-English speaking backgrounds. Issues of health service access may be issues for people from non-English speaking backgrounds. 5.3.2.3 Individual-level study population, sampling frame and sampling method The individual-level study utilised the Ingleburn Baby Information System (IBIS) database. This database was initiated in 1995 and is based on data collected by Child and Family Health Nurses on all mothers who attend the first well baby clinic (home visit or clinic based) after discharge from the post-natal ward. This information is completed on a clinical record form which is scanned into a Windows© based PC program (Appendix D). IBIS does not collect information on a number of relevant variables, including maternal age and parity, which are routinely collected during pregnancy and childbirth and entered into the OBSTET database. Data linkage to obstetric data was only possible for the Local Government Areas of Fairfield, Camden, Campbelltown and Wollondilly. This linkage enabled analysis of the relationship of EDS to maternal age. The IBIS dataset contained 234,034 cases as at October 2006. Population-based collection started in Campbelltown in 1998 followed by Bankstown in 2000, Fairfield and Wingecarribee in 2001 and Liverpool in 2002. I have elected to undertake an EDA for the calendar years of 2002 and 2003 as all geographical areas are covered and this will allow subsequent individual level, group level and multi-level analysis using the same data set. The frequency of cases, 2002-2003, available for analysis is shown below.

131


Year by SWSAHS Sector Count Sector BNK

FAIR

LIV

MAC

WIN

Total

2002.00

2300

2319

1954

3381

469

10423

2003.00

2319

2818

2634

3391

406

11568

4619

5137

4588

6772

875

21991

Total

BNK (Bankstown), FAIR (Fairfield), LIV (Liverpool), MAC (Macarthur including Campbelltown, Camden and Wollondilly), WIN (Wingecarribee).

Table 3: Frequency of IBIS cases 2002 to 2003 by Health Service Sector 5.3.2.3 Missing Data 5.3.2.4 EDS missing data In the first two years of the collection of data the EDS was not completed consistently. Interviews with nurses revealed that they often did not complete the EDS if the women were obviously not depressed. Consequently in 2000 and 2001 there were significant numbers of missing EDS data. In the years 2002 and 2003 there remained 30 percent missing data. This level of missing EDS data is problematic and would be of significant concern if this was the final confirmatory data analysis. As this analysis is exploratory, I have elected to proceed with using this earlier data. Comparison of those who completed the EDS and those who did not indicates that those with factors likely to be associated with depressive symptoms were less likely to have an EDS completed than those who were not. For each of the variables in Table 4 a cross tabulation was undertaken of those who completed the EDS with those who did not. Statistically significant differences existed for financial situation, accommodation, maternal education, breast-feeding and suburb duration. Financial difficulties were more likely, for example, to be reported by mothers who did not have an EDS. If depressive symptoms are associated with financial difficulty, then the observed association will be less than expected. There was a systematic nature to the missing data. A plot of missing data by sector also indicated that missing EDS data was more likely in the Macarthur Sector.

132


Pearson ChiSquare

df

Sig

Country of Birth(D)

0.671

1

0.413

Aboriginal(D)

2.575

1

0.109

Variable

Phone Access(D)

2.082

1

0.149

Car Access(N)

5.514

2

0.063

Unemployed Father(D)

3.421

1

0.064

Financial Situation(D)

34.825

2

0.000

Marital Status(N)

0.752

2

0.687

Accommodation(D)

16.393

1

0.000

Smoking

0.434

1

0.510

Maternal Education

59.860

4

0.000

Breast Feeding

23.782

1

0.000

Suburb Duration

13.444

2

0.001

D (Dichotomous), N (Nominal)

Table 4: Comparison of EDS Missing Data Cases Hair and colleagues (1998, p 51) state that the only approach to non-random or “missing at random (MAR)” is the specifically designed modelling approach. All other approaches will introduce bias. I have assessed that deleting cases will bias the results by deleting more cases from Macarthur and more cases with difficult financial circumstances. By contrast the imputation method that I have used below will artificially increase the explanatory power of the analysis. I have elected not to use the modelling approach but rather have used two methods, one which may underestimate the associations and the other which may artificially overestimate. I undertook the analysis using: 1. Selection of only those cases where the EDS was completed 2. Imputation of missing EDS cases using median scores. Mean (or median in the case of skewed data) is one of the more widely used methods. The general approach replaces the missing variables with a mean or median value of that variable based on all valid responses. A modification, which I have used, uses a median value based on the two closest responses. All exploratory analyses were done on both datasets. The findings were consistent for both datasets. I have reported in the Results Section on the analysis of the deleted case dataset. Analysis of the imputed dataset is at Appendix E.

133


5.3.2.5 Other Missing Data Data for self reported financial situation was missing for 15 percent of cases. This level of missing data is acceptable (Hair et al., 1998) and consequently I did not impute. The missing cases will, however, be deleted during the statistical analysis. There were missing data from other variables. In all cases the percentage was less than for financial situation.

5.3.3 The Study Variables and Univariate Analysis 5.3.3.1 Outcome (Dependent) Variable The EDS was administered at the time of first visit usually in the home (Appendix C). Validation studies of the EDS have demonstrated 68 percent - 86 percent sensitivity and 78 percent - 96 percent specificity and in an Australian sample, 100 percent sensitivity and 89 percent specificity. Positive predictive value has been reported between 70 percent-90 percent. The EDS has been validated for a number of other languages and ethnic groups, including: Spain (Garcia-Esteve, Ascaso et al. 2003), Turkey (Inandi, Elci et al. 2002), Nepal (Regmi, Sligl et al. 2002), Norway (Eberhard-Gran, Eskild et al. 2001), Japan (Yoshida, Yamashita et al. 2001), India (Banerjee, Banerjee et al. 2000), French speaking Quebec (Des Rivieres-Pigeon, Seguin et al. 2000), China (Zhang, Chen et al. 1999), Germany (Bergant, Nguyen et al. 1998), Sweden (Bagedahl-Strindlund and Monsen Borjesson 1998), Brazil (Da-Silva, Moraes-Santos et al. 1998), Arabic (Ghubash, AbouSaleh et al. 1997), Malaysia (Kit, Janet et al. 1997), and Portugese (Areias, Kumar et al. 1996). South Western Sydney studies have found that Vietnamese and Arabic translations of EDS were acceptable to the women and appear to be suitable screening instruments for postnatal distress and depression in these populations (Matthey, Barnett et al. 1997; Barnett, Matthey et al. 1999). The EDS has not been validated for indigenous Australians. The distribution of the EDS Scale variable was expected to be skewed. The distribution and descriptive statistics of the EDS Scale are shown at Figure 29 and Table 5 below.

134


2,000

Frequency

1,500

Mean =5.48 Std. Dev. =4.4 N =15,389 1,000

500

0 -10

0

10

20

30

EDscore

Figure 29: Distribution of the EDS Variable

Descriptive Statistics

EDscore N

Statistic

15389

Range

Statistic

29

Minimum

Statistic

0

Maximum

Statistic

29

Mean

Statistic

5.48

Std. Error

15389

.035

Std. Deviation

Statistic

4.400

Variance

Statistic

19.361

Skewness

Statistic

1.054

Std. Error Kurtosis

Valid N (listwise)

Statistic Std. Error

.020 1.309 .039

Table 5: Descriptive Statistics for the EDS Variable

135


Dichotomous EDS Variable The study in this Chapter reports on both EDS > 9 and EDS > 12. Buist and colleagues (2002) note that using EDS > 12 is a sound option for reducing false positives. Using EDS > 9 is also useful as many women experience considerable dysfunction and want assistance (Brown, Davey et al. 2001; Buist, Barnett et al. 2002). Consequently two dichotomous variables (EDS9 and EDS12) were created from each of the two datasets. Each of the resulting variables has a value of 0= EDS 0-9 or 1=EDS>9 and 0= EDS 0-12 or 1=EDS>12 respectively. 5.3.3.2 Comparison (Independent) Variables As discussed earlier, EDA is in the post positivist tradition and shares the interaction of prior knowledge and data analysis. Thus I have allowed my prior knowledge and the findings of the concurrent literature reviews and qualitative studies to influence the selection of variables for study. At the individual level I allowed any variable that was numerically robust and plausible to enter the initial EDA. My prior knowledge assisted this process as did the findings of the literature review, interviews and focus groups. One omission from this process was the failure to include variables related to maternal expectation. Maternal expectation was found to be important from the qualitative studies. Consequently all analyses in this Chapter were repeated with “Maternal Expectation” included.

136


Variable

Type of Data

Transformations Used

Demographic Country of Birth (D)

Dichotomous

Aboriginal or Torres Strait Islander (D)

Dichotomous

Age of Mother

Continuous

Sex of Baby (D)

Dichotomous

Marital Status (C)

Nominal

Household Size (O)

Ordinal

Blended Family (D)

Dichotomous

Children Under Five (O)

Ordinal

Ordinal

Dichotomous

Socioeconomic Accommodation (C)

Nominal

Employment of Mother

Nominal

Employment of Father

Nominal

Financial Situation

Ordinal

Car Access

Nominal

Phone Access

Dichotomous

Education of Mother

Nominal

Dichotomous

Dichotomous

Dichotomous

Health Mother Rating of Health

Ordinal

Rating of Child's Health

Ordinal

Breast Feeding

Dichotomous

Smoking

Dichotomous

Mothers Expectation

Ordinal

Dichotomous

Dichotomous

Pregnancies Planned Pregnancy

Dichotomous

Previous Miscarriage

Dichotomous

Previous Child Death

Dichotomous

Previous Stillbirth

Dichotomous

Previous Child Disability

Dichotomous

Previous Termination

Dichotomous

Previous SIDS

Dichotomous

Table 6: Comparison (Independent) variables studied

137


Variable

Type of Data

Transformations Used

Social Network Suburb Duration

Ordinal

Regret Leaving Suburb

Ordinal

Dichotomous

Support Network

Ordinal

Dichotomous

Practical Support

Nominal

Dichotomous

Emotional Support

Nominal

Dichotomous

Attachment Mother Responds to Child

Dichotomous

Comforts Baby

Dichotomous

Enjoys Contact with Baby

Dichotomous

Baby Characteristics Baby Trouble Sleeping

Ordinal

Dichotomous

Baby Demanding

Ordinal

Dichotomous

Baby Content

Ordinal

Dichotomous

Baby Difficult Feeder

Ordinal

Dichotomous

Baby Difficult to Comfort

Ordinal

Dichotomous

Health of Child

Ordinal

Dichotomous

Table 6: Comparison (Independent) variables studied (Continued)

138


Descriptive statistics were examined on each variable. Attention was given to missing variables, outliers, and distribution statistics. Nominal variables were examined to inform later development of dummy variables. Ordinal variables were examined to assess their fit to a normal distribution. Almost all individual level variables required some form of recoding, transformation or assigning of dummy variables. For some analysis, such as the later Bayesian studies, the variables were also centred about their mean. The data transformations were both theoretically and data driven. The process was iterative with constant assessment of improvement versus the need for additional transformation. Dichotomous variables were often coded as 1 or 2 in the dataset thus requiring a recoding to 0 or 1. The order of the coding was designed to ensure that likely “risks” of depressive symptoms were assigned “1”. Dummy variables were created. Indicator and Orthogonal (Polytomous) coding was used. Nominal variables were recoded to 0 and 1 using the Indicator Dummy Code method of n1 vectors. Ordinal variables were treated in three ways in the initial exploratory analysis. 1

As continuous variables (if normally distributed and 5 or greater vectors).

2

As Orthogonal (Polytomous) variables

3

As Indicator variables.

The large number of categorical variables and need for dummy variables, was problematic creating a conflict with the aim of parsimony and planned multi-level analysis. The exploratory analysis sought to identify dichotomous or scale variables that were meaningful and yet did not lose statistical significance.

139


5.3.4 Bivariate Exploration Bivariate analysis of the independent variables against the outcome variables was undertaken. The emphasis in this chapter is on presenting detail of the individual level findings that will be used in later multi-level analysis.

Dichotomous Individual-Level Outcome Variables For the dichotomous outcomes the relationship with nominal, and ordinal variables was assessed using the likelihood and Pearson chi-square tests and Univariate logistic regression. Estimates of odds ratios and confidence intervals were made. For the ordinal variables scatter plots were also undertaken. The significant results are presented in the Results Section.

Continuous Outcome Variable For the continuous outcome of EDS scale a limited number of univariate linear regressions were undertaken. The results of the univariate linear regressions are not presented in this Study.

Multinominal Outcome Variable Analysis of a possible multinominal outcome variable (EDS 0-9, 10-12, >12) was not undertaken for this study.

5.3.5 Exploratory Factor Analysis Factor analysis is a statistical tool for analysing scores on large numbers of variables to determine whether there are any identifiable dimensions that can be used to describe many of the variables under study (Munro 2001). It is often assumed that the co-variation between variables is due to some underlying common factors. Factor analysis examines the inter-correlations mathematically to enable the underlying traits to be identified. Factor analysis is a dimensional clustering technique that allows for the possibility that a variable is only partially in a cluster. The “factor loading” of a variable indicates the extent to which it is part of the factor cluster (Asher, Weisberg et al. 1984, p 342) Factor analysis may be exploratory or confirmatory. As most of the relevant variables are categorical, I will use a variant of exploratory factor analysis, Categorical Principal Components Analysis (CatPCA), and an exploratory approach to confirmatory factor analysis and structural equation modelling (i.e. specification search (AMOS 18.0)).

140


Exploratory factor analysis can be used for instrument development, data reduction and theory development. In theory development factor analysis is used to discover a structure that can be meaningfully interpreted. The researcher begins without preconceived expectations about the nature of the structure that will emerge and the structure is allowed to unfold from the data (Munro 2001). Factor analysis is also used for abductive reasoning as in this critical realist theory building study (Haig 2005). 5.3.5.1 Categorical Principal Component Analysis (CatPCA) Factor analysis, and the related principal component analysis (PCA) approach, is based on a matrix of correlations between variables and hence data assumptions for correlations and linear regression apply including the requirement for interval data that is normally distributed. The data in this study is mainly categorical and does not meet those assumptions. Hence the commonly used statistical approaches to factor analysis and PCA could not be used. One of the new algorithmic models used for measuring latent variables is Principal Component Analysis (PCA) with Optimal Scaling (Gifi 1990; Meulman, Van der Kooij et al. 2004). CatPCA is the nonlinear equivalent of PCA but unlike PCA, CatPCA can manage categorical variables and does not require classical statistical assumptions, like multivariate normality. CatPCA simultaneously reduces the dimensionality of the data and turns categorical variables into quantitative variables using optimal scaling. The quantitative measure obtained by CatPCA (object scores) takes into account the possible multidimensionality, the nature of variables, and their importance in determining the measure. The quantitative measures have coordinates that allow the categories or dimensions to be represented in a geometric display thus making data interpretation easier. I used CatPCA to identify dimensions in the data that may represent underlying latent (unmeasured) variables. The information thus gained was used to inform the logistic regression modelling, development of measurement models and specification search, and theory generation. The number of dimensions was based on ensuring that the eigenvalues were greater than 1. The Shepard (scree) plot for the first seven dimensions is at Figure 30.

141


5.3.5.2 Specification Search (AMOS 18.0) When a hypothesised model is rejected by confirmatory factor analysis or structural equation modelling, the researcher can decide whether or not to respecify and re-estimate the model. If the model is respecified and re-estimated it is considered “post hoc analysis” and must now be framed within an exploratory rather than confirmatory mode. Although CFA procedures and software continue to be used these analyses are exploratory in the sense that they focus on the detection of mis-fitting parameters in the original hypothesised model. Such post hoc analyses are conventionally termed “specification searches” and are available in AMOS software (Schumacker 2006; Byrne 2010, p 89). Measurement models are that part of a Structural Equation Model (SEM) which deals with the latent variables and their indicators. The measurement model is conventionally drawn with unmeasured covariance between each pair of latent variables, straight arrows from the latent variables to their respective indicators and straight arrows from the error and disturbance terms to their respective variables. The unmeasured covariance is drawn as a two-headed covariance arrow unless there are strong theoretical reasons not to do so. The measurement model is evaluated using SEM goodness of fit measures. It is recommended that the chi-square be reported. The findings from the CATPCA and Logistic Regression were, used together with prior knowledge from the literature review and qualitative research findings, to prepare measurement models with three and four latent factors for specification search in AMOS 18.0.

5.3.6 Use of Causal models Prior to undertaking regression analyses it is important to propose the hypothetical causal networks being considered. Hernandez-Diaz and colleagues (2006) use the example of birth weight to examine how paradoxical conclusions may be drawn if preliminary causal inference is not assessed. They argue that “paradoxes” can appear in any research field when statistical adjustments ignore causal relations among variables. In their study they show how standard adjustment (stratification or regression) for variable affected by exposure may create bias by introducing spurious (non causal) association between the exposure and the outcome. Care must be taken because adjustment for factors that might be on the causal pathway between the exposure and the outcome is often unwarranted (Hernandez-Diaz, Schisterman et al. 2006). Care should also be taken to not control for a variable that is a descent (effect) of the outcome understudy (Greenland, Pearl et al. 1999).

142


Pearl (1993) proposed the following criterion, named “back-door” which provides a graphical method of selecting a set of factors for adjustment. It states that a set S is appropriate for adjustment if two conditions hold: 1.

No element of S is a descent of X

2.

The elements of S “block” all “back-door” paths from X to Y, namely all paths that

end with an arrow pointing to X. In this criterion, a set S of nodes is said to block a path p if either 1.

p contains at least one arrow-emitting node that is in S, or

2.

p contains at least one collision node that is outside S and has no descent in S.

Investigators can use their prior expert knowledge to propose hypothetical causal networks. Diagrams known as directed acyclic graphs (DAGs) can be used to represent those networks (Pearl 1995). Similar to the causal networks used in Chapter 4, the diagrams link variables (nodes) by arrows (directed edges) that represent direct causal effects (protective or causative) of one variable on another. DAGs are acyclic because the arrows never point from a given variable to any other variable in its past (ie. Causes precede their effects). The absence of an arrow between two variables indicates that the investigator believes there is no effect. In this emergent and exploratory phase of theory building I will use DAGs to limit the possibility of spurious findings from the multivariate methods used in the exploratory data analysis. DAGs will be used to make explicit why variables are entered, or not entered, into the logistic regression. The assumptions made are tentative and may not hold true through the remainder of the study.

143


5.3.7 Logistic Regression 5.3.7.1 Analysis Technique Candidate variables were selected for the multivariate model if the univariate test had a pvalue 40

123

1.86

11,728

54.7

9,708

45.3

2,768

14.8

Average

14,329

76.7

Difficult

1,582

8.5

652

3.1

School cert Year 10

5,529

26.1

HSC Year 12

5,098

24.1

TAFE

5,036

23.8

University

4,197

19.8

662

3.1

1 to 5

17,956

83.0

6 to 10

3,563

16.5

109

0.5

10,376

48.3

Own house

1,575

7.3

Rent privately

5,596

26.1

Rent public housing

1,360

6.3

Living with parents

2,483

11.6

Live in Caravan

36

0.2

Live in refuge

39

0.2

No Maternal Age (years)* 10 Accommodation Mortgage

* Maternal age available only from Fairfield, Campbelltown, Camden and Wollondilly LGA only

** A = Private - own house, mortgage, rent privately or living with parents/B = Public - public housing, caravans, refuge *** Subjective scale from 1-10: 1-3 difficult / 4-7 average / 8-10 good

Table 7: Characteristics of Study Sample, 2002 to 2003

148


Characteristic

Number

Percent

Employment of Father Full Time

14,588

71.6

Casual FT

392

1.9

Part Time

489

2.4

Casual PT

288

1.4

Self Employed

2,131

10.5

Unemployed

1,355

6.6

Student

233

1.1

Pension

315

1.5

Maternity Leave

49

0.2

Home Duties

83

0.4

454

2.2

15,578

72.2

Occasional

3,416

15.8

Never

2,589

12.0

21,014

98.0

424

2.0

Yes

14,329

66.6

No

7,202

33.4

Yes

2,965

14.2

No

17,963

85.8

1,955

9.0

Married

16,626

76.4

Partner

3,195

14.7

Other Car Access Regular

Phone Access Yes No Planned Pregnancy

Blended Family

Marital Status Single

* Maternal age available only from Fairfield, Campbelltown, Camden and Wollondilly LGA only

** A = Private - own house, mortgage, rent privately or living with parents/B = Public - public housing, caravans, refuge *** Subjective scale from 1-10: 1-3 difficult / 4-7 average / 8-10 good

Table 7 (Cont.): Characteristics of Study Sample, 2002 to 2003

149


5.4.2 Prevalence of Postpartum Depression The distribution of EDS was previously described in Section 4.4.2.1. The mean EDS Score was 5.48 (95% CI: 5.41 - 5.55). The distribution was skewed (Skewness Statistic 1.054, SE 0.02). At a mean postnatal age of 3.77 weeks (95% CI: 3.62-3.92) the prevalence of EDS >9 was 16.93 (95% CI: 16.34 – 17.52) and the prevalence of EDS > 12 was 7.37 (95% CI: 6.96 – 7.78). 95 % CI Mean

Lower

Upper

EDS Score

5.48

5.41

5.55

Postnatal age at First Visit

3.77 wk

3.62 wk

3.92 wk

EDS >9

16.93%

16.34%

17.52%

EDS >12

7.37%

6.96%

7.78%

Table 8: Prevalence of EDS at First Visit

150


5.4.3 Bivariate Analysis EDS > 9 Variable

Chi Square

Sig

EDS > 12

Chi Square

Sig

Type

df

D

1

145.02

0.00

0.49

0.48

1

0.66

0.42

0.49

0.48

Demographic Country of Birth (D) Aboriginal or Torres Strait Islander (D)

D

Sex of Baby (D)

D

1

0.38

0.54

0.05

0.83

Marital Status (C)

C

2

38.25

0.00

78.87

0.00

Household Size (O)

O

2

6.49

0.04

7.68

0.02

Blended Family (D)

D

1

8.38

0.00

16.05

0.00

Children Under Five (O)

O

3

21.72

0.00

5.94

0.11

Accommodation (C)

N

6

38.07

0.00

77.44

0.00

Accommodation (D)

D

1

13.05

0.00

24.07

0.00

Employment of Mother

N

10

44.61

0.00

65.44

0.00

Employment of Father

N

10

49.44

0.00

60.15

0.00

Employed Yes/No

D

1

7.53

0.01

10.17

0.00

Financial Situation

O

9

275.56

0.00

285.22

0.00

Financial Situation G/A/D

C

2

214.89

0.00

225.21

0.00

Car Access

N

2

85.84

0.00

94.25

0.00

Phone Access

D

1

2.80

0.09

5.11

0.02

Education of Mother

N

5

9.28

0.10

10.43

0.06

Planned Pregnancy

D

1

36.56

0.00

52.56

0.00

Previous Miscarriage

D

1

0.09

0.77

0.17

0.68

Previous Child Death

D

1

0.08

0.78

1.42

0.23

Previous Stillbirth

D

1

3.40

0.07

2.38

0.12

Previous Child Disability

D

1

2.50

0.11

0.49

0.48

Previous Termination

D

1

0.34

0.56

0.10

0.75

Previous SIDS

D

1

2.30

0.13

1.52

0.22

Socioeconomic

Pregnancies

O - Ordinal, D - Dichotomous, N - Nominal

Table 9: Chi Square exploration of EDS possible associations

151


EDS > 9 Variable

Type

EDS > 12 Chi Square

df

Chi Square

1

340.66

0.00

479.44

0.00

4

543.91

0.00

608.78

0.00

4

177.05

0.00

126.14

0.00

Sig

Sig

Health Mother Rating of Health Mother Rating of Health Rating of Child's Health

D O O

Breast Feeding

D

1

4.87

0.03

20.22

0.00

Smoking

D

1

0.76

0.38

1.59

0.21

O

2

31.68

0.00

40.86

0.00

3

53.33

0.00

76.18

0.00

1

30.23

0.00

43.57

0.00

Social Network Suburb Duration Regret Leaving Suburb Regret Leaving Suburb

D

Support Network

O

4

459.09

0.00

403.40

0.00

Support Network

D

1

304.68

0.00

281.11

0.00

Practical Support

N

2

229.52

0.00

192.90

0.00

Practical Support

D

1

213.27

0.00

160.95

0.00

Emotional Support

N

2

367.95

0.00

345.86

0.00

Emotional Support

D

1

340.63

0.00

312.51

0.00

1

1.88

0.17

2.67

0.10

1

14.10

0.00

14.04

0.00

1

21.96

0.00

39.66

0.00

4

318.90

0.00

221.46

0.00

O

Attachment Mother Responds to Child Comforts Baby Enjoys Contact with Baby

D D D

Baby Characteristics Baby Trouble Sleeping

O

Baby Demanding

O

4

325.14

0.00

243.52

0.00

Baby Content

O

4

277.18

0.00

207.52

0.00

Baby Difficult Feeder Baby Difficult to Comfort

O

4

130.49

0.00

70.71

0.00

O

4

239.11

0.00

175.33

0.00

Health of Child

O

4

177.05

0.00

126.14

0.00

O - Ordinal, D - Dichotomous, N - Nominal

Table 9: Chi Square exploration of EDS possible associations (Continued)

152


Variable

Categor ies

B

SE

0.504 0.399

OR Exp( B)

95% CI Lowe Upper r

Wald

df

Sig

0.047

116.925

1

0.000

1.656

1.511

1.814

0.072

30.371

1

0.000

1.490

1.293

1.717

7.586

2

0.023

Demographic Country of Birth (D) Sole Parent (D) Household Size (C)

1–5 6 -10

-0.055

0.063

0.761

1

0.383

0.947

0.837

1.071

> 10

0.671

0.260

6.641

1

0.010

1.957

1.174

3.260

0.199

0.065

9.329

1

0.002

1.221

1.074

1.387

0.266

0.089

8.934

1

0.003

1.304

1.096

1.552

0.200

0.089

5.039

1

0.025

1.221

1.026

1.454

0.215

0.015

211.533

1

0.000

1.239

1.204

1.276

198.811

2

0.000

Average

0.545

0.080

46.847

1

0.000

1.725

1.476

2.017

Poor

1.372

0.101

185.560

1

0.000

3.944

3.237

4.805

68.014

2

0.000

Blended Family (D)

Socioeconomic Accommodation (D) Unemployed Father (D) Financial Situation (S) Financial Situation (C)

Car Access (C)

Good

Regular Occasio nal No

0.242

0.061

15.747

1

0.000

1.273

1.130

1.434

0.506

0.064

62.224

1

0.000

1.659

1.463

1.882

Mothers Health (D)

1.331

0.081

268.703

1

0.000

3.785

3.228

4.438

Mothers Health (S)

0.594

0.031

377.541

1

0.000

1.812

1.706

1.923

Child's Health (S)

0.383

0.031

156.867

1

0.000

1.467

1.382

1.558

Mothers Expectations

0.851

0.030

794.724

1

0.000

2.342

2.207

2.484

0.292

0.047

38.151

1

0.000

1.339

1.220

1.469

0.144

0.026

31.293

1

0.000

1.154

1.098

1.214

0.224

0.048

22.027

1

0.000

1.251

1.139

1.374

32.212

2

0.000

Health

Pregnancies Planned Pregnancy (D)

Social Network Suburb Duration (S) Suburb Duration (D) Suburb Duration (C)

Regret Leaving Suburb (C)

3+ 2 yrs

0.205

0.062

10.766

1

0.001

1.227

1.086

1.387

1 or
9, Significant Associations

154


OR Variables

Categories

95% CI

B

SE

Wald

Df

Sig

Exp(B)

Lower

Upper

Country of Birth (D)

0.544

0.064

71.942

1

0.000

1.722

1.519

1.953

Sole Parent (D)

0.711

0.086

69.091

1

0.000

2.036

1.722

2.408

11.205

2

0.004

Demographics

Household Size (C)

1–5 6 – 10

0.040

0.083

0.234

1

0.629

1.041

0.885

1.225

> 10

0.968

0.290

11.102

1

0.001

2.632

1.490

4.651

0.307

0.084

13.477

1

0.000

1.359

1.154

1.600

46.022

6

0.000

Blended Family (D) Socioeconomic Accommodation (C)

Mortgage Owned

0.027

0.130

0.044

1

0.834

1.028

0.796

1.327

Private rent

0.198

0.076

6.751

1

0.009

1.219

1.050

1.416

Public Rent

0.401

0.120

11.093

1

0.001

1.493

1.179

1.890

Parents

0.385

0.095

16.222

1

0.000

1.469

1.218

1.771

Caravan

0.464

0.739

0.394

1

0.530

1.590

0.374

6.764

Refuge

1.841

0.370

24.783

1

0.000

6.305

3.054

13.017

Accommodation (D)

0.366

0.109

11.282

1

0.001

1.441

1.164

1.784

Unemployed Father (D)

0.271

0.116

5.414

1

0.020

1.311

1.044

1.647

Financial Situation (S)

0.245

0.020

151.448

1

0.000

1.277

1.229

1.328

163.712

2

0.000

Financial Situation (C)

Car Access (C)

Good Average

0.513

0.118

19.063

1

0.000

1.671

1.327

2.104

Poor

1.487

0.135

120.573

1

0.000

4.425

3.394

5.771

67.792

2

0.000

Regular Occasional

0.357

0.081

19.600

1

0.000

1.428

1.220

1.673

Poor

0.636

0.082

60.970

1

0.000

1.890

1.611

2.217

Mother’s Health (D)

1.510

0.088

291.633

1

0.000

4.527

3.807

5.384

Mother’s Health (S)

0.704

0.042

286.121

1

0.000

2.021

1.863

2.193

Child’s Health (S)

0.385

0.042

85.619

1

0.000

1.470

1.355

1.595

Breast Feeding (D)

0.209

0.063

10.893

1

0.001

1.233

1.089

1.396

Mothers Expectation (S)s

1.019

0.042

576.893

1

0.000

2.771

2.550

3.012

Planned Pregnancy

0.469

0.063

55.256

1

0.000

1.599

1.413

1.809

Previous Stillbirth

0.228

0.109

4.368

1

0.037

1.256

1.014

1.556

Suburb Duration (S)

0.197

0.035

32.104

1

0.000

1.218

1.138

1.304

Suburb Duration (D)

0.362

0.067

29.141

1

0.000

1.436

1.259

1.638

32.779

2

0.000

Health

Pregnancies

Social Network

Suburb Duration (C)

3+

155


OR Variables


95% CI

Categories

B

SE

Wald

Df

Sig

Exp(B)

Lower

Upper

2 -3 yrs

0.280

0.086

10.710

1

0.001

1.324

1.119

1.565

1 yr or less

0.395

0.070

31.699

1

0.000

1.484

1.293

1.702

70.442

3

0.000

Yes a lot Yes a little

0.161

0.077

4.394

1

0.036

1.175

1.011

1.365

No not much

0.419

0.096

19.140

1

0.000

1.521

1.260

1.835

No not at all

0.826

0.105

62.463

1

0.000

2.284

1.861

2.803

0.261

0.033

63.675

1

0.000

1.298

1.218

1.384

0.492

0.069

51.009

1

0.000

1.635

1.429

1.872

Support Network (S)

0.401

0.026

239.715

1

0.000

1.493

1.419

1.571

Support Network (D)

0.919

0.067

188.989

1

0.000

2.507

2.199

2.858

127.666

2

0.000

Regret Leaving Suburb (S) Regret Leaving Suburb (D)

Practical Support (C)

Yes Sometimes

0.526

0.085

38.415

1

0.000

1.693

1.433

2.000

No

1.090

0.106

105.387

1

0.000

2.974

2.415

3.662

Practical Support (S)

0.540

0.048

127.181

1

0.000

1.717

1.563

1.886

Practical Support (D)

0.708

0.071

100.185

1

0.000

2.031

1.768

2.333

209.954

2

0.000

Emotional Support (C)

Yes Sometimes

0.856

0.094

82.553

1

0.000

2.354

1.957

2.831

No

1.388

0.113

151.982

1

0.000

4.007

3.213

4.996

Emotional Support (S)

0.730

0.050

210.877

1

0.000

2.075

1.880

2.290

Emotional Support (D)

1.045

0.076

187.730

1

0.000

2.844

2.449

3.303

Comforts Baby (D)

1.111

0.546

4.135

1

0.042

3.038

1.041

8.866

Enjoys Contact with Baby (D)

1.811

0.417

18.895

1

0.000

6.118

2.703

13.844

Baby Trouble Sleeping (S)

0.406

0.035

133.746

1

0.000

1.501

1.402

1.609

Baby Demanding (S)

0.392

0.031

157.531

1

0.000

1.479

1.392

1.572

Baby Content (S)

0.390

0.045

74.686

1

0.000

1.477

1.352

1.614

Baby Difficult Feeder (S) Baby Difficult to Comfort (S) Health of Child (S)

0.208

0.039

28.207

1

0.000

1.231

1.140

1.330

0.458

0.039

135.301

1

0.000

1.581

1.463

1.707

0.385

0.042

85.619

1

0.000

1.470

1.355

1.595

Attachment

Baby Characteristics

S – Scale, D - Dichotomous, C-Category (Indicator)

Table 11: Bivariate Logistic Regression, EDS > 12, Significant Associations

156


5.4.4 Categorical Principal Component Analysis Categorical Principal Component Analysis was undertaken using all variables, except for the dependent variables, and set for analysis of seven dimensions. Component Loadings Variable

Dimension 1

Baby Content (N) Maternal Expectations Marital Status (N) Household Size (O) Number of Child Under 5 (O) Accommodation (N) Regret Leaving Suburb (O) Employment of Mother (N) Employment of Father (N) Access to Car (O) Education of Mother (O) Mothers Health (O) Practical Support (O) Emotional Support (O) Health of Child (O) Baby Trouble Sleeping (O) Baby Demanding (O) Baby Difficult Feeder (O) Baby Difficult to Comfort Social Support Network (O) Suburb Duration (O) Financial Situation (O) Responds to Child N Comfort Child N Enjoys Contact N Country of Birth Breast Feeding Unplanned Pregnancy Total (Eigenvalue) Cronbach’s Alpha

.097 .116 -.127 .061 .010 .153 .098 -.131 .136 .157 -.117 .138 .105 .125 .141 .092 .110 .063 .112 .159 .075 .168 1.149 1.130 1.104 .091 .082 .113 4.156 .788

2 -.274 -.265 .417 -.212 -.081 -.562 -.329 .516 -.459 -.521 .407 -.418 -.341 -.365 -.400 -.257 -.315 -.199 -.283 -.432 -.281 -.529 .331 .311 .296 -.299 -.148 -.366 3.662 .754

3 .667 .303 .305 -.190 -.138 -.328 -.057 .280 -.244 -.206 .265 -.065 .090 .085 -.012 .752 .736 .612 .742 .028 -.078 -.144 -.038 -.016 -.006 -.061 -.156 -.246 3.218 .715

4 -.128 .165 .559 .022 .052 -.338 .026 .240 -.089 .096 .181 .423 .205 .263 .491 -.176 -.189 -.097 -.167 .424 -.030 .082 -.009 -.007 -.007 .527 -.269 -.341 1.871 .483

5 .082 -.162 .176 .640 .755 -.050 -.241 -.134 .028 -.090 -.091 -.166 .171 .127 -.168 .044 .055 .015 .049 .023 -.334 -.017 .013 .003 .001 .037 -.116 .063 1.379 .285

6 -.087 -.064 -.032 -.242 -.085 -.002 .091 -.048 -.070 .121 .144 -.497 .520 .458 -.493 -.067 -.079 .016 -.042 .300 .230 .027 .004 .004 .005 -.003 -.253 .012 1.320 .251

7 -.063 .088 -.141 -.264 .046 -.216 -.136 .007 -.168 -.248 -.226 .214 .395 .334 .144 -.043 -.060 -.042 -.019 .100 -.384 .016 -.022 -.008 -.003 -.427 .529 .018 1.273 .222

Variable Principal Normalization.

Table 12: CATPCA – Seven Dimensions/Factors

157


4.5 4 3.5 Eigenvalues

Variance

3 2.5 2 1.5 1 0.5 0 1

2

3

4

5

6

7

Dimensions

Figure 30: Shepard (Scree) plot of the first seven dimensions The eigenvalues used in Categorical Principal Component Analysis (CPCA) are equivalent to those of classical PCA. They are measures of how much variance is accounted for by each dimension. The eigenvalues can be used as an indication of how many dimensions are needed. As a general rule the eigenvalues should be larger than 1. Principal component seeks to locate the first component dimension in space that accounts for the maximum proportion of the variance of the variables.

158


Figure 31: Bi Plot of CPCA – Dimension 1 and 2 The long vector (lines) are relatively long indicating that the first two dimensions account for a large amount of the variance of all of the quantified variables. The variables form a bundle having a large positive loading on the first dimension. These variables (Comfort Child, Enjoys Contact, and Responds to Child) may be considered to be representative of maternal attachment. The vectors in this bundle are orthogonal (perpendicular) to the other vectors of which means that this set of variables is uncorrelated with the second set of variables. The second dimension includes all remaining variables with the largest loadings being for the variables: accommodation, employment of mother, access to a car, financial situation, marital status, education of mother, social support network, mothers’ health, and unplanned pregnancy. The second dimension might represent a latent factor related to

social and economic disadvantage.

159


Figure 32: Bi Plot of CPCA – Dimension 3 and 4 The variables baby not content, trouble sleeping, demanding, difficult feeder, and difficult to comfort, form another bundle. The variables in dimension three might be considered to be related to infant behaviour and temperament. The variables marital status, country of birth, health of child, social support network, mothers’ health, unplanned pregnancy, accommodation, not breastfeeding, emotional support and employment of mother form a fourth bundle that is difficult to define. These variables are associated with the second dimension and may therefore provide some useful information about the “common factor”. The dimension might be related to social

marginalisation.

160


Figure 33: Bi Plot of CPCA – Dimension 5 and 6 The fifth dimension is predominantly composed of the variables number of children under five and household size. There is a negative loading on “suburb duration” and “regret leaving suburb”. This dimension is not strongly associated with variables in other dimensions and may relate primarily to the stress of raising three or more children under

five. The variables “mothers’ health” and “child’s health” are negatively associated with social, emotional and practical support. They form a sixth dimension that might represent some underlying factor common to the ideas of support and wellbeing.

161


5.4.5 Assumptions of Causal Inference Before entering variables into a regression model it is important to make explicit assumptions of causal inference. As discussed earlier, care should be taken to not control for variables that are descents (effects) of the outcome understudy (Greenland, Pearl et al. 1999). The analysis will use the “Back-Door Criterion” for selecting covariates for adjustment. Based on the qualitative study (Chapter 4) I propose that maternal depression causes: 1.

Poor maternal attachment or “bonding”

2.

Failure to initiate or continue breastfeeding Z1

X

Z2

Maternal Depression

No Breastfeeding Poor Attachment

Figure 34: DAG for Attachment, Breastfeeding, Smoking The assumption explicit in this DAG is that “not breastfeeding” and “poor attachment” are not causes of maternal depression but may be caused by maternal depression. Based on this analysis the following variables will not be adjusted in the logistic regression: 1. Mother responds to Child 2. Comforts Child 3. Enjoys Contact with Baby 4. Breastfeeding. Based on the findings of the qualitative study, I have assessed that all other variables might be on the causal pathway and they will therefore be included in the logistic regression analysis. I have elected to include maternal self-reported health in the analysis as a possible “upstream” measure of maternal stress. These assumptions may prove to be incorrect but in this emergent and exploratory process I have no prior knowledge of possible causal pathways.

162


5.4.6 Logistic Regression Results 5.4.6.1 Logistic Regression EDS > 9 95% C.I for EXP(B) B

S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

Mothers Expectation(O)

0.629

0.038

272.257

1

0.000

1.875

1.740

2.021

Accommodation(D)

-0.135

0.141

0.910

1

0.340

0.874

0.662

1.153

Sole Parent (D)

0.018

0.144

0.016

1

0.900

1.018

0.768

1.349

No Regret Leaving(D)

0.092

0.071

1.660

1

0.198

1.096

0.953

1.260


-0.202

0.122

2.741

1

0.098

0.817

0.643

1.038

Baby difficult sleep(O)

0.149

0.047

9.944

1

0.002

1.161

1.058

1.274

Baby Demanding (O)

0.090

0.042

4.586

1

0.032

1.094

1.008

1.188

Baby Difficult Feed(O)

0.034

0.043

0.624

1

0.430

1.035

0.950

1.127

Baby Difficult Comfort(O)

0.031

0.051

0.387

1

0.534

1.032

0.935

1.139

Baby Not Content(O)

0.177

0.053

10.982

1

0.001

1.193

1.075

1.325

Practical Support(D)

0.232

0.084

7.645

1

0.006

1.262

1.070

1.487

Emotional Support(D)

0.472

0.093

25.729

1

0.000

1.603

1.336

1.924

Social Support(O)

0.137

0.032

18.580

1

0.000

1.147

1.078

1.221

Financial Situation(O)

0.094

0.020

22.511

1

0.000

1.098

1.057

1.142

Unplanned Pregnancy((D)

0.148

0.066

1.159

1.019

1.319

Household Size (O)

5.015

1

0.025

2.252

2

0.324

Household Size(O) (1)

-0.072

0.085

0.720

1

0.396

0.931

0.789

1.099

Household Size(O)(2)

0.440

0.370

1.418

1

0.234

1.553

0.752

3.207

Blended Family(D)

0.066

0.090

0.533

1

0.465

1.068

0.895

1.273

Suburb Duration(D)

0.079

0.065

1.500

1

0.221

1.083

0.953

1.229

15.988

2

0.000

Car Access(O)(1)

-0.297

0.087

11.570

1

0.001

0.743

0.626

0.882

Car Access(O)(2)

0.116

0.094

1.497

1

0.221

1.123

0.933

1.351


0.235

0.066

12.771

1

0.000

1.264

1.112

1.438

Car Access(O)

Maternal Health(O)

0.371

0.044

69.593

1

0.000

1.449

1.328

1.581

Child Health(O)

-0.036

0.045

0.654

1

0.419

0.964

0.883

1.053

Constant

-6.338

0.200

1004.536

1

0.000

0.002

Table 13: Initial Logistic Regression of EDS > 9 Table 13 shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >9 entered into the initial logistic regression model. Employing a 0.05 criterion of statistical significance, mother’s expectations, country of birth, financial situation, maternal health, unplanned pregnancy, social support, practical support, emotional support, baby with difficulty sleeping, demanding, not being content, and one dummy variable coding access to a car, had significant partial effects. The correlation matrix revealed no signs of multicollinearity. The following predictors were not significant: accommodation, sole parenthood, unemployed father, baby difficult feeding, baby difficult to comfort, household size,

163


blended family, suburb duration, child health and no regret leaving the suburb. The Hosmer and Lemeshow Test was a not significant (Chi-square = 6.412, df 8, p = 0.601) indicating that the data fit the model well. The model was able to correctly classify 98.3 percent of EDS >9 for an overall success rate of 84.1 percent. The area under the ROC curve was 74.2 (95% CI: 72.8-75.5) indicating an adequate fit.

95% C.I for EXP(B) B

S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper


0.636

0.038

281.631

1

0.000

1.889

1.754

2.035

Baby Difficult Sleeping(O)

0.163

0.045

13.138

1

0.000

1.177

1.078

1.285

Baby Demanding(O)

0.104

0.041

6.529

1

0.011

1.110

1.025

1.202

Baby not Content(O)

0.183

0.053

12.088

1

0.001

1.200

1.083

1.331


0.238

0.084

8.072

1

0.004

1.268

1.076

1.494


0.480

0.093

26.822

1

0.000

1.615

1.347

1.937

Social Support(O)

0.136

0.032

18.548

1

0.000

1.146

1.077

1.219


0.089

0.019

21.075

1

0.000

1.093

1.052

1.135

Unplanned Pregnancy(D)

0.151

0.064

5.637

1

0.018

1.163

1.027

1.317

16.700

2

0.000

Car Access Car Access(O)(1)

-0.302

0.086

12.210

1

0.000

0.739

0.624

0.876

Car Access(O)(2)

0.116

0.092

1.586

1

0.208

1.123

0.937

1.346

Country of Birth(D)

0.214

0.064

11.103

1

0.001

1.238

1.092

1.404

Maternal Health(O)

0.353

0.040

77.166

1

0.000

1.423

1.316

1.540

-6.296

0.195

1040.678

1

0.000

0.002

Constant

Table 14: Forward Stepwise (Conditional) Logistic Regression of EDS >9 Table 14 shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >9 entered into a Forward Stepwise LR with probability of 0.05 for entry. Employing a 0.05 criterion of statistical significance, mothers expectation, country of birth, financial situation, maternal health, unplanned pregnancy, social support network, practical support, emotional support, baby with difficulty sleeping, demanding, not being content, and one dummy variable coding access to a car, had significant partial effects. The correlation matrix revealed no signs of multicollinearity. Stepwise with probability for entry of 0.25 added unemployed Father and Suburb Duration but neither were statistically significant at the 0.05 level and both had low Wald Chi Square. Tables 15 and 16 are the preliminary and final models respectively after manual entering and removal of variables.

164



S.E.

Wald

df

Sig.

Maternal Expectation(O)

0.642

0.035

333.564

1

0.000

1.900

1.774

2.035

Baby Difficult Sleep(O)

0.137

0.042

10.556

1

0.001

1.147

1.056

1.246

Baby Demanding(O)

0.081

0.038

4.484

1

0.034

1.084

1.006

1.169

Baby Difficult Feeding(O)

0.051

0.039

1.686

1

0.194

1.052

0.974

1.136

Baby Not Content(O)

0.171

0.049

12.317

1

0.000

1.186

1.078

1.304


0.206

0.075

7.468

1

0.006

1.229

1.060

1.425


0.437

0.083

27.588

1

0.000

1.548

1.315

1.822

Social Support(O)

0.143

0.028

25.408

1

0.000

1.154

1.091

1.220


0.082

0.017

22.170

1

0.000

1.086

1.049

1.124

Unplanned Pregnancy(D)

0.152

0.057

6.954

1

0.008

1.164

1.040

1.302

Country of Birth(D)

0.229

0.058

15.550

1

0.000

1.258

1.122

1.410

Mothers Health(O)

0.352

0.037

90.317

1

0.000

1.421

1.322

1.528

-6.260

0.181

1192.086

1

0.000

0.002

Constant

Exp(B)

Lower

Upper

Table 15: Parsimonious Preliminary Main Effect Model EDS >9 For the preliminary model (Table 15) the Hosmer and Lemeshow Test was a not significant (Chi-square = 4.74, df 8, p = -0.785) indicating that the data fit the model well. The model was able to correctly classify 98.0 percent of EDS >9 for an overall success rate of 83.2 percent. The area under the ROC curve was 74 (95% CI: 72.8 – 75.3) indicating an adequate fit.

df

Sig.

Exp(B)

95% C.I for EXP(B) Lower Upper

B

S.E.

Wald

Maternal Expectation(O)

0.642

0.035

338.182

1

0.000

1.901

1.775

2.036

Baby Difficult Sleep(O)

0.138

0.042

11.032

1

0.001

1.148

1.058

1.246

Baby Demanding(O)

0.092

0.038

5.975

1

0.015

1.096

1.018

1.180

Baby Not Content(O)

0.171

0.048

12.534

1

0.000

1.186

1.079

1.304


0.205

0.075

7.481

1

0.006

1.228

1.060

1.422


0.446

0.083

29.109

1

0.000

1.562

1.328

1.837

Social Support(O)

0.149

0.028

28.040

1

0.000

1.161

1.099

1.227


0.086

0.017

25.094

1

0.000

1.090

1.054

1.127

Country of Birth(D)

0.225

0.058

15.070

1

0.000

1.252

1.118

1.402

Mothers Health(O)

0.365

0.037

98.047

1

0.000

1.440

1.340

1.548

-6.216

0.178

1213.109

1

0.000

0.002

Constant

Table 16: Final Model EDS >9 For the final model (Table 16) the Hosmer and Lemeshow Test was a not significant (Chisquare = 3.682, df 8, p = -0.885) indicating that the data fit the model well. The model was able to correctly classify 98.1 percent of EDS >9 for an overall success rate of 83.2 percent. The area under the ROC curve was 74 (95% CI: 72.7 - 75.2) indicating an adequate fit.

165


Diagnostics

chgdev9

6.00

4.00

2.00

0.00 0.00000

0.20000

0.40000

0.60000

0.80000

1.00000

Predicted probability

Figure 35: Deviance versus Predicted Probability EDS >9 In Figure 35 deviance of studentized residuals are plotted against predicted logistic probability. The curve that extends from the lower left to the upper right corresponds to cases in which the dependent variable has a value of 0. The change in deviance plot illustrates that there are no significantly outlying cases.

Analog of Cook's influence statistics

0.02500

0.02000

0.01500

0.01000

0.00500

0.00000 0.00000

0.20000

0.40000

0.60000

0.80000

1.00000


Figure 36: Cooks Distances versus Predicted Probabilities EDS >9 In Figure 36 the Cook’s distances plot generally follows that of the change in deviance. There are cases in the upper right that are marginal outliers.

166


5.4.6.2 Logistic Regression EDS > 12 95% C.I for EXP(B) B Mothers Expectation(O)

S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

0.774

0.056

192.589

1

0.000

2.169

1.944

2.420

-0.234

0.202

1.350

1

0.245

0.791

0.533

1.175

Sole Parent (D)

0.240

0.185

1.673

1

0.196

1.271

0.884

1.828


0.202

0.101

4.022

1

0.045

1.224

1.005

1.491


-0.195

0.171

1.301

1

0.254

0.823

0.589

1.150


0.180

0.067

7.158

1

0.007

1.197

1.049

1.365

Baby Demanding (O)

0.164

0.058

7.950

1

0.005

1.179

1.051

1.321


-0.019

0.061

0.097

1

0.756

0.981

0.870

1.106


Accommodation(D)

-0.029

0.070

0.171

1

0.680

0.972

0.847

1.114

Baby Not Content(O)

0.163

0.077

4.470

1

0.034

1.177

1.012

1.369


0.250

0.115

4.771

1

0.029

1.284

1.026

1.607


0.569

0.122

21.762

1

0.000

1.766

1.391

2.243

Social Support(O)

0.158

0.043

13.474

1

0.000

1.171

1.077

1.275


0.129

0.029

20.142

1

0.000

1.138

1.076

1.204


0.165

0.095

2.986

1

0.084

1.179

0.978

1.421

4.402

2

0.111

Household Size (O) Household Size(O) (1)

0.048

0.121

0.160

1

0.690

1.049

0.828

1.329


0.919

0.441

4.335

1

0.037

2.506

1.055

5.951

Blended Family(D)

0.099

0.127

0.604

1

0.437

1.104

0.860

1.416

Suburb Duration(D)

0.102

0.094

1.176

1

0.278

1.107

0.921

1.330

3.902

2

0.142

Car Access(O)(1)

-0.103

0.122

0.709

1

0.400

0.902

0.711

1.146

Car Access(O)(2)

0.194

0.130

2.226

1

0.136

1.214

0.941

1.565


0.096

0.097

0.971

1

0.325

1.101

0.909

1.332

Car Access(O)

Maternal Health(O)

0.492

0.064

59.329

1

0.000

1.635

1.443

1.853

Child Health(O)

-0.121

0.065

3.510

1

0.061

0.886

0.780

1.006

Constant

-8.397

0.299

790.979

1

0.000

0.000

Table 17: Logistic Regression EDS > 12 Table 17: shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >12 entered into the initial logistic regression model. Employing a 0.05 criterion of statistical significance, mother’s expectation, no regret leaving, financial situation, maternal health, social support network, practical support, emotional support, baby demanding, difficulty sleeping, not being content, and one dummy variable of household size, had significant partial effects. The correlation matrix revealed no signs of multicollinearity.

167


The following predictors were not significant: accommodation, sole parenthood, unemployed father, baby difficult feeder, baby difficult to comfort, unplanned pregnancy, blended family, suburb duration, car access, country of birth, and child health. The Hosmer and Lemeshow Test was a not significant (Chi-square = 2.743, df 8, p = 0.949) indicating that the data fit the model. The model was able to correctly classify 99.6 percent of EDS >9 for an overall success rate of 99.1 percent. The area under the ROC curve was 79.3 (95% CI: 77.4 – 81.1) indicating an adequate fit.


S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper


0.781

0.055

199.066

1

0.000

2.183

1.959

2.433


0.208

0.099

4.433

1

0.035

1.231

1.014

1.493


0.196

0.061

10.279

1

0.001

1.217

1.079

1.372

Baby Demanding (O)

0.199

0.054

13.484

1

0.000

1.220

1.097

1.356


0.246

0.114

4.674

1

0.031

1.279

1.023

1.599


0.566

0.121

21.973

1

0.000

1.761

1.390

2.232

Social Support(O)

0.158

0.041

14.568

1

0.000

1.171

1.080

1.270


0.134

0.028

23.679

1

0.000

1.143

1.083

1.207


0.220

0.091

5.802

1

0.016

1.246

1.042

1.489

Maternal Health(O)

0.443

0.059

56.969

1

0.000

1.558

1.388

1.748

-8.318

0.278

897.418

1

0.000

0.000

Constant

Table 18: Forward Stepwise (Conditional) Logistic Regression of EDS >12 Table 18 shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >9 entered into a Forward Stepwise LR with probability of 0.05 for entry. Employing a 0.05 criterion of statistical significance, mothers expectation, no regret leaving, financial situation, maternal health, unplanned pregnancy, social support network, practical support, emotional support, baby with difficulty sleeping, and baby demanding had significant partial effects. The correlation matrix revealed no signs of multicollinearity. The Hosmer and Lemeshow Test was a not significant (Chi-square = 14.857, df 8, p = 0.06) indicating that the data fit the model poorly. The model was able to correctly classify 99.6 percent of EDS >9 for an overall success rate of 93.1 percent. The area under the ROC curve was 788 (95% CI: 77.1 – 80.4) indicating an adequate fit. Backward Stepwise (Conditional) regression gave similar results. Tables 19 and 20 are the preliminary and final models respectively after manual entering and removal of variables.

168



S.E.

Wald

df

Sig.


0.765

0.050

237.118

1

0.000

2.148

1.949

2.368


0.131

0.056

5.501

1

0.019

1.139

1.022

1.271

Baby Demanding (O)

0.173

0.049

12.474

1

0.000

1.189

1.080

1.309


0.583

0.099

34.442

1

0.000

1.791

1.474

2.176

Social Support(O)

0.214

0.035

36.804

1

0.000

1.239

1.156

1.327


0.151

0.024

38.748

1

0.000

1.163

1.109

1.219

Maternal Health(O)

0.471

0.053

79.625

1

0.000

1.601

1.444

1.776

-8.132

0.246

1090.605

1

0.000

0.000

Constant

Exp(B)

Lower

Upper

Table 19: Parsimonious Preliminary Main Effect Model EDS >12 For the preliminary model (Table 19) the Hosmer and Lemeshow Test was a not significant (Chi-square = 7.645., df 8, p = -0.469) indicating that the data fit the model poorly. The model was able to correctly classify 99.6 percent of EDS >12 for an overall success rate of 92.4 percent. The area under the ROC curve was 78.3 (95% CI: 76.6 – 80.0) indicating an adequate fit.


S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper


0.757

0.050

229.866

1

0.000

2.131

1.933

2.350


0.148

0.056

6.900

1

0.009

1.160

1.038

1.295

Baby Demanding (O)

0.155

0.050

9.697

1

0.002

1.167

1.059

1.287

Sole Parent (D)

0.444

0.114

15.282

1

0.000

1.559

1.248

1.947


0.575

0.100

33.012

1

0.000

1.777

1.461

2.162

Social Support(O)

0.203

0.037

30.695

1

0.000

1.226

1.141

1.317


0.132

0.025

27.694

1

0.000

1.141

1.086

1.199

Maternal Health(O)

0.460

0.053

74.259

1

0.000

1.583

1.426

1.758

Country of Birth(D)

0.091

0.084

1.174

1

0.279

1.095

0.929

1.291

-8.051

0.248

1054.068

1

0.000

0.000

Constant

Table 20: Final Model EDS > 12 For the final model (Table 20) the Hosmer and Lemeshow Test was a not significant (Chisquare = 2.921 df 8, p = -0.939) indicating that the data fit the model well. The model was able to correctly classify 99.6 percent of EDS >12 for an overall success rate of 92.5 percent. The area under the ROC curve was 78.5 (95% CI: 76.8 – 80.1) indicating an adequate fit. Country of Birth was not significant but its inclusion in the model improved the fit to a Hosmer and Lemeshow Test of p=0.939.

169


Diagnostics 10.00

chgdev12

8.00

6.00

4.00

2.00

0.00 0.00000

0.20000

0.40000

0.60000

0.80000


Figure 37: Deviance versus Predicted Probability EDS >12 In Figure 37 deviance of studentized residuals are plotted against predicted logistic probability. The curve that extends from the lower left to the upper right corresponds to cases in which the dependent variable has a value of 0. The change in deviance plot illustrates that there are no significantly outlying cases.


0.03000

0.02000

0.01000

0.00000 0.00000

0.20000

0.40000

0.60000

0.80000


Figure 38: Cooks Distances versus Predicted Probabilities EDS >12 In Figure 38 the Cook’s distances plot generally follows that of the change in deviance. There are cases in the upper left that may be outliers.

170


5.6

Theory Generation

Categorical PCA was used in Section 5.4.4 as part of the exploratory data analysis (EDA). In that use the CatPCA was principally used to detect patterns and phenomenon in the data. Haig (2005) argues, however, that while factor analysis (in this case CatPCA) can be used for EDA it should be “properly construed as a method for generating explanatory theories”. He further argues:

“that EFA helps researchers generate theories with genuine explanatory merit; that factor indeterminacy is a methodological challenge for both EFA and confirmatory factor analysis, but that the challenge can be satisfactorily met in each case; and, that EFA, as a useful method of theory generation, can be profitably employed in tandem with confirmatory factor analysis and other methods of theory evaluation.” (Haig 2005, p 303) As part of a critical realist approach to theory building, the earlier sections of this Chapter were principally about description and phenomenon detection (Danermark, Ekstrom et al. 2002; Haig 2005). In this section I will utilise two forms of factor analysis for theory generation purposes.

5.6.1 CatPCA Revisited Examination of the above CatPCA results reveals strong loading of variables on two dimensions that probably relate to theoretical concepts of maternal attachment and infant temperament. Examination of Figure 32 suggests that there are at least two further significant dimensions, one relating to disadvantage or social exclusion and the other to support, country of birth and mothers’ perceived health. The nature of the remaining dimensions is less clear. I have revisited the CatPCA analysis by excluding variables related to maternal attachment and infant temperament. The purpose is to improve the visualisation of the remaining component loadings.

171


Component Loadings Dimension 1 Maternal Expectations

2

3

4

5

6

7

-.177

.329

-.268

.055

.076

-.164

.278

.507

.558

.250

.010

-.082

.276

-.175

Household Size (O)

-.266

-.065

.714

.022

-.037

.113

.215

Number of Child Under 5 (O)

-.123

-.031

.694

-.144

.259

.216

.232

Accommodation (N)

-.649

-.370

-.016

-.007

-.211

.038

.093

Regret Leaving Suburb (O)

-.334

-.016

-.218

-.051

-.178

.485

.378

Employment of Mother (N)

-.590

-.266

.094

-.091

.014

-.089

-.211

Employment of Father (N)

-.524

-.118

.112

.060

-.159

-.096

-.311

Access to Car (O)

-.580

.072

-.020

-.066

-.270

.080

-.259

.486

.201

-.092

-.121

-.218

.009

.336

Mothers Health (O)

-.439

.405

-.018

.542

.148

-.092

.172

Practical Support (O)

-.318

.293

-.099

-.523

.386

-.018

.033

Emotional Support (O)

-.343

.345

-.099

-.459

.328

-.024

.025

Health of Child (O)

-.405

.484

.001

.542

.086

-.075

.151

Social Support Network (O)

-.432

.465

-.065

-.263

.080

.096

-.091

Suburb Duration (O)

-.299

-.022

-.275

-.123

-.442

.407

.147

Financial Situation (O)

-.572

.093

-.017

-.009

-.005

-.183

-.127

Country of Birth

-.319

.477

.212

.026

-.401

.062

-.161

Breast Feeding

-.209

-.303

-.208

.335

.584

.563

-.191

Unplanned Pregnancy

-.436

-.356

.008

-.061

.036

-.287

.444

Total Eigenvalue Cronbach’s Alpha

3.619

1.966

1.393

1.330

1.294

1.050

1.040

.762

.517

.297

.261

.239

.050

.040

Marital Status (N)

Education of Mother (O)

Object Principal Normalization.

Table 21: CPCA – Infant temperament and Maternal Attachment variables removed The first dimension includes variables previously described for the dimension social and

economic disadvantage and included the variables accommodation, employment of mother and father, access to a car, financial situation, marital status, education of mother, maternal health, and unplanned pregnancy. It is notable that poor social support network also loads on this latent variable. Dimension two has similar loadings to the previous analysis with marital status, country of birth, health of child, social support network, mothers’ health forming and ill-defined bundle. This dimension appears to be related to social isolation and migration (ethnicity). As previously discussed dimension three (previously five) is related to household size and number of children under five. Dimension four loads mothers health, child health, practical support and emotional support. The underlying factor might represent the ideas of support

and wellbeing.

172


Figure 39: Bi Plot of CPCA – Dimension 1 and 2 with infant related variables removed Visualisation of the component loadings suggests clustering along two axes. One axis, (called social isolation and migration), includes country of birth, social support network, emotional support, practical support and self reported maternal and child health. Financial situation is closely related. The second axis includes the remaining variables but with high loading for the variables: marital status, unplanned pregnancy, education of mother, accommodation, employment of mother and father, access to car, and financial situation. This axis seems to be related to social and economic disadvantage.

173


Figure 40: Bi Plot of CPCA – Dimension 3 and 4 with infant related variables removed The third and fourth dimensions had similar loadings to those previously described in the first analysis for the fifth and sixth dimensions respectively. The third dimension is predominantly composed of the variables that may be related to the stress of raising three or more children under five. The fourth dimension includes self-reported health, child health and three support variables. Paradoxically the support variables have a negative loading compared to selfreported health suggesting that in some situations support may not be protective. This bundle might represent some underlying factor related to support and wellbeing The fifth dimension included not breast feeding, lack of emotional and practical support, Australian birth, short suburb duration, and having access to a car. The sixth dimension includes no regret leaving the suburb, short suburb duration, not breast feeding and increased number of children under five.

174


5.6.2 CFA Exploratory Specification Search (AMOS 17.0) The purpose of the DAG and regression exercises was to further examine the relationships in the data by excluding possible spurious associations. This was done by excluding variables that may be “descendent” of the phenomenon “depressive symptoms” and by “controlling” for possible confounding in the data. Based on those analyses I have elected to proceed with further “theory generation” analysis using exploration of both three and four factor confirmatory factor analysis (CFA) measurement models. The DAG analysis indicated that attachment is a descendent of “depressive symptoms” and therefore variables related to that dimension will not be included. Based on the previous analysis and integration with the qualitative study findings I assessed that a three factor measurement model would probably represent the main theoretical concepts of infant temperament, social disadvantage and support or isolation. The finding in the CatPCA of two latent variables related to “support” suggests that a four factor measurement model may also be helpful. The three factor model scree plot of the best fit of AIC indicated that two models may represent the best fitting models to the data. These two models are shown in Figure 42 below. The loading of the variables on the three latent variables suggest that latent variable F3 is related to infant behaviour or temperament and F2 is related to the concept of support. Latent variable F1 always loads on sole parenthood, maternal expectations, poor selfreported health, financial difficulty, no social support and mother not born in Australia Those loadings suggest that F1 relates to social and economic disadvantage and marginalisation. It also includes social support and mother not born in Australia. A four factor model may tease these dimensions out.

175


Figure 41: Three Factor Specification Search – Two best models

The scree plot for “Best Fit for AIC” of the Four Factor Model Specification Search is shown below (Figure 41).

Figure 42: Plot of Best Fit for AIC Four Factor Measurement Model Specification Search

176


The Four Factor Measurement Model scree plot (Figure 41) of the “best fit of AIC” indicated that two models may represent the best fitting models to the data. The two models were essentially the same, apart from the loading of F2 on practical support. As with the three factor models there is a latent variable (F4) loading on variables related to infant behaviour. F3 is loading on three measures of lack of support, sole parenthood and financial difficulties. F2 has complex associations with maternal expectations being met (i.e. it has a negative sign), not born in Australia, no social support, and in the second model, no practical support. F1 is strongly associated with unmet maternal expectations (i.e. it has a positive sign), poor self reported health, and difficult financial situation. The interpretation of these complex and paradoxical findings is in Section 5.6.3.

Figure 43: Four Factor Specification Search – Two best models

177


5.6.3 Emerging Theory As expected in all four factor models there was consistent loading of infant temperament variables on F4 and the no support variables on F3. The three factor models were similar with consistent loading of infant temperament on F3 and low support variables on F2. The association of sole parenthood and financial difficulties with F3 is consistent with interpreting F3 to represent a latent variable of “isolation and unsupported”. The four factor models consistently loaded and financial situation and maternal health on F1 and social support and country of birth on F2. These findings suggest that the two latent variables F1 and F2 represent “economic marginalisation” and “ethnic social marginalisation” respectively. The inclusion of maternal expectations into the measurement model did not alter these findings or interpretations. Unmet maternal expectation was strongly associated with F1 and negatively associated with F2. The association of economic marginalisation with unmet expectations is consistent with findings emerging from the qualitative studies. The negative loading of maternal expectations on F2, together with “not born in Australia” and “no support”, suggests that for these mothers their expectation were exceeded.

178


The findings from the individual level qualitative study (Chapter 4) can be used here to suggest that the two latent variables F1 and F2 represent “economic isolation” and “ethnic social isolation” respectively. Having no support is represented by F3 and the counterfactual, having support, might be protective. The infant temperament related F4 is a possible cause of maternal stress. Maternal expectations are clearly important but their role is complex. Figure 44 represents a tentative illustration of the emerging causal inference arising from the associations found in this Chapter and interpreted with mechanistic findings from the concurrent qualitative study (Chapter 4).

Economic isolation (+) Ethnic social isolation

(+) Depressive Symptoms

(-) Social support

(+) (?)

Infant Temperament

Maternal Expectation

Figure 44: Emerging Causal Inference

179


5.7

Conclusion

The purpose of this Chapter was to describe the findings of quantitative exploration of individual level factors that might be associated with the phenomenon of perinatal depression, and perinatal adversity. The bivariate exploration identified associations with maternal expectations, marital status, blended family, household size, number of children under five, breast feeding, country of birth, sole parenthood, accommodation, financial situation, employment of mother and father, car access, phone access, planned pregnancy, self-reported health, suburb duration, regret leaving the suburb, support network, emotional support, practical support, and measures of infant behaviour and maternal attachment. Analysis of likely causal relations suggested that breastfeeding and measures of maternal attachment should not be conditioned for in the logistic regression as they were most likely outcomes of depression rather than causative factors. The exploratory logistic regression consistently identified parsimonious set of variables that included: maternal expectations, sole parenthood, country of birth, financial situation, self reported health, support network, emotional support, practical support and measures of infant behaviour. Exploratory factor analysis using categorical principal component analysis and AMOS(18.0) CFA specification search indicated, that there are at least four latent factors which I have called: economic isolation, ethnic social isolation, social support and infant temperament. The role of maternal expectation is important but complex with possible different mechanisms among Australian born mothers and migrant mothers. Critical analysis of this Chapter can be found in Chapter 10 – Conclusion, Limitations and Implications. The following Chapter will describe the qualitative exploration of group level structures and mechanisms that may influence the phenomenon of postnatal depression.

180

Part C: Group Level

Part C: Group Level Exploratory Analysis Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







181

Part C: Group Level

182

Chapter 6: Ecological Qualitative

Chapter 6: Qualitative Study at the Group Level Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







183


6.1

Introduction

The purpose of this Chapter is to describe the qualitative exploration of group level factors that might be associated with the phenomenon of perinatal depression, perinatal adversity and their possible impact on developmental and health outcomes for the foetus and infant. The study of individual level factors was previously reported in Chapter 4 using the same methodology and methods to that used here. That study identified six key concepts for consideration in theory development. They were: stress; infant temperament; expectations and dreams; marginalisation and “being alone”; loss of power and control; and support and nurturing. The findings of Chapter 5 were complimentary identifying the following variables as being associated with postnatal depression: support, infant temperament, maternal self reported health, financial stress and not being born in Australia. The focus of the overall study is to examine the role of neighbourhood context in relation to the developmental origins of health and disease. The Study reported in this Chapter seeks to identify the possible mechanisms and pathways by which the neighbourhood and other “Group Level” factors may operate. As previously noted this study was undertaken concurrently with the group level quantitative studies (Chapter 7) which both informed, and were informed by the qualitative study findings.

184


6.2

Aims

The aims of this chapter are to: 1. Undertake a qualitative study of group level factors associated with perinatal depression and the developmental origins of health and disease 2. Use the findings to inform the emerging theory. The specific analysis questions were: 1. Are there more depressed mothers in some suburbs than others? 2. Why is there more depression in some suburbs? 3. What are the characteristics of suburbs that have a high or low numbers of depressed mothers? 4. Are there phenomenon at a city, state and national level that might increase or decrease mothers depression?

6.3

Methods

I have previously discussed the philosophical and methodological approaches (Chapter 2) and methods (Chapter 4) used in this study. In particular the Method used is the same as that described in Section 4.3. As previously open coding was the predominant approach taken. Two preset categories, social support and disadvantaged “depressed” communities, were used. As in Chapter 4 the qualitative study findings will be presented in three parts. The phenomenon and concepts identified in the initial and focused coding will be presented first followed by the findings of the situational analysis. Finally the findings of early theory generation will be presented prior to taking them forward to Chapter 7 for theory construction.

185


6.4


The qualitative research results are reported thematically below. Confidentiality is preserved for comments and quotes by only identifying them as mothers group or experts. Photographs taken in the study communities are presented in figures to illustrate some of the themes and issues that emerged. While experts talked freely about the importance of community or neighbourhood level factors the mothers were less certain. Mothers generally believed that individual level factors were more important because “each individual that is in that suburb is in different circumstances anyway”. This view is about the importance of circumstances reaching into the individual level experience is consistent with that articulated by Clarke (2005 ,p 71-72)

“The conditions of the situation are in the situation. There is no such thing as "context." The conditional elements of the situation need to be specified in the analysis of the situation itself as they are constitutive of it, not merely surrounding it or framing it or contributing to it. They are it. Regardless of whether some might construe them as local or global, internal or external, close-in or far away or whatever, the fundamental question is ‘How do these conditions appear-make themselves felt as consequential-inside the empirical situation under examination?’ “

6.4.1 Community-Level Social Support Networks As discussed in Chapter 4, support was seen by mothers and experts as an important protection against maternal depression. That support had a number of sub-categories including partner support, family support, emotional support, and access to services. In addition, support networks were often mentioned by both experts and mothers as being important for providing mothers with support. Those support networks usually included family members and friends and seemed to operate mainly at the level of the individual mother.

186


Figure 45: Focused Network of Social Support There seemed, however, to be an element of social support that might operate at the group level and be stronger in some communities that others. For example, one mothers group felt that wealthy suburbs would have more support and that that was because they have less to worry about, and others felt that the support was related to access to mothers groups, services and other resources in the community. By contrast all focus groups, and some experts, identified community safety as impacting on support networks and maternal isolation. One focus group spoke of the isolation that resulted from fear to go out to the park or other places where they might meet others. This link between the categories of “support”, “social network” and “isolation” was explained by one mother as “sometimes may be there is lack of support, if you don’t have a big network, you just might feel isolated”. The themes of isolation and marginalisation have a relationship with social support and were discussed previously. One mothers group drew attention to the role of language and culture noting that “may be some people don’t speak English, we have a special language group… if they don’t know they just isolate… may be some culture they don’t know how to have friends … sometimes we can’t mix with other people, you have to move with the same religion [or] culture.” Another group felt that some communities might have more maternal

187


depression because there were more new immigrants who might not have “family or support” or ”an established social network”. That same group felt that communities with less depression would have a “strong community bond and know where the community health nurse and that sort of thing is”. They also felt that that community would be “close knit”. These mothers groups’ comments suggested a close link between the categories of “support”, “social networks” and “community bonds”. One expert spoke of “bonding networks” as an element of “social capital” where those who are like, support each other but often exclude groups who may not be like”. By contrast that same expert spoke of “bridging networks which are networks between dissimilar populations. The more of those you have the higher levels of social capital”. There also seemed to be physical aspects of communities that might contribute to social support networks. One mothers group talked about the importance of parks and playgrounds where they would come across other parents. The importance of town centres was raised by both experts and parents as places where mothers can meet each other. Access to the community shopping centres was seen by one expert as being important for social networks and “to feel linked”. There was also comment that the development of large shopping malls had drawn people away from local community centres and had had a negative impact on local social networks. The lack of public transport from suburbs to the large community centres contributed to isolation and lack of social networks. A number of experts believed that social networks could protect mothers from the impact of living in disadvantaged depressed communities. One commented that “through networks and supports … reduction of stress is likely to result in less alcohol consumption … smoking and less likely to be depression”. Another expert stated that “the reality is if there is 100 people you know or 30 people you can talk to, the rest of society’s impact on you is hard to detect”. A related category was connectedness. Several experts spoke of the positive role that community connectedness might play and one spoke of “belonging to community” which I have interpreted as being a part of connectedness. Connectedness was seen as a protective or resilience factor that might enable a “mother to ride out depression better than someone in another neighbourhood who didn’t have access to that good connection to community”. A closely related category was connection to neighbours and several experts spoke of these related categories in terms of providing social support, possibly through the presence of increased social networks.

188


Home visiting was clearly a way by which mothers receive support from services. Several mothers spoke of the help and support given by the community child health nurses. Two experts also spoke of research showing a positive impact of regular nurse home visiting on maternal and infant outcomes. Those experts did not relate this directly to impacts on maternal depression but to longer term outcomes for both mother and infant. Mothers also spoke of the importance of antenatal classes and the nurse home visits to their involvement with mothers groups which provided “those connections from the beginning to continue all the way through”.

6.4.2 Social Capital The initial literature review reported in Chapter 3 did not suggest a relationship between postnatal depression and social capital and the data from the focus groups with mothers, did not identify social capital as a category. Thus the link between perinatal depression and social capital does not appear to be a direct one. The category “social capital” emerged from the initial expert interviews principally in relation to the related categories of “social networks”, “social support” and “connectedness”. Social capital was also discussed by those experts in relation to negative terms such as “depressed community”, “segregation” and “marginalisation”.

Note: As will be discussed later, there are at least two different schools of “social capital”. Kawachi and colleagues argue that social capital has both individual and group level attributes and that it can be conceptualised as both social capital and as resources embedded in networks. There is, therefore, an overlap between concepts in this section and those discussed in the previous section in relation to community-level social support networks.

189


Figure 46: Focused Network of Social Capital One expert noted that “there are higher rates of child abuse amongst poor populations that also have low social capital”. The link was drawn to maternal perinatal emotional health and associated physical and emotional abuse. It was thus speculated that where there was no social capital there would be marginalisation and no social support networks. The same key informant drew links between high social capital and “inclusiveness as a culture” as being not that much different to a homogenous population where marginalisation is not tolerated or encouraged. Thus a community with strong social capital would have social support networks and no marginalisation. Social capital was further described as being affected by “levels of the size of the gap between rich and poor” with research showing a “consistent linear relationship of bigger the gap between rich and poor [the] lower [the] levels of social capital”. The implication of this is, [the] idea that there would be low social capital where there is inequality. Other experts, however, talked of low social capital existing in disadvantaged “depressed communities”, and in communities where there was segregation or marginalisation of people along ethnic and religious lines. Similarly trust, safety, tolerance and high rates of volunteer activity were considered to be part of social capital and were phenomenon that would not be found in “depressed

190


communities”. One expert spoke of “bridging networks which are networks between dissimilar populations” as contributing to social capital. “Bonding networks” were described as being between people who were like and “they often exclude groups who may not be like”. Communities with strong social capital were also described as places where women would have a voice and where there were high levels of education of both women and men. Several experts spoke of “building community” and the importance of “investing in the people and the community to make the community theirs”. Physical attributes were described as contributing to social capital including parks, play grounds, community meeting places, and community gardens. One expert spoke of the importance of local football teams. She noted that “we had a wonderful system of football teams in Sydney and it is completely disrupted .. it got nationalised and that regional cohesion got lost … The football teams should be investing in the community”. Another expert spoke of the role of big business and community leaders in investing in the community. She talked of a community activity to build a community garden on council land, but the council leased the land to big business for the building of a franchise fast food outlet. Both of these key informants gave examples where business had failed to contribute to social capital. Another expert, however, spoke of the positive contribution made by local businesses to community development activities. She also spoke of local governments in Sydney that were known to be making significant efforts to build the social capital of their communities.

6.4.3 “Depressed” Community A very strong theme emerging was the image of a “depressed” community where there might be elements of both physical and social environmental deprivation. This theme was to be expected given the nature of the interview question. The category “depressed” community had a large number of sub-categories that together give a fuller picture of a depressed community. They include: low family income, high rates of tobacco smoking, high rates of unplanned pregnancy, high rates of sole parenthood, community not looking cared for, high rates of crime, domestic violence, gangs walking the street, lack of buses, lack of space for people to socialise in, fewer amenities, houses that are either cold or hot, a lack of telephones, rental and public housing, poor quality housing, lack of trust, racism, migrant isolation, transient populations, vacant housing, young unemployed people on the streets in the evenings, high numbers of mentally unwell residents, high population density, lack of connectedness, lack of social inclusion, and a depressed mood in the community.

191


Figure 47: Focused Network of “depressed” community When mothers were asked to describe a community with high rates of depression they described such a community as being “grubby”, “having transport problems”, “not being looked after”, “people doing it tough”, “being in the middle of nowhere and nowhere to go to”, “isolation”, “new immigrants to the country … don’t have established social networks”, “environments that is not probably not healthy for your child”, “there is violence”, “where you can’t get to say, a mothers’ group, don’t have family close by”, ”you can’t get out and about to shops”, “stuck in the middle of nowhere … being couped up in a house with a kid 24/7”, “violent people then you choose not to go the park … you don’t want to especially now you have a baby, you want to keep him safe … then you are isolated”. The picture given by these mothers is of a “grubby” poorly cared for community with violence, poor transport and resulting isolation. The emphasis is on isolation and this isolation I have identified as a major emerging category. Isolation was also related by mothers to geographical isolation, “if you are in the middle of nowhere and nowhere to go to”, and social isolation, for example, if parents were new to a country and had no social networks. Experts drew on their field experience and knowledge of literature. One expert described areas of disadvantage “where there is higher unemployment rates often associated with high alcohol drinking, cigarette smoking, violence … and higher rates of unwanted

192


pregnancies”. That same expert commented that “poor communities, disadvantaged communities, have lower social capital and [there is] more dysfunction, more violence, more stress, [and] more depression” thus drawing an inferential link between disadvantaged communities, stress and depression. Other experts gave colourful descriptions of disadvantaged communities that they had worked in – “streets are barren, there aren’t people walking in the streets … youth walk around in gangs, so mums won’t let [children] walk, … houses are poorly kept … gardens overgrown … lounges on the front veranda” Both mothers and experts spoke of the importance of transport. Communities were described as having an irregular bus service where there was an expectation that people would have a car but “there is only one car in the family, the partner in the house has the car. The woman is at home. It is an effort to pack babies and kids up and take them out and do shopping and do all those things”. The lack of transport is related to poor access to services and amenities. This access to services I have identified, as a major emerging theme that is discussed further below. One expert spoke of social inclusion as being a protective factor. They stated that “its more likely that there are excluded and isolated people out there in the disadvantaged areas”. That same expert noted, however, that where people don’t connect up it doesn’t necessarily mean it has to be a disadvantaged area “but a place where people are fearful of what’s going on, even if there’s not a high crime rate there’s a perception of crime, and lack of safety in the area”. Another expert talked of a housing estate that is on the wrong side of the railway track and highway where parents don’t have access to private transport and many don’t have phones, so they are socially isolated. They don’t have money so every day is a struggle financially and they have neighbours they don’t trust because there is lots of trouble in their district, the crime, different race of people.” That expert also spoke of racism in that community “you find the highest levels of racism, in particular against Aboriginal people in communities where they’re living next door to struggling Aboriginal people.” The categories of social exclusion, isolation and racism were thus linked to the major category “depressed” communities. The picture was not that clear, however, when looked at in terms of ethnic diversity. One expert reflected on her experience of racism and ethnic diversity in a disadvantaged community “I have often wondered in [Suburb] the kind of interplay, racism is going up and people don’t think that multiculturalism makes life better … I’ve often wondered if that as an area becomes more depressed and disadvantaged, and

193


there’s less opportunities and lots of structural problems, and less connectedness. Whether that means people will go at each [other] more and won’t be about to support each other more. Where as if you’ve got those other things in place the ethnic diversity comes into play in a good way … And because there’s scarce resources and people are fighting for scarce resources and then its us against them kind of thing. Where as if there’s enough, then everyone can trust and interact with each other more. Then it just acts as a greater strength”. Comments related to racism were all made within the context of discussion about disadvantage. Other related categories such as ethnic diversity, ethnic clusters and segregation were not always related to the major category of “depressed community” and are discussed separately below.

6.4.4 Access to Services at the Group-level Access at the individual-level was identified in Chapter 5 as a major concept [theme] that protects mothers from isolation. The quantitative studies in Chapter 6 identified access to telephone and car as important variables that were not significant in the final logistic regression when controlling for other variables such as financial difficulty.

Figure 48: Focused Network of Access At the group level, access is discussed here in the context of place. Access has also emerged as a major concept at the group-level with related sub-concepts that include: access to medical care, access to shopping centres, access to baby checks, access to

194


child care, access to services, location of services, access to books, information, transport, coordination of services, services not talking to each other and courage to make a call for help. Mothers did not use the phrase “access” which was generally used by experts. Mothers did, however, talk of transport problems often in relation to comments on isolation – “I think transport problem is significant, it is hard to get out and about, I suppose there is isolation”. When talking about communities with a low rate of depression mothers linked this to information - “knowing where the community nurse is and that sort of thing” and access to services – “prominent Community Health Centre”, “I guess it is more to do with plenty of supply of resources, if you need them, just having in your mind that it is there”. Mothers spoke of the importance of the services provided by midwives in the hospital including services for postnatal depression “like the questionnaire and stuff like that are very good”. Having someone to “look over a mother .. especially if there is not a lot of support from the partner at least there is someone there to identify that they might have postnatal depression”. The home visiting nurse services and mothers groups were considered to be important services to access. Experts included discussion of access when discussing the themes of “depressed” community and social capital. One expert when discussing “depressed” communities described them as being further away from resources. Those resources included community amenities such as shops, banks, parks, telephones, and services such as child care, community health centres and medical services. Health services were seen as being able to provide a support system for mothers during and after pregnancy where the mother “feels happy and supported”. Difficulties with access to services were identified by experts including the multiple referrals of some mothers to multiple services and the failure to refer others who had identified needs. A failure of services to communicate with each other was identified by one expert who felt there should be better coordination of services. The accessing of services was seen by both mothers and experts as being particularly difficult for new immigrants who did not have established networks, had difficulty communicating their needs and accessing interpreter services.

195


6.4.5 Big Business Big businesses were seen as important players in the provision of support to communities and mothers. Big business and media were variably described as having both positive and negative influences. The focused network is thus complex and at times paradoxical. Maternity Leave

Power of industry

Lack of space for people to socialise

Access to shopping centres

Building community

Less amenities Meeting places

is associated withBusiness Park is a cause of Job security is a cause of

Perceived image of parenthood is a cause of

Housing

is associated with is associatediswith associated with development

Lack of buses

mistakes

is a cause of

is associated with

is associated with

Advertising

Shopping Centre is part of

is part of

is associated with

is a cause of

Shopping centre good for connection

is part of

is a cause of

is associated with is a cause of

is a cause of

is a cause of

Planning location of industry

is associated with

is associated with

Social capital~

Big Business is a cause of

Contradicts

is a cause of

is associated with

Business social responsibility

is associated with

Shopping centre draws people from community space

is part of Contradicts

is a cause of

Opportunities for entertainment

is a cause of

Globalisation is a cause of

is associated with

Football teams is associated with

is associated with

Global Economy Business Media

Recruit boys to rugby teams

is a cause of

Disrupts the community

The mall

Figure 49: Focused Network of Big Business The impact of sports franchises on local communities was discussed by one expert who felt that an important contribution to community had been removed when sports codes had been “nationalised”. Another expert spoke of the impact that shopping malls were having on small community centres where both people and services were attracted away from small communities to the large town centres. “we are drawing people out of our public spaces which in fact creates a sense of safety and community to a closed environment which means you are only seeing people when you shop, you are only feeling safe in those shopping centres .. They are not going to a local park because the Council has locked it up.” While this was seen by some as having a negative impact, others felt that the new shopping malls provided opportunities for mothers to network and to access services. The owners of large shopping malls have the ability to make both positive and negative impact. One expert was clearly upset at the limited involvement that local shopping centre

196


owners were having in their community and spoke of the impact on children’s play space “we should actually be building the shopping centres around the playground, making the playground in the shopping centre. … Then what are we doing, we are actually taking away the ground space from the children and saying the place where you play is the shopping centre”. The large shopping centres were also seen as placing financial pressure on families – “those big shopping centres are the place to meet people. The focus of the eye is drawn a bit like the dot on the blank wall, the eye is drawn to the spending of money as the way to happiness and connecting to people”. The city councils are also impacted by big business. An example was given by one expert of the failure of local government to support the development of a community gardening project, electing instead to lease the land to a large fast food franchise.

6.4.6 Supportive Social Policy A theme emerged in relation to what I have called supportive social policy. Experts and mothers spoke of the importance of social policy matters such as financial support for families, maternity leave, and free child care. One expert spoke often of comparisons with the Scandinavian countries while others talked of matters related to local council social planning.

197


Figure 50: Focused Network of Supportive Social Policy Mothers spoke of the importance of financial help as a way of decreasing postnatal depression. This was seen as the national Governments responsibility. Related to the financial support was a question related to whether Medicare was available to support mothers with postnatal depression. There was concern expressed regarding the means testing for rental support and a view expressed that “financial assistance” should be for everyone. The lack of government financial support was linked by one mother to stress. She stated that “usually by the time you have a child you are living beyond your means anyway”. Another mother spoke of policy for universal paid maternity leave and noted that extended paternity leave would also help. She felt that some mothers would get some help from their husbands if they stayed home longer. Experts supported the view that financial support was important. One expert linked financial inclusions and food security with the “ability to be in control of your life”. In relation to this loss of control she noted that “if a woman is feeling out of control and in an unhealthy relationship there would be links to maternal depression and other poor outcomes”. One expert felt that there needed to be a shift in employment so that people did not have to travel so far to get work. She suggested there be more small business parks so that people did not need to travel to the city for work. Another expert felt that the local councils had “lifted their game in terms of social planning” but also noted that there were limited resources. She further noted that the shift in population had come before the state infrastructure was in place such as roads, public transport, schools and hospitals. As a result the non-Government sector had had a massive influence with a focus on community development programs. Another expert spoke of the active work of councils such as Fairfield with active investment in community development initiatives such as playgroups, community parks and transition to school initiatives. This expert felt that there was a big difference between the activities undertaken by the Fairfield Council and others in the region. She felt that at the State level they could develop models for councils to follow in relation to activities that might reduce maternal depression. Another expert talked of the excellent work being undertaken by Marrickville Council. This expert felt that some local councils “become so immersed in the politics and bureaucracy, their ability to partner purposefully and openly is limited by that political nature of the beast”. She drew a link with lack of social cohesions

198


and noted that some councils, were not investing in community work while others such as Fairfield and Campbelltown were. Another expert spoke of the development of social segregation based on “what school you go to”. This was illustrated as “so you have neighbourhood kids where one is going in his 4 wheel drive off to the private school and the other is walking to his state school and they are strangers next door”. This link between social segregation and gaps between rich and poor was made by other experts noting that “what is happening is there is once again this equality thing, where everybody is poor you don’t have big differences in socioeconomic status [and] there appears to be less depression”. That same expert went on to note that where there are those differences associated with higher crime and marginalisation probably the marginalisation is the big one”. There was a link made between this theme of “supportive social policy” and “big business”.

6.4.7 Ethnic Segregation or Diversity The importance of a relationship between ethnicity and maternal depression was raised early by an international expert. The individual level quantitative study identified an association with country of birth and the group level measures of ethnic segregation and diversity were associated with increased rates of depression in the ecological regression studies. Experts interviewed further identified a link between social cohesion and ethnic segregation.

199


Isolation

is a cause of is a cause of

Migrant isolat ion is property of

"Depressed" Communit y~

Segregat ion protects from

is property of protects from

is a cause of is a cause of is a cause of is a cause of is property of is property of

Mood of social environment Social Net w ork~

is property of protects from protects from

Ethnic clusters~

is associated with

is associated with

is part of

T olerance

is property of

Religous is part of segregation

bonding netw orks

Social capit al~

Racism

is a cause of is associated with

is property of is part of

is a cause of

is a cause of

Et hnic diversit y

is associated with

is associated with is property of is associated with

is property of

Multicult ural Communities

Arranged marriage

Culture

Non English Speaking

Figure 51: Focused Network of Supportive Social Policy One expert when speaking of social cohesion spoke of the divide in Bankstown between Arabic and Vietnamese groups. She spoke of further divides within the Arabic speaking community with Arabic speaking families coming from 22 nations. For her social cohesion was around “looking at where are the glue points [are] where we can glue people together to form the fabric of the community”. In contrast to the ethnic segregation that she experienced in Bankstown, the same expert spoke of integration in the suburb of Sadleir where Muslim women were participating in the multicultural group rather than their Muslim women’s group. She quoted the mothers as saying “but we want to be part of our community. If we go there we are not part of the community, we are just other Muslim women. By coming here we are part of the community. We want to know other people”. By contrast in Bankstown some ethnic groups did not want to meet with others because they didn’t like the way women from other cultures parented. “One of the key issues, why they were getting together, was around parenting, and in a sense in a negative way it was this strong sense of racial arrogance that was coming out in these young women. I was a bit blown away from having this experience with these other women to having this experience with these women who did not emphatically want to relate to other women of another race”.

200


With that different experience among the culturally diverse communities she wondered whether it was related to the size of the various ethnic populations. “What I found startling was, what we find in Bankstown is there is a huge Vietnamese community and a huge Arabic community. Is it because they are so large that they end up becoming quite separate? Where as in the other two, Liverpool and Fairfield, does the salt and peppering of the multinational nature of it mean that there is a greater degree of integration?”

Some of the field experts highlighted the factional groups within the larger ethnic populations. For example one expert spoke of the splits within the Vietnamese community with some families affiliated to communism or to the French. Others may be Chinese Vietnamese or tribal Vietnamese. “It doesn’t mean that if you are living around Vietnamese that you have your community there. Sometimes you have to be careful who [you] are linking with within the community. May be they have a language within themselves.”

The segregation may not be a problem until there is a pregnancy and infant. One expert explained it as follows “They might be self-sufficient. Working in a restaurant they are liaising with Vietnamese that is fine. But once you have a pregnancy they will have to go out to access services [and] be part of the broader society for schooling and for everything. That is why the isolation is highlighted because they don’t know the system. They might have been fully integrated into Cabramatta and this pregnancy has to bring them out of their comfort zone.”

Also highlighted was the role that Health and Community services might be playing in creating segregation. One expert felt that services were so “fully into addressing the needs of particular religions, or specific groups, that we are creating segregation because it is an easy solution for us”. She gave the example of “Like say Arabic Muslim from Bankstown wouldn’t go to Lakemba. They are different. Although they are both Muslims they wouldn’t mix”. Mothers in the focus groups also highlighted the possible role of ethnic clusters. One mother when asked why some women might get more depressed than others wondered if it may be because there are clusters of people from different cultures living in those communities. She said that she knew that depression was more prevalent in mothers of a certain ethnic group and therefore if there was a high concentration of those people in the suburb there would be more depression.

201


There appeared to be a contradiction in what would be best for mothers. One expert felt that if every Muslim women “who identify let’s say as Suni” was sent to a particular establishment then we would be creating a cluster or group that would not know anything except that. She felt it was a risk not to expose them to other groups. “So we are addressing this sense of belonging but then we are creating nuclease where this cluster of people, where they don’t know anything of the broader society”

A different expert noted that people from similar ethnic groups seek each other out. Many of those ethnic communities are a “supportive family orientated group” where the women are very supportive of each other. They may come from large families where there are many sisters, aunties and grandmothers and the women often do not work. “So in terms of parenting and having babies, for women from some of those ethnic backgrounds, the Arabic and African groups, are very supportive and very helpful.”

The concurrent group-level exploratory factor analysis and spatial exploratory data analysis suggested a possible interplay between social disadvantage, and ethnic integration or segregation. Indices of ethnic segregation were higher in those suburbs with social disadvantage. One expert had wondered about this relationship in the suburb of Miller. “I have often wondered in Miller the kind of interplay, racism is going up and people don’t think that multiculturalism makes life better. I’ve often wondered if that as an area becomes more depressed and disadvantaged and there’s less opportunities and lots of structural problems, and less connectiveness. Whether that means people will go at each more and won’t be able to support each other. Where as if you’ve got those other things in place, the ethnic diversity comes into play in a good way”.

202


6.5


In the previous section I described the concepts that emerged from initial and focused coding. As is Chapter 4 I will now use two kinds of situational analysis, situational maps and social worlds/arenas maps, to further “open up” the data. As with the previous section the situational analysis drew on findings from the literature review and quantitative analysis. In addition I brought to this part of the analysis my own observations and experience from working in the South West Sydney community. This is a methodological break from the inductive approach taken up to this point. Clarke (2005, p 85) defends this approach as follows:

“The last caveat is perhaps the most radical. Researchers should use their own experiences of doing the research as data for making these maps. There is a saying in the world of qualitative inquiry that the person doing the research is the "research instrument." I am further asserting that that instrument is to be used more fully in doing situational analyses. … Beginning even before a research topic is decided upon, we notice and store information, impressions, and images about topic areas and issues. … Part of the process of making situational maps is to try and get such information, assumptions, and so on out on the table and, if appropriate, into the maps.” It is here in the situational analysis, therefore, that I as a population health practitioner, must bring to the table my knowledge of the “situation” in South West Sydney, New South Wales and Australia,

6.5.1 Situational Analysis Mapping The messy working maps described by Clarke were used extensively throughout the analysis but are not presented here. The more structured map that (Figure 52) presents the data in terms of the situational elements such as individual and collective human actors, silent actors, discursive constructions, political elements, spatial elements and major issues and debates. Figure 53 focuses this analysis on the mother’s situation within neighbourhood, community, city, State and national arenas. As in Chapter 4, the situational analysis mapping and analysis of relationships between elements did not contribute significantly to the findings. This is not surprising as the study design had been informed by critical realism and the research map of Layden (1993). Consequently the interviews had sought to elucidate these same social structures and

203


relationships. Although most phenomena and their relationships had previously been identified, the following more abstract concepts were strengthened and drawn together: 1. Social support networks, social cohesion and social capital 2. Services planning and delivery and social policy 3. Global economy, business and media. The analysis highlighted the importance of other arenas, processes and structural elements that were discussed in Chapter 4. Findings of the analysis were then used to inform the building of the Social worlds/Arenas Map and together both analyses contributed to this Chapters overall findings.

204

INDIVIDUAL HUMAN ELEMENTS/ACTORS

NONHUMAN ELEMENTS/ACTANTS

Mother Mothers’ Partner Family members, sister, grandmother Mothers friends Midwife and child and family nurse Neighbours Birth trauma Immigrants New people Mentally ill residents “Youth” roaming the streets Criminals

Huggies advert Television Magazines and newspapers Cars, bus, train Phones Shops and Malls Call centres providing advice Information, books, web sites Physical safety of neighbourhood Social space, Community centres Community gardens “amenities” Libraries Vacant housing Banks Medical centres

COLLECTIVE HUMAN ELEMENTS/ACTORS

IMPLICATED/SILENT ACTORS/ACTANTS

Social networks Groups of mothers, Play groups Hospitals, Obstetric Services Child and Family Nursing service Non Government Organisation helpers Area Health Service Government Departments TAFE, Schools Interpreter services Church Mosque Business Clubs Football teams

Politicians Political party members Unions Employers Government, Centre link Housing, Community Services Council Land owners Television producers Advertisers Producers of products for infant care and mothers “Big business” “small business”


DISCURSIVE CONSTRUCTIONS OFINDIVIDUAL AND/OR COLLECTIVE HUMAN ACTORS

DISCURSIVE CONSTRUCTION OF NONHUMAN ACTANTS

Family dysfunction Fear Hope, purpose Cleanliness Happiness “Broke” “Battlers” “ “Bottom of pile” Isolation Nurture Survival Social hierarchy or Class, poverty Drug centre of Sydney Stress Unhealthy life-style behaviours Partner violence and drugs Not welcome Racism

Estates with new houses are good Apartment living is not good “toxic community” Unreal advertising of motherhood Crowding of some homes Depressed neighbourhoods Estates of bliss Hot houses Cold houses Mouldy houses Damaged houses Lack of sunlight

POLITICAL/ECOMONIC ELEMENTS

SOCIOCULTURAL/SYMBOLIC ELEMENTS

Safety from violence of community Marginalisation of poor Marginalisation of ethnic groups Region is relatively poor Differences between rich and poor Sydney is a divided city

Religious groups have different norms Racism Ethnic segregation Aussie Gentry of Macarthur Workers of Campbelltown

TEMPORAL ELEMENTS

SPATIAL ELEMENTS

Families NSW Programme started 2000 Integrated perinatal care started in SWS Innovative programmes started in SWS Traumatic pregnancy Previous losses Settlement history

Distanced to amenities Variation in distribution of race and religion Movement of amenities to big centres New Estates with Nice Big Homes Apartments House Neighbourhood Public housing estates

MAJOR ISSUES/DEBATES (USUALLY CONTESTED)

RELATED DISCOURCES (HISTORICAL, NARRATIVE, AND/OR VISUAL)

Muslims are not wanted Teen pregnancy is bad Pregnancy out of marriage is bad Unaffordable housing Homelessness Mums need to work to survive

Macarthur is for gentry Campbelltown and Liverpool for workers Teen pregnancy is bad Pregnancy out of marriage is bad Outsiders vulnerable Every one will own a home Every family will own a car A nice house looks like this A bad house looks like that.

OTHER KINDS OF ELEMENTS Family dysfunction Loss of expectations Loss of control Unplanned pregnancy

Figure 52: Ordered Situational Map: Maternal Depression, Community

205


6.5.2 Social Worlds and Social Arenas

Figure 53: Social worlds/Arenas Map of Mothers Home and Neighbourhood What is this map telling us? Firstly there is the domestic or home social arena drawn as the largest dotted square that was explored in Chapter 4. That arena remains central to this analysis and is made up of a few actors: mother, baby, father and others such as mothers’ sister, mother, midwife, visiting nurse, and friends. The home is the domain where mother and baby spend most of the day. It is important to again note that there may be expectations, loss of control, loneliness and poor connection to support. The immediate neighbourhood arena was also discussed in Chapter 4. It is here that physical and local social aspects of the environment may impact on mothers. There may be mothers groups, buses and services available but for some, these are absent. The neighbourhood may feel unsafe or friendly and this may affect mothers’ willingness to leave the home.

206


The Map illustrates two other possible neighbourhood arenas. One of these has a “nice” local village, library, community hall, and nice looking homes and physical environments. There is no comment about ethnicity because most people are “aussie”. The other neighbourhood arena has mixed ethnic peoples and is known as a “drug capital” with high levels of crime, distrust and perceived lack of safety. Those neighbourhoods are polarised images of suburbs in South West Sydney. The expert arena has universities that historically have been physically distant from the community although their experts worked closely with the neighbourhood arenas. The expert arena is closely overlapped with the policy and services and social cohesion arenas. These are two areas where universities have been active. Big expert hospitals are in the expert arena with maternity services closely associated. Antenatal classes are close to the hospitals while mothers groups are imbedded in local neighbourhoods. The expert voice is strong and South West Sydney has a history of providing expert leadership in innovative services and programmes. The Mall plays an important role in the lives of all who live and work in the city and community arena. The Mall is designed for cars and may not always be mother and baby friendly. The Mall is associated with “big business”, media, glamour and expectations of what might be. The Mall has replaced the Town Centre as the place to go to meet friends and “hang out”. Absent from the Community Arena are “community gardens” and league teams. There are, however, League Clubs and fast food outlets. Football teams are emerging. The community has a poor bus and train service making it difficult for mothers to get to The Mall. Important in this arena are jobs. South West Sydney jobs are vulnerable to changes in global economic fortunes. Dominant is the policy and services arena. The earlier analysis highlighted the importance of access to services and the important role that supportive government policy plays. South West Sydney has large public housing estates and relatively poor access to services. Services are not equitably distributed or delivered as evident in the visualisation of access to child and family nursing services in Chapter 7. General practitioner services are described as a “cottage” industry in many disadvantaged neighbourhoods. There are few mothers groups and most antenatal classes are held at the hospitals. Federal and State policy plays an important role in the lives of the families in South West Sydney where the term “battlers” is used to describe the daily struggle of families.

207


Sydney has been described as “A City Divided”. The reason for this is clear in Figure 12. Global and local economics play a powerful role in the lives of families. As illustrated in Chapter 4 media raise expectations for families a similar phenomenon to the impact that The Mall may have on some families. There are stack differences in wealth not only between North and South West Sydney but also within local communities. Parts of South West Sydney were settled by the Gentry of New South Wales and that history still divides the region. The power of Big Business to shape lives was illustrated in the voices of local practitioners and mothers. The loss of local League teams and community gardens are two examples. The Social Worlds/Arenas Map is traversed by the Social Cohesion Arena. Social cohesion and its related concepts of social networks and social capital have a positive influence on families and maternal depression. The voices of mothers, experts and the quantitative data explored in Chapters 5 and 7, all suggest the protective influence of support and social integration. Global forces are increasing the migration of peoples of different ethnic and cultural backgrounds to South West Sydney. The diversity of peoples is a feature of most South West Sydney suburbs. Previous quantitative studies have found that postnatal depression is associated with mothers not being found in Australia and the spatial studies in Chapter 7 indicated that rates of depression are higher in suburbs with ethnic segregation.

6.5.3 Social support networks, social cohesion and social capital Support and nurturing of mothers was identified in Chapter 4 as a strong emerging protective theme. That analysis also identified that family, services, policy and support networks all played a role in providing a nurturing environment. The earlier analysis in this Chapter confirmed the importance of community-level social support networks, social capital and ethnic integration. All of these concepts point toward the role of connectedness or social cohesion as being protective of mothers. The situational analysis confirmed these findings and suggested that current matters of ethnic segregation, diversity, integration were part of a broader idea of social cohesion. Important historical elements were identified in South West Sydney related to the phases of settlement and identity. These elements had a spatial component evident in the spatial maps in Chapter 7. Silent in the data is the historical settlement of Macarthur by land owning gentry and the role that the Nepean River played in dividing social Class. In contrast to the close knit Macarthur communities, other parts of the Region have large migrant communities and new public and private housing estates with varying degrees of

208


social cohesion and social capital. The situational analysis suggests that both economic and policy forces play an important role in shaping the level of social support available to mothers.

6.5.4 Services planning and delivery and social policy The earlier focused coding identified group-level access to services and supportive social policy as key concepts in protecting mothers from postnatal depression. The situational analysis confirmed their important role and suggests that they are linked. The division between them probably related to the hierarchical nature of policy and service delivery in the Australian context. The mothers were clear that income support, nurse phone calls, and mothers groups all contributed to supporting them when they felt “out of control” or depressed. While experts had clearly played an important role in the region their contribution was silent in the data. Drawing on my own knowledge and experience I placed the Expert Arena between influencing policy and service delivery and strengthening social cohesion. An unexpected finding during the research was the limited number of mothers groups available. This was surprising considering the potential that they have as both being protective and supporting mothers with depression and other difficulties. The answer may lie in the nature of current NSW government policy again pointing to the important role that services planning and social policy play in influencing the situation for mothers vulnerable to depression.

6.5.5 Global economy, business and media. Finally it is clear from the data that the influence of the global economy, big business and media reaches through all levels of the Social Arenas to impact significantly on the situation for mothers and infants. Related findings in Chapter 4 include: “being broke”, economic marginalisation, occupational class, absence of power and control and the generation of expectations and dreams. Strong global economic forces are also responsible for movement of migrants and refugees to the region of South West Sydney, Sydney, Australia. Some are economic migrants while others are from past and present conflicts. Local jobs are vulnerable to global economic fortunes and the landscape of local communities are influenced as much by the planning decisions of Big Business as they are by planners and policy makers. Thus local communities have peoples who once fought each other living side by side.

209


The “Huggies” comment by one mother drew attention to the important role played by media and advertising in the ordinary lives of mothers. The role that media might play in raising the expectations of mothers was previously discussed. Related are the pressures put on families by images of large five bedroom homes (McMansions), four by four wagons to drop the kids off at school, and the glitter of advertising and merchandise in “The Mall”. The earlier analysis also included the comment on the impact to the local community of losing the local League team as a result of a media related franchise agreement.

6.6

Theory Generation

The situational analysis was the third stage of data analysis in this chapter. The focus was on possible situations experienced by mothers with newborn infants. The theoretical concepts of: social support networks, social cohesion and social capital; services and social policy; and global economy, business and media were briefly explored. The previously identified, and related, concepts of poverty, class and “depressed community” remain significant for the process of theory building. “Depressed community” in particular is a strong contender as a proximal cause of depression. The following conceptual network illustrates how the concepts identified in this Chapter might create the contextual conditions necessary for perinatal depression to occur or be maintained.

210


"Depressed" Community~

is cause of

Global Economy Business Media

is cause of

is cause of

is cause of

Services & Social Policy

protects protects from from

Economic Social Marginalisation

is cause of

protects from

Social Cohesion and Capital

is cause of

Stress~

Figure 54: Emerging Conceptual Framework of Maternal Stress

Figure 55: Conceptual Model of Mothers Mood The above models will be taken forward to Chapter 8 where I will explore and undertake further theory construction toward a mid-level Theory of Maternal Depression, Stress and

Context.

211


6.7

Conclusion

The purpose of this Chapter was to describe the findings of the qualitative exploration of group level factors that might be associated with the phenomenon of perinatal depression, perinatal adversity and their possible impact on the developmental and health outcomes for the foetus and infant. The qualitative study was undertaken concurrently with the literature review and exploration of quantitative data. The key phenomenon identified included: community-level social support networks, social capital, “depressed community”, Big Business, supportive social policy, access to services at the group level, and ethnic segregation and diversity. Situational analysis refined these to: social support networks, social cohesion and social capital; services and social policy; and global economy, business and media. The following Chapter will report on the exploration of group level quantitative data using spatial and multilevel exploratory data analysis and regression modelling methods.

212

Chapter 7: Ecological Quantitative

Chapter 7: Exploratory Data Analysis of Group Level Ecological Factors Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







213


7.1

Introduction

The purpose of this Chapter is to describe the quantitative exploration of group level ecological factors that may influence the phenomenon of maternal depression during the perinatal period. As previously, the study has been informed by the concurrent qualitative studies and the findings of the individual level analysis. As noted in the introduction to Chapter 5, the critical realist and emergent nature of this study requires the methods of exploratory data analysis (EDA), exploratory factor analysis (EFA) and exploratory spatial data analysis (ESDA) be used in a different manner from that commonly used in epidemiological and confirmatory analytical studies. The findings will be used principally to generate explanatory theory. The critical realist nature of this study does not limit the use of regression or Bayesian methods, as used in this Chapter, to generate theory but as with other post-positivist traditions it does require that the findings are treated as tentative and fallible. I have deliberately delimited the scope of the analysis in this Chapter to exclude multilevel factor analysis, path analysis, structural modelling and multilevel analysis of cross-level mediation and moderation. This was assessed as being beyond the scope of the current study but as discussed elsewhere places limits on explanatory inference.

214


7.2

Aims

The aims of this chapter are to: 1. Identify group-level factors that may be associated with increased postnatal depressive symptoms 2. Explore a number of ecological and multilevel models to determine the most parsimonious set of group level predictors. 3. Use the findings to inform emerging theory. The specific objectives are to: 1. Identify a set of group level variables that could be associated with individual and aggregated maternal depressive symptoms 2. Visualise and explore the distribution of the aggregated maternal depressive symptoms 3. Visualise and explore the distribution of identified group level variables 4. Identify groupings of variables that may be associated with latent vectors 5. Use both frequentist and Bayesian statistical methods to explore the ecological relationship between the group level variables and aggregated rates of maternal depressive symptoms 6. Explore a number of nested multilevel spatial models to identify possible group level predictors of individual level maternal depressive symptoms. The specific analysis questions are: 1. Does the distribution of aggregated depressive symptoms and identified covariates suggest possible associations that might be further examined? 2. What group level factors might have a causative role in the phenomenon of maternal depression and depressive symptoms? 3. How are the factors related to each other and, therefore, what latent, unmeasured factors might be important for further study?

215


7.3

Methods - Overview

7.3.1 Theoretical Approaches As noted earlier recent commentators have argued that epidemiology and social epidemiology in particular require a stronger theoretical approach (Kaplan 2004, p 127). Kaplan further stated that “perhaps nowhere is the need for social epidemiologic theory more apparent, than in the study of “place” effects on health”. One of the main problems in the study of neighbourhoods and health is the lack of development of theories about plausible social, psychological, and biological links between specific features of the neighbourhoods and specific health outcomes (Curtis and Rees Jones, 1998). Rajaratnam and colleagues (2006, p 547) in their review of 31 maternal and child health relevant articles note that “although most authors provided theoretical explanation of their choice to examine broad neighbourhood constructs, few were explicit about why certain indicators were selected to measure these constructs”. Diex Roux (2003) further notes that the collection of new data on theoretically defined areas is complex and may be impractical over a broad range of areas. Strategies that combine the use of existing standardised data on a broad range of areas with new data collection on a subset of areas could be an alternative. Diex Roux argues that qualitative research may contribute to the development of hypotheses that can be tested in large, quantitative datasets and may be of help in understanding the results of quantitative studies. “The integration of quantitative and qualitative approaches would constitute a major innovation in epidemiology generally” (Diez Roux 2003). Exploratory Data Analysis (EDA) as used in this Chapter is a quantitative inductive or emergent approach to the development of theory from empirical data. As discussed earlier EDA is in the post positivist tradition and shares the interaction of prior knowledge and data analysis. I have allowed my prior knowledge and the findings of the concurrent qualitative and quantitative studies to influence the selection of group-level variables for this ecological study from within existing large data sets. I have sought to include variables that might be indicators of concepts emerging from interviews, focus groups and the individual level quantitative studies. Ecological studies focus on comparison of groups rather than individuals and have been used by social scientists and epidemiologists for more than a century often as simple “descriptive” [or EDA] analyses (Morgenstern 2000). The limitations of ecological studies are generally attributed to the well established ecological bias, or fallacy, inherent in

216


making inferences regarding individual-level associations based on group-level data (Morgensten 1982; Piantadosi, Byar et al. 1988; Greenland 1992). The converse and equally important fallacy is the ‘atomistic fallacy’ that can occur by focusing exclusively at the individual level and thus missing the context in which the individual action occurs (Blakely and Subramanian 2006). Ecological studies make use of maps and spatial statistical tools to explore and analyse aggregate and environmental data. Such exploratory spatial data analysis (ESDA) has a strong tradition (Haining 2003), and will be used extensively in this Chapter to visualise, explore and analyse group level variables that may influence perinatal depression. Multilevel analysis allows for the simultaneous examination of the effects of group level and individual level variables on individual level outcomes (Duncan, Jones et al. 1995). This approach helps to clarify the sources and magnitude of ecologic and cross-level bias, and allows for the separation of biological, contextual, and ecological effects (Morgenstern 1998). I will use multilevel analysis in this Chapter to explore whether the observed ecological effects from the ESDA remain after controlling for possible confounders measured at the individual level, thus identifying the compositional or contextual nature of the ecological effects. Subsequent theory building will draw on the inductive inference of these findings. There have been methodological difficulties in drawing causal inferences from multi-level studies of principally related to compositional and endogeneity issues and cross-level mediation (Pickett and Pearl 2001; Subramanian, Lochner et al. 2003; Oakes 2004; Subramanian 2004; Diez Roux 2008). The delimitations that I have imposed on the analysis in this study will thus limit causal inference. The implications of these will be discussed throughout the Chapter and again in Chapter 10. I have used Bayesian methods extensively in this Chapter for both spatial and multilevel exploratory data analysis. The use of Bayesian methods has been useful both for philosophical and technical reasons. Bayesianism is a philosophy of inference that, like Poppers refutationism, holds an objective view of scientific truth and a view of knowledge as tentative or uncertain. Rothman and Greenland (1998, p 22) cite Greenland (Greenland 1998) and argue that “Bayesian philosophy provides a methodology for sound reasoning and, in particular, provides many warnings against being overly certain about one’s conclusions. The use here of Bayesian approaches is consistent with critical realism. In particular Lipton (2004, p120) observes that “Bayesianism poses no threat to

217


Inference to the Best Explanation” the principal mode of explanatory inference used in this study. Central to Bayesian inference is the representation of uncertainty about parameters or hypotheses as probabilities. Congdon (2006) notes that under this framework we can calculate the probability that a parameter lies in a given interval (similar to confidence intervals) and that prior knowledge about a parameter, such as a relative risk, are important aspects of the inference process. Technically I have used the Bayesian statistical software WinBUGS because of its flexibility to handle random effects models such as those used for spatial and multilevel analysis. Browne and Draper (2006) have suggested a hybrid modelling strategy, in which likelihood-based (or frequentist) methods are used in the model exploration phase and Bayesian diffuse-prior methods are used for reporting of final inferential results (Draper 2008). In this Chapter I will use likelihood methods for some data exploration but where spatial conditional autocorrelation (CAR) models are required I have used Bayesian WinBUGS software. Further consideration will be given to theoretical matters in Chapter 8 – Theory Construction. Methodological matters related to the selection of variables for study and methods of analysis will be covered in the following sections as appropriate and limitations of this study will be discussed in Chapter 10.

218


7.3.2 Sample and Design 7.3.2.1 Study Design The study is an ecological study of aggregated rates of postnatal depressive symptoms in South Western Sydney Area Health Service from 2002-2003. The exploratory data analysis will include descriptive analysis, cluster analysis, factor analysis, mapping, standard linear regression, and Bayesian spatial and multi-level analysis. For this exploratory phase of the study I have elected to use the 2001 Census data for the majority of the group level candidate variables. NSW crime statistics and aggregated individual level data are also used. Candidate group level variables have been selected based on the findings of the previous quantitative and qualitative studies. Factor analysis of the candidate group variables will be undertaken to identify possible latent factors. The factor analysis will be used to inform both the inductive exploratory data analysis and the later abductive theory building as proposed by Haig (2005). The ecological analysis will also utilise both spatial and multilevel analysis thus addressing the limitations of earlier ecological study designs that used principally standard linear or logistic regression methods. 7.3.2.2 Study Setting The study setting for the ecological study is different to that of the individual level exploratory data analysis which was undertaken on all births in the former south west Sydney. As discussed below I have elected to undertake group level analysis at the suburb level. The ABS 2001 Census was only reported for urban suburbs in parts of the local government areas of Bankstown, Fairfield, Liverpool, Campbelltown and Camden (Figures 52). A subset of individual level data was extracted from those selected suburbs (n=14622) for use in the later multilevel analysis.

219


Figure 56: South West Sydney Local Government Areas and Suburbs - 2001 Census

220


7.3.2.3 Group level unit of study The individual-level data available for study was coded by suburb of residence. Individuallevel information was not available at the ABS Collection District level. The suburb of residence was chosen as the closest group-level administrative unit to naturally occurring local neighbourhood environments. The limitations of using administratively defined areal units are well described (Diez Roux 2001; O'Campo 2003; Diez Roux 2004). The family unit might have been considered as a second-level group population. The data set does not, however, contain sufficient information on other family members other than the mother and infant to make this a useful analytical unit. I have, therefore, treated the mother, infant and family as individual-level data for this study. The ABS Postal Areas are a close approximation of Post Codes. Those Post Code administrative units do not, however, represent natural local communities as they may be experienced by residents. There are seven local government administrative areas and 11 Statistical Local Areas. Those administrative areas might have been considered for inclusion but the resulting multi-level model would have lacked sufficient statistical power. For this exploratory data analysis study I have, therefore elected to limit analysis to a twolevel multi-level model using suburb group level variables derived from the ABS 2001 Census, NSW Crime Data, and aggregation of selected individual-level variables.

221


7.3.2.4 Study Suburbs There were 101 suburbs available to study using the 2001 Census maps. The following (Table 22) are the population, area, population density and number of women of child bearing age for each suburb (ABS Census 2001).

Suburb Name Abbotsbury

Pop.

Area Sq Km

Density Pop/Sq Km

WCBA

3861

1.32

2933

948

Airds Ambarvale Ashcroft Bankstown Bass Hill Birrong Blairmount Bonnyrigg Bonnyrigg Height

4113 7569 3273 25507 8642 2877 562 8872 7107

2.63 2.86 1.18 6.34 4.84 1.43 3.20 3.13 2.22

1561 2643 2780 4023 1786 2005 176 2837 3198

979 1868 679 6097 1777 574 71 2200 1788

Bossley Park

14444

4.46

3240

3414

Bow Bowing Bradbury Busby Cabramatta Cabramatta West Camden

1637 9092 4062 19487 6538 3090

2.31 7.14 1.23 4.34 1.87 4.36

709 1274 3297 4491 3503 708

416 1998 886 4599 1468 568

Camden South Campbelltown Canley Heights Canley Vale Carramar

4668 9991 9717 10135 3128

4.12 8.01 2.60 3.10 1.04

1134 1247 3739 3266 3016

1014 2225 2238 2243 659

Cartwright Casula Cecil Hills Chester Hill Chipping Norton Claymore Condell Park Currans Hill Eagle Vale

2264 11254 5829 9993 9027 3579 9013 4415 5924

0.93 6.37 2.89 4.39 7.55 1.48 4.67 5.36 2.26

2425 1768 2016 2278 1195 2420 1928 824 2619

442 2646 1525 1995 2064 892 1906 1260 1486

Suburb Name Holsworthy Horningsea Park Hoxton Park Ingleburn Kearns Lansvale Leumeah Liverpool Long Point Lurnea Macquarie Fields Macquarie Links Miller Milperra Minto Moorebank Mount Annan Mount Pritchard Narellan Narellan Vale Old Guildford Padstow Padstow Heights Panania Picnic Point Prairiewood Prestons Punchbowl Raby Regents Park Revesby

Area Sq Km

Density Pop/Sq Km

2922

7.74

377

749

2660 3367 14328 2895 2444 9081 20596 357 8026

2.31 1.95 14.66 2.82 2.96 3.89 6.37 2.12 3.30

1150 1726 978 1028 826 2335 3232 168 2434

755 919 3280 717 498 2003 4855 86 1754

13755

6.87

2001

3205

328 3161 3947 11334 5987 6667

2.25 1.25 5.62 9.06 12.83 7.16

146 2536 702 1250 467 932

82 683 830 2697 1249 1711

8977 3653 6980 2179 12162

3.52 4.14 3.65 0.96 5.37

2550 883 1913 2266 2266

1966 851 1839 481 2626

3420 10853 5218 3317 8509 17197 6384 3856 11233

1.96 4.58 3.75 2.14 8.44 4.94 2.74 1.55 5.34

1746 2369 1391 1552 1008 3483 2328 2494 2104

671 2217 1060 724 2275 3925 1636 780 2273

Pop.

Source: ABS 2001 Census. W-CBA: Women of Child Bearing Age (15-45 yrs)

Table 22: Characteristics of Study Suburbs

222

WCBA


Suburb Name

Pop.

East Hills Edensor Park Elderslie Eschol Park Fairfield Fairfield East Fairfield Height Fairfield West Georges Hall Glen Alpine Glenfield Green Valley Chullora Greenacre Mount Lewis Greenfield Park Hammondville Harrington Park Heckenberg Hinchinbrook

2551 9338 2647 2813 14648 2332 5477 10360 7966 4193 7012 11562 461 20716 1100 5386 3458 3027 2964 9975

Area Sq Km 1.41 2.96 4.22 3.41 4.42 1.49 1.47 3.23 6.41 3.23 7.29 3.22 2.81 6.35 0.35 1.66 1.92 2.78 0.93 3.52

Density Pop/Sq Km 1805 3156 628 825 3316 1566 3714 3212 1242 1298 961 3588 164 3263 3114 3254 1804 1091 3197 2837

WCBA 527 2343 546 711 3274 498 1181 2301 1697 992 1556 3032 94 4441 231 1313 779 791 649 2582

Suburb Name Revesby Heights Rosemeadow Ruse Sadleir Sefton Smithfield St Andrews St Helens Park St Johns Park Villawood Voyager Point Wakeley Warwick Farm Wattle Grove West Hoxton Wetherill Park Woodbine Yagoona Yennora

Pop.

Area Sq Km

1272 7952 5820 2923 5137 10382 6287 5769 5774 6831 767 4887 4613 9635 4745 6407 2729 14569 1286

0.73 2.83 2.67 0.92 1.83 4.56 2.62 5.56 1.83 3.29 2.47 1.68 4.62 2.48 5.78 11.28 1.55 5.59 0.76

Density Pop/Sq Km 1732 2809 2182 3168 2806 2278 2395 1037 3156 2078 310 2908 999 3893 821 568 1760 2606 1701

WCBA 264 2085 1261 609 1077 2227 1572 1519 1273 1379 213 1188 1070 2589 1285 1472 643 3032 235

Source: ABS 2001 Census. W-CBA: Women of Child Bearing Age (15-45 yrs)

Table 23: Characteristics of Study Suburbs (Cont.) The suburb distribution of mothers born in Australia and not born in Australia is shown in Figure 57

Source: IBIS Data 2002-2003

Figure 57: Birth place of mothers

223


Figure 58: South West Sydney 2001 Census Suburbs The spatial distribution and names of the 2001 Census Suburbs used in this study are shown in Figure 58.

224


7.3.3 Missing Data and Zero Cells Analysis of missing EDS data was discussed previously in Chapter 5. At the 2001 Census there were no suburbs with missing population or covariate data. Similarly there were no zero cells for either the Census or IBIS aggregated covariate variables.

7.3.4 Outcome (Dependent) Variables The Postnatal Edinburgh Depression Scale was discussed previously in Chapter 5 and is at Appendix C. The aggregated outcome variables available for study was the percentage EDS > 9 (deleted missing data), EDS > 9 (imputed missing data), EDS > 12 (deleted missing data) and EDS > 12 (imputed missing data). The percentage was calculated by dividing the observed cases by the number of cases surveyed in that suburb. The expected cases were also calculated based on the total observed cases in all 100 suburbs divided by the total surveyed cases in all 100 suburbs.

EDS9P N

Valid Missing

Mean Std. Error of Mean Median Std. Deviation Variance

EDS12P

EDS9MP

EDS12MP

101

101

101

101

0

0

0

0

16.221

7.119

14.163

5.692

.628

.365

.704

.373

16.666

6.929

14.208

5.298

6.281

3.654

7.048

3.732

39.461

13.353

49.682

13.930

Skewness

.046

.213

.404

1.321

Std. Error of Skewness

.241

.241

.241

.241

1.666

.333

.203

3.521

.478

.478

.478

.478

Kurtosis Std. Error of Kurtosis Range

36.956

16.666

33.333

21.810

Minimum

.000

.000

.000

.000

Maximum

36.956

16.666

33.333

21.810

Table 24: Descriptive Statistics for the Group-level Outcome Variables I will report in this Chapter only on the analysis of EDS > 12 (deleted missing data). Other analyses are reported in the Appendices and had similar findings.

225


7.3.5 Comparison (Independent) Variables 7.3.5.1 Selection of Group Level Variables I sought here to include variables that might be indicators of the concepts emerging from the interviews and focus groups in Chapters 4 and 6 and from consideration of the individual level latent factors emerging from the exploratory factor analysis in Chapter 5. A constraint imposed by the emergent study design (see Section 2.8) was that extant literature on group level theoretical concepts was not able to be used. The domains emerging at the stage when group level variables were selected were: 

Social Networks, Capital or Cohesion



“Depressed” or disadvantaged Community



Access to Services



Ethnic Segregation or Integration



Big Business



Social Policy



Global economy



Media

I assessed that four of these might be measured at the suburb level, namely: social networks, capital and cohesion; “depressed” or disadvantaged community, access to services and ethnic segregation or integration. At the suburb level there was an extensive range of possible variables in the 2001 Census dataset and the derived SEIFA indexes. It was not possible, however, to include all possible variables or to identify candidate variables for all emerging explanatory concepts and latent factors. For some emerging concepts, such as social cohesion and social capital, there were no candidate variables in the Census data. Attempts to secure relevant data from the NSW Health Survey were unsuccessful. I therefore used aggregated variables (social networks and “no regret leaving suburb”), while acknowledging the possibility of same-source bias identified by Radenbush and others (Duncan and Raudenbush 1999; Radenbush and Sampson 1999) cited by O’Campo (2003). I deliberately did not review theoretical literature related to the selection of variables as this would not have been consistent with the emergent approach. Methodological approaches to classifying and measuring ecological variables have been reviewed by Blakely and colleagues (Blakely and Woodward 2000; Blakely and Subramanian 2006). Of consideration here is the distinction between derived and integral variables. Almost all the variables used in this study were derived or aggregate variables. They were either derived from the 2001 Census or from the IBIS survey data. Integral variables are

226


attributes of groups or places that are not reducible to the individual level. Examples used here are: crime rates, population density and dwelling related measures. The multicultural nature of the population of South West Sydney and themes emerging from the qualitative studies necessitated the selection of measures of ethnic diversity, integration or segregation. Three measures of ethnic diversity were selected based on the advice given to the US Bureau of the Census by Galster (2004). Galster argued that measurement of diverse neighbourhoods is conceptually distinct from the measurement of segregation. Measures of diversity measure differentiation of population within the areal unit (Census tract) instead of across them. In addition they are able to measure multiple groups in contrast to many measures of segregation that are usually designed for only two groups (usually segregation of blacks and whites). A full discussion is beyond the scope of this section. Those recommended were: Diversity Index (Maly 2000), Entropy Index (Modarres 2004), and the Simpson’s Index (Simpson 1949). The interviews also drew attention to the possible importance of extremes of wealth and poverty within areal units. I used here, the Index of Extremes (ICE) as proposed by Massey (2001) and cited by Casciano and Massey (2008). These four measures are described in more detail at Appendix F. Table 25 summarises the ecological variables selected for analysis including the emerging domain it was selected for and whether it is derived or integral.

227


Variable Name

Description

Concept Selected for

Type/ Source

Density

Total population per square kilometre (2001 Census)

Depressed Community

Integral

Entropy Index

Entropy Index (2001 Census, see Appendix F)

Ethnic

Derived

Entropy Log Index

Normalised Log of Entropy Index (See Appendix F)

Ethnic

Derived

Simpson Index

Simpson Index (See Appendix F)

Ethnic

Derived

Maly Index

Maly Index (See Appendix F)

Ethnic

Derived

Social Capital

Derived

Social Capital

Derived

Access

Derived

Access

Derived

Access

Derived

Depressed Community

Derived

Depressed Community

Derived

Depressed Community

Derived

Volunteerism No Volunteerism Universal Nurse Visit Rate Universal Home Visit Percent Nurse visit rate Poor Families

Rich Families

ICE

Smoking

Percent of mothers smoking at first home visit (IBIS 2002-2003)

Unplanned Pregnancy

Percent of mothers who planned the pregnancy (IBIS 2002-2003

Public Housing

Percent of families in public housing (IBIS 20022003)

Breastfeeding

Percent of mothers breast feeding (IBIS 20022003)

No Social Support Social Support No Practical Support Practical Support No Emotional Support Emotional Support No Regret Leaving Regret Leaving Poor Health

228

Voluntary work by total population > 15 years (2006) No Voluntary work by total population > 15 years (2006 Census) Average number of first nurse visit, 2002-2003, by infants resident, 2001 Census Average number of first nurse visit, 2002-2003, by infants resident, 2001 Census Average number of all nurse visits, 2002-2003, by infants resident, 2001 Census 50 percent of NSW median family weekly income, 2001 Census. Number of families with weekly income $0-$499 per week by total families Percentage of with income greater than the NSW median family income, 2001 Census. Number of families with weekly income greater than $800$999 per week Number of rich families minus number of poor families divided by the total families per suburb (2001 Census)

Percent of mothers with no social support network (IBIS 2002-2003) Percent of mothers with social support network (IBIS 2002-2003) Percent of mothers with no practical support network (IBIS 2002-2003) Percent of mothers with practical support network (IBIS 2002-2003) Percent of mothers with no emotional support network (IBIS 2002-2003) Percent of mothers with emotional network (IBIS 2002-2003) Percentage of mothers who would not regret leaving the suburb Percent of mothers who would regret leaving the suburb (IBIS 2002-2003) Percent of mothers reporting poor health (IBIS 2002-2003)

Suburb norm of health behaviour Suburb norm of health behaviours

Derived

Derived Derived

Suburb norm of health behaviours

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Outcome

Derived


Variable Name Good Health Owned Dwelling

Description Percent of mothers reporting good health (IBIS 2002-2003) Percent of owner occupied dwellings (2001 Census)

Rented Dwellings

Percent of rented dwellings (2001 Census)

Unemployment

Percent unemployment (2001 Census)

Different address last 5 years Same Address last 5 years Different Address last 1 year

Percent of families who lived at a different address 5 years previous(2001 Census) Percent of families who lived at the same address 5 years previous (2001 Census) Percent of families who lived at a different address 1 years previous (2001 Census)

Vacancy Rate

Percent of vacant dwellings (2001 Census)

Apartments

Percent of dwellings as apartments (2001 Census)

Single Houses High Apartments Class 3 Class 2

Percent of dwellings as single dwelling houses (2001 Census) Percent of dwellings as multistorey apartments (2001 Census) Blue collar working occupation class rate (2001 Census) Administration and retail occupation class rate (2001 Census)

Class 1

White collar occupation class rate(2001 Census)

Occupational Extremes

Index of Extremes of occupation class(2001 Census)

One Parent Families

Percent of one parent families(2001 Census)

Low Schooling IRSD IRSD Decile Violent Crime

Percent of individuals with no schooling or less than year 9 (2001 Census0 Index of Relative Social Disadvantage(2001 Census) Decile of Index of Relative Social Disadvantage (2001 Census) Violent Crime Rate. Reported violent crime rate at post code level applied to relevant suburbs (NSW Crime Bureau)

Concept Selected for

Type/ Source

Outcome

Derived

Depressed Community Depressed Community Depressed Community

Integral Integral Derived

Social Capital

Derived

Social Capital

Derived

Social Capital

Derived

Depressed Community Depressed Community Depressed Community Depressed Community Depressed Community Nil Depressed Community Depressed Community Depressed Community Depressed Community Depressed Community Depressed Community Depressed Community

Derived Integral Integral Integral Derived Derived Derived Derived Derived Derived Derived Derived Integral

Table 25: Independent variables used in the group level analysis

229


7.3.5.2 Univariate Analysis and Transformation Descriptive statistics and histograms were examined for each variable. Attention was given to missing variables, outliers, and distribution statistics. As all of the group level variables were either percent, rates or scale none of the group level variables required recoding, transformation or assigning of dummy variables. The variables were, however, standardized by subtracting the mean and dividing by the standard deviation for each variable. This transformation eliminates the bias introduced by the differences in the scales of the several attributes or variables used in the analysis (Hair, Anderson et al. 1998). Standardization of variables is also recommended for both multilevel modelling and Bayesian analysis using WinBUGS software (Spiegelhalter, Thomas et al. 2003).

7.3.6 Spatial Visualisation - Mapping 7.3.6.1 Outcome Variable Mapping

Mapping of Standardized Rates The most basic exploratory spatial data analysis method is the visualisation of spatial data. The simplest mapping form is the depiction of disease rates at specific locations. Standardised Ratios (such as standardised mortality ratios) are commonly used to map disease distribution. This likelihood approach compares the observed cases with the “expected” within small geographical areas. The Standardised Ratio is an estimate of the relative risk within each small area (i.e. the ratio describes the odds of being in the disease group rather than the background group) (Lawson, Browne et al. 2003, p 4). I calculated standardised ratios of the observed EDS counts, (O(i)) divided by the expected count (E(i)). The expected count (E(i)) for each suburb was computed using the observed EDS ratios for the SWS region (i.e. total EDS>9 and total EDS>12 divided by total women surveyed) multiplied by the number of mothers surveyed in each suburb. The number of women surveyed in each suburb may not have represented the true distribution of women who might have been surveyed. I therefore also undertook standardisation using “women of child bearing age”. The rates standardised by numbers surveyed was highly correlated with the standardisation by WCBA. I have, therefore, used rates based on numbers of women surveyed in all analyses.

230


Hierarchical Bayesian Smoothing of Rates The crude maps developed using the likelihood method described above often feature large outlying relative risks in areas where the population is small. Hence those maps usually show high uncertainty and the standardised rates can vary greatly (Johnson 2004). They also fail to catch similarity of relative risks in nearby or adjacent areas. To address this, different methods of smoothing have been developed all based on the phenomenon that observations close together in space are more likely to share similar properties than those that are far apart. Bayesian hierarchical approaches to spatial autocorrelation are currently used as a standard in epidemiology and environmental health literature (Zhu, Gorman et al. 2006). Unlike the conventional statistical inference which derives the average estimates of parameters, hierarchical Bayesian modelling produces parameter estimates for each individual analysis unit by borrowing information from all analysis units. This is the Bayesian "borrowing of strength" effect. In a standard Poisson model the variance is required to be equal to the mean but many Poisson models have more variances and are called over-dispersed Poisson models. Hierarchical Bayesian modelling identifies these "extra variances” as either spatially-correlated effects or heterogeneity effects and is able to achieve spatial smoothing by borrowing information from all individuals. Bayesian Hierarchical Methods are described in more detail in Section 7.3.10. Bayesian software package WinBUGS Version 1.4 (Spiegelhalter, Thomas et al. 2003) computes a set of posterior means for the relative risks (RR) given the observed EDS counts which are then used to create maps (using the included GeoBUGS software) to visualize the high- or low-risk suburbs. I used WinBUGS and GeoBUGS to produce smoothed estimates of suburb disease relative risks for EDS>9 and EDS>12 (Figure 64). The related WinBUGS code is shown in Figure 59.

231


# Spatial model, smooth EDS9 model { for (i in 1:N) { EDS12O[i] ~ dpois(mu[i]); log(mu[i]) 12 cases (Kulldorff 1997; Kulldorff, Feuer et al. 1997). In this study, the SaTScan software was applied to the 101 South West Sydney neighbourhoods in order to generate possible clusters of EDS>9 and EDS>12. The

232


calculation required coordinates for cases and population. The use of suburb [county] centroid locations as surrogates for case and population locations has been successfully demonstrated (Kulldorff 1997; Kulldorff, Feuer et al. 1997). For this analysis the centroids were calculated by ArcView 9.3 from the ABS 2001 Census suburb boundary files. No covariates were applied. One iteration was undertaken using the default setting which maximizes the cluster size at 50 percent of the total study population. The Monte Carlo simulation used to test significance was set at 999 iterations. The software was set to generate both high and low clusters. Only statistically significant clusters were retained for analysis. Non-cluster areas were aggregated together into one cluster area and assumed to have a relative risk of 1.0. Clusters were mapped using Arc-View 9.3 (Environmental Systems Research Institute 2008) in order to identify their physical location. 7.3.6.3 Comparison Variable Mapping Visualisation also included the mapping of the spatial distribution of the candidate comparison variables using ArcView 9.3. A chloropleth map is a map that uses shades of colour (or greyscale) to classify values into a few broad classes in a similar manner to a histogram for non spatial data. Chloropleth maps are visually appealing and thus commonly used. They are, however, a crude summary of the data and can be readily altered by manipulating the class cut-offs (Banerjee, Carlin et al. 2004). Visual inspection of the chloropleth maps is, nevertheless, an important step in exploratory (spatial) data analysis. Visual clusters and increasing (or decreasing) spatial trends may be observed if present. Manipulation of the class cut-offs (display manipulation) is an important analytical tool in EDA of spatial data and is examined in detail by Andrienko and Andrienko (2006). I manipulated the cut-offs as part of my visual exploration of the distribution of both outcome and comparison variables. In Section 7.4.2.3 I have presented the chloropleth maps using the preset Quantile classifications in ArcView 9.3 unless otherwise stated in the map legend. The chloropleth maps presented are for the raw rates with no Bayesian smoothing (unless stated). Chloropleth maps are also used in subsequent spatial results sections where a Spatial Model is chosen for detailed reporting.

233


7.3.7 Bivariate and Correlation Analyses The group-level outcome and comparison variables were individually assessed using descriptive statistics and histograms. All had normal distributions. Bivariate analysis (or univariate regression) of the independent variables against the aggregated outcome variables was undertaken using likelihood-based OLS linear regression in SPSS. The analysis was undertaken using both standardized and non-standardized variables. Significant results of the bivariate linear regression are presented at Appendix G. The presence of multicollinearity was assessed using a correlation matrix. This correlation matrix was also useful in informing the selection of variables for inclusion in the group-level studies. The problem with a correlation matrix is that it assesses only for the relationship among two variables without adjustment for the other variables. For this reason the presence of multicollinearity was also assessed using the correlation matrix for parameter estimates obtained as a product of the linear regression analysis and during the diagnostic phases of the exploratory factor analysis and later Bayesian studies. The Correlation Matrix is at Appendix G.

7.3.8 Exploratory Factor Analysis As previously discussed in Section 5.3.5 factor analysis is a statistical tool for analysing scores on large numbers of variables to determine whether there are any identifiable dimensions that can be used to describe the variables understudy (Munro 2001). I have used exploratory factor analysis in this study principally as a tool to aid in the building of theory. As previously noted in a truly exploratory approach, a researcher uses factor analysis to discover a structure that can be meaningfully interpreted or used for abductive reasoning (Munro 2001, p307; Haig 2005). For theoretical EFA where there is expected to be correlation between factors the most appropriate method of extraction is “common factor analysis” with oblique rotation. I therefore used Principal-axis factoring with the oblimin method of oblique rotation as available in SPSS. I also checked the outcomes using Principal Component Analysis (PCA) and orthogonal rotation. The results were the same. To avoid repetition, further details of the assumptions and diagnostics used are reported in Appendix K. A second purpose of EFA is data reduction where the factor analysis is used to develop composite variables representing of the latent variables. Another approach is to use the highest factor loadings as surrogates for the latent variable or to use all variables. The EFA was used for all three purposes in this study.

234


There is debate regarding the best approach for deductive theory testing studies. It has been argued that instead of using EFA to get composite indices of underlying dimensions of suburb characteristics it is best to use the measures of single suburb characteristics because of the greater ease of interpreting the results and consequently identifying policy implications. The individual measures may, on the other hand, be distal indicators of processes that will require complex policy interventions. There is also likely to be high inter-correlations among individual suburb characteristics making interpretation of analyses that employ single neighbourhood variables difficult and possibly misleading. The benefit of inductive theory building using EDA is that all approaches can be used. I chose initially to examine each individual variable and used the factor loadings to inform that model selection. The initial findings of the multilevel models were, however, difficult to interpret and I subsequently elected to undertake selected analysis using EFA composite variables. The EFA variables were saved using both the Anderson-Rubin and standard regression methods. The Anderson-Rubin Method is a method of estimating factor score coefficients which ensures orthongonality of the estimated factors. The scores that are produced have a mean of 0, a standard deviation of 1, and are uncorrelated. The Regression Method scores that are produced have a mean of 0 and a variance equal to the squared multiple correlations between the estimated factor scores and the true factor values. The scores may be correlated even when factors are orthogonal (Hair, Anderson et al. 1998, p115; Osborne, Costello et al. 2008). The Anderson-Rubin factor scores were used in the spatial and multilevel studies reported in this Chapter.

7.3.9 Likelihood OLS Linear Regression Ecological studies have previously been undertaken using likelihood statistical methods including linear regression and logistic regression methods. In those studies the ecological study is carried out on geo-referenced data but the focus is not usually on the spatial distribution but on the linkage between the dependent variable and the measured covariates (Lawson, Browne et al. 2003). The disadvantage of these approaches is that they do not take into account the spatial nature of the data and are subject to ecological fallacy if the findings are used to make inference at the individual level. These shortcomings are addressed in the later Bayesian hierarchical spatial and multilevel analyses. Likelihood methods remain useful for exploring associations among variables in ecological or group-level studies provided their short comings are recognised. As part of the EDA I

235


examined the relationship between the variables using standard likelihood methods. In this ecological study the outcome variables are aggregated variables and have normal distributions. Consequently standard linear regression and exploratory factor analysis are used. As discussed above univariate descriptive statistics and histograms were examined on all outcome and comparison (independent) variables. There was no missing data at the ecological level. Missing data at the individual level was discussed in Section 5.3.3. Bivariate analysis of the independent variables against the outcome variables was undertaken. All correlations were statistically significant but with varying r values (see Appendix G). The relationships were linear. The correlation matrix was examined and highly correlated variables were not analysed together. All candidate variables were statistically significant in bivariate analysis and thus were candidates for the multivariate analysis. All were also biologically or theoretically important. For the initial analysis I adopted the analytic approach that argues for the inclusion of all scientifically relevant variables. This approach was possible given the large sample size and statistical significance of all candidate variables in bivariate analysis. Stepwise methods for model building are largely discredited in statistical literature (Greenland 1989; Austin and Tu 2004). As noted earlier they remain useful, however, as part of an exploratory data analysis (Tabachnick and Fidell 2001). I used “Enter”, forward, and backward analytical approaches and assessed model fit using R Square and F Statistics. Backward regression proved to be the most useful in extracting parsimonious models. I have presented the results of both Forward and Backward Regression in Appendix H. I limited the presentation of diagnostics to those undertaken for the Backward Regressions (see Appendix H). I assessed multicollinearity by looking at the correlation coefficients in the Linear Regression Output Correlation Matrix. A weakness of this approach is that multicollinearity between one variable and a combination of variables will not be shown. A subjective approach is to also inspect the standard error for each variable. Where there is a very large SE this implies that multicollinearity exists. Where multicollinearity was suspected the variable with the largest SE was omitted from the model. Assessment of the models was undertaken by reviewing the F test and assessment of model fit using R

236


square. The strength of individual (standardized) covariates in estimating the outcome was assessed and their direction. The further diagnostics undertaken included: the Dubin-Watson test for independence, examination of residual statistics, checking for outliers at 3 and 2 standard deviations, histogram and P-P plotting of the regression standardised residual, scatter plots of the regression standardized residual to standardised predicted values and standardised variables, and identification of influential points using plots of Cooks Distance against Centered Leverage values. Interaction terms were not entered into the models. The purpose of the exploratory multivariate analysis was to identify indicator variables that might be important to the development of theory and later spatial and multilevel analysis. Interaction, mediation and the matter of model external validation, is given further consideration, later in this Chapter and in Chapter 10.

7.3.10

Bayesian Hierarchical Spatial Regression

7.3.10.1

Introduction

As discussed above, the OLS linear regression treats each suburb in isolation so there is no recognition in the model of the spatial relationships that exist between the set of suburbs and no possibility of directly modelling spatial variability (Law and Haining 2004). The diagnostics undertaken for the non-spatial linear regression revealed the presence of significant spatial structure in the distribution of residuals (Appendix H). Law and Haining (2004) note the problem of residual spatial autocorrelation and its consequences for drawing inferences from a model in spatial modelling using the normal linear model. To address this problem they propose fitting logistic models with random variation in the logistic parameter where that random variation is spatially structured. Random effects models may also be useful where, due to data limitation, the analyst is concerned about the effects of important missing covariates, or predictors, in drawing inferences. Where the missing covariates are themselves likely to be spatially structured the specification of the random effect should include a spatially structured element. A key feature of Bayesian spatial and ecological regression models is that they incorporate random spatial effects into the modelling of potential association between the outcome under study and covariate effects at the ecological level. These random spatial

237


effects may reflect unmeasured confounders and thus the model makes it possible to ascertain whether the residual effects suggest spatial patterns or clusters (MacNab 2004). The Bayesian approach is called “hierarchical” because it uses multiple levels of analysis in an iterative way. As noted earlier, unlike the conventional statistical inference which derives the average estimates of parameters, hierarchical Bayesian modelling produces parameter estimates for each individual analysis unit by borrowing information from all analysis units. This enables the analysis of spatially-correlated (structured) and heterogeneity (unstructured) effects (Zhu et al 2006). 7.3.10.2

WinBUGS

The growth of interest in Bayesian approaches to statistical analysis owes much to the development of Markov Chain Monte Carlo (MCMC) methods and software, such as WinBUGS (Spiegelhalter, Thomas et al. 2003), for implementing it . The MCMC make it possible to fit Bayesian models that would otherwise be computationally intractable (Law and Haining 2004). In contrast to the frequentist approach, in Bayesian statistics, parameters are treated as random variables expressed in term of probabilities. Bayesian inference focuses on the modification of prior beliefs about the values of the parameters by new data (Law and Haining 2004). In WinBUGS, MCMC is used “to produce sample drawings from the joint posterior density once it has converged to stationarity” (Law and Haining 2003). Samples taken before the convergence (the “burn in”) are discarded. The methods that I used are described by Lawson, Browne and Vidal Rodeiro (2003), Zhu, Gorman and Horel (2006), Vidal-Rodeiro et al (2003), Haining (2003) and Law and Haining (2004). 7.3.10.3

Data

The outcome and comparison data used for group level analysis is presented above in Section 7.3.5. Technical notes on selected variables are at Appendix F. No additional primary data was used in the Bayesian Hierarchical models described here. All grouplevel covariates were analysed. Only the standardized variables were used. As discussed above the derived Anderson-Rubin factor scores were used in some models. 7.3.10.4

Modelling strategy

The modelling strategy was to first fit a log-normal model for relative risk using observed and expected counts for the outcome variables EDS >9 and EDS > 12. This basic model allowed for subsequent covariate adjustment, for spatial correlation of risk in nearby areas and for the addition of unstructured variance terms (BYM Model). In the ‘Besag, York and

238


Mollie’ (BYM) model for relative risks (Besag, York et al. 1999), “area-specific random effects are decomposed into a component that takes into account the effects that vary in a structured manner in space (clustering or correlated heterogeneity) and a component that models the effects that vary in an unstructured way between areas (uncorrelated heterogeneity)” (Lawson, Browne et al. 2003, p 123). I then added to this basic log-normal model a conditional autoregressive (CAR) component. The resulting WinBUGS code is in Figure 59. To fit the model a file containing data on the adjacency matrix for the study area was generated using the Adjacency Tool in GeoBUGS 1.1 (Spiegelhalter, Thomas et al. 2003). This model was used to derive the smoothed EDS relative risks discussed in Section 7.3.6.1 above. For the bivariate and multivariate models, I added fixed covariate terms resulting in the WinBUGS code in Figure 60. # Spatial model model { for (i in 1:N) { EDS_OBSER[i] ~ dpois(mu[i]); log(mu[i]) 12 Following Bayesian smoothing (Figure 64), visual inspection of the smoothed RR confirms the presence of two possible clusters in the northern suburbs with a lesser cluster in the south. Many of the suburbs that had previously shown high rates, now have average or low rates after smoothing. Suburbs with low rates of EDS are shown in an eastern part of the study area which borders the Georges River, a middle western and a southern group of suburbs. The two later groups of suburbs are in semi-rural areas where there are new housing developments.

250


Spatial Cluster Analysis SaTScan software was used to test for clusters of EDS >9 and EDS >12 and to identify the suburbs with statistically significant clusters (Figure 65). The spatial scan statistic confirms the findings of the previous visual analysis revealing statistically significant high rate clusters of EDS >9 in the north western and eastern areas. The smaller cluster in the south was not statistically significant. The statistic was also able to confirm low rate clusters of EDS > 9 in the suburbs bordering the Georges River and in the southern parts of the study area.

(a)

(b)

Figure 65: Spatial scan statistic clusters (a) EDS >9 (b) EDS > 12 Analysis of EDS > 12 showed a similar pattern but with cluster of EDS > 12 extending across the northern suburbs. The semi-rural areas in Camden and parts of Campbelltown had low rate clusters.

251


7.4.2.3 Visualisation of Candidate Covariates

Measures of Disadvantage (and bliss)

(a)

(b)

Figure 66: Index of Relative Social Disadvantage (a) Score (b) NSW Decile Figure 66(a) above shows the distribution of relative social disadvantage in the study suburbs as measured by the Australian Index of Relative Social Disadvantage. Figure 66(b) is a map of the IRSD NSW deciles. From this map it is clear that almost the whole study region is socially deprived compared to NSW as a whole. There are only a few suburbs that are in the State upper deciles. Both maps show a clear cluster of deprivation in the north west of the study area with smaller clusters in the north east and south east. Those clusters of deprivation are supported by other indicators of deprivation studied and are in the same geographical areas as the clusters of EDS > 9 and EDS > 12 discussed above. There appear to be two clusters of low social disadvantage, one in the south west and the other in the north east. Those areas of low disadvantage are in the same geographical areas as the clusters of low EDS > 9 and EDS > 12 rates identified in the spatial scan statistic (Figure 65). The impression gained from this visual inspection of the distribution of aggregated EDS scores and SEIFA scores is that there is a correlation between social disadvantage and high EDS scores. This finding is in keeping with the findings from the individual level EDA and qualitative studies.

252


(a)

(b)

Figure 67: (a) Family Poverty (b) Family Affluence Figure 67 show maps of family poverty and family affluence. Poverty was defined as the proportion of suburb families with a weekly income below 50 percent of the median income for NSW, and affluence as the proportion of families with a weekly income above the median income for NSW in 2001 (Saunders and Adelman 2005). As would be expected, visual inspection of these maps of poverty and affluence show similar clusters to those discussed above. The poverty, however, is clearly concentrated in a smaller group of suburbs than that suggested by the SEIFA maps above. The north west cluster is concentrated in an area surrounded by suburbs of lower degrees of poverty. By contrast the poor suburbs in the south area are surrounded, in most cases, by suburbs of affluence. The indicator of concentrated extremes (ICE) measures these contrasts within suburbs but not between suburbs (Figure 68(a)).

253


As discussed previously Massey (2001) proposed the use of a measure of concentrated disadvantage and affluence. The Index of Concentration at the Extremes (ICE) is calculated as the number of affluent families minus the number of poor families divided by the total number of families. The spatial distribution of ICE is mapped in Figure 68(a) and is almost identical to the distribution of poverty and affluence in Figure 67 and of the 2001 Census unemployment rate (Figure 68 (b)).

(a)

(b)

Figure 68: (a) Index of Concentrated Extremes (b) Unemployment Rate

254


The qualitative studies suggested that social class may be an important determinant of health outcomes including depressive symptoms. In this study low social class includes blue collar workers and high social class includes managers and professional occupation groups. The spatial distribution of class is mapped in Figure 69. As would be expected there is clustering of low social class in areas of social disadvantage and of high social class in areas with relative affluence.

(a)

(b)

Figure 69: Social Class (a) Low Social Class (b) High Social Class

255


Community violent crime rates are mapped in Figure 70(a) below. The rates of crime were measured at the postal area level and the same rates were assigned to the suburbs making up those postal areas. The chloropleth map shows high rates of violent crime in the southern suburbs and to a lesser degree in the poorer suburbs of the central north. Violent crime rates were low in those suburbs that had affluent families and high social class.

(a)

(b)

Figure 70: Population distribution (a) Violent Crime (b) Vacant Dwellings Vacant dwellings on Census night are shown in Figure 70(b). The distribution is different to that seen previous with high rates in the north eastern suburbs. There are also high rates in the poorer suburbs of Liverpool and Campbelltown local government areas.

256


The population density map chloropleth (Figure 71(a)) shows two areas of high population density at the 2001 Census. The largest is in the north western suburbs of Liverpool and Fairfield and in the western suburbs of Bankstown. The clusters are in similar locations to the clusters of EDS > 9.

(b) Figure 71: (a) Population Density (b) One Parent Family One parent families are known to have significant financial and other hardship. In the ecological EFA (Appendix K) sole parenthood has the highest loading on Factor 1 (disadvantage). The chloropleth map shows high rates of one parent families in those suburbs experiencing poverty (Figure 71(b)). The spatial distribution is also close to that of rental accommodation (Figure 73(b)) and apartment dwellings (Figure 72(a)).

257


Dwelling Type and Ownership The study area is characterised by areas of concentrated apartment dwellings in the northern suburbs and a high percentage of single dwellings in the growth suburbs of the west. There are also more likely to be single dwellings in the affluent suburbs. The suburbs with a high percentage of apartment dwellings (Figure 72(a)) appear on visual analysis to also have increased rental accommodation and social disadvantage.

(a)

(a)

Figure 72: Dwelling Type (a) Apartments (b) Single Houses As expected the maps of rented and owned homes are similar. Suburbs with high rental accommodation will have low home ownership. The pattern of rental accommodation is similar to that for apartments with some exceptions. There are suburbs with high rental accommodation and single dwellings particularly in the central and southern areas of the study area.

258


(a)

(b)

Figure 73: Home ownership (a) Owned (b) Rented The map of aggregated mothers self-reported rental accommodation status, Figure 74 below, is very similar to that reported from the 2001 Census (Figure 73(b)). This provides an element of validation for the ecological data aggregated from the IBIS data set.

(a) Figure 74: Mothers self-reported rental accommodation

259


Measures of Social Capital, Cohesion and Networks Changes in address have been proposed previously as possible measures of community cohesion or connectedness. Figure 75 shows the distribution of families that had different addresses within the previous 1 and 5 years. The spatial distribution is similar and subsequent correlation studies confirmed this. Analysis reveals high mobility in the wealthy suburbs where there is a growth in population moving into newly built single dwellings. By contrast there is also high mobility in disadvantaged suburbs with a high percentage of rental accommodation. The suburbs in the north eastern and western areas of the study area have relatively low changes in address suggesting greater social connectedness. As might be expected there are visual similarities to the map of Factor Six Anderson-Rubin scores (Figure 121(b)).

(a)

(b)

Figure 75: Different Address (a) 1 year previous (b) 5 year previous

260


Aggregated scores of mothers self-reporting “no regret” at leaving a suburb are shown in Figure 76(a). The distribution pattern in the northern suburbs is similar to that for mothers reporting a limited support network (Figure 76(b)). In the southern suburbs the high rates of “no regret leaving” are not associated with high rates of “no support”. The high rates of “no support“ in the northern suburbs shows a pattern that is similar to the distribution of EDS > 9 and EDS > 12 and the measures for disadvantaged reported above.

(a)

(b)

Figure 76: (a) No Regret Leaving (b) No Support The distribution of high rates of “no regret” and “no support” in Figure 76 are opposite to those for recent changes of address. Note that the Ecological EFA (Appendix K) loads both “no support” and “different address at five years” on the same factor but with different direction of the signs.

261


There were limited measures of social cohesion available for this study. Years of schooling (Figure 77) and volunteerism (Figure 78) were two that were used. In the EFA they both loaded on Factor 2 together with Entropy, index of relative social disadvantage and nurse visiting rate.

(a) Figure 77: Schooling less than Year 8 The distribution of “schooling less than year 8” is shown in Figure 77 with high rates in the northern suburbs. The distribution is very similar to that for percentage of volunteerism in Figure 78(b) and both show distributions similar to the EDS smoothed maps.

262


(a)

(b)

Figure 78: (a) Percent Not Volunteer (b) Percent Volunteer

Measures of Diversity and Segregation For measures of ethnic diversity and segregation within suburbs I elected to use the entropy, Simpson and Maly indices (Galster 2004). These measures are described in Appendix F. The following maps demonstrate no significant difference in the distribution of the Entropy Index, Log Entropy or Simpson Index. Later correlation analysis confirms this and I elected to use the Entropy Index in later analyses. The maps in Figure 79 and 80 show that the suburbs in the south of the study area are more ethnically homogenous than those in the northern suburbs. Used in this way the entropy index is a measure of diversity within suburbs. Thus the three indices demonstrate increased ethnic diversity in the northern suburbs.

263


(a)

(b)

Figure 79: Entropy (a) Log Entropy Index (b) Entropy Index

(a)

(b)

Figure 80: Entropy (a) Simpson Index of Segregation (b) Maly Index of Diversity The Maly Index is 1 (completely heterogenous) when all the groups in the neighbourhood match their metropolitan share. The Maly index in this study is high demonstrating that

264


most suburbs have high ethnic diversity. The distribution of the Maly Index was different to the other three and in the later EFA it loaded with measures of support, population density and changes in address.

Access to Health Services Two measures of access to health services were used. They were both related to the activities of the early childhood child and family nursing service. NSW has a policy of providing a home visit to every newborn infant and their mother. As can be seen in Figure 81(a) there is good coverage of the whole area. There is no noticeable trend regarding areas having a higher percentage of universal home visits. By contrast the nurse visiting rate (Figure 81(b)) for infants less than 1 year of age shows a definite trend with the northern suburbs having a lower nurse visiting rate than the southern suburbs.

(a)

(b)

Figure 81: Nurse Access (a) First home visits (b) Nurse visits per infant in first year

265


Health and Health Behaviours Aggregated rates of self-reported poor health are shown in Figure 82(a). Visual inspection of the chloropleth map shows high rates of self-reported poor health in the north western suburbs and those southern areas that had also had high rates of reported depressive symptoms. Notably few suburbs in the north western areas had high rates of self reported poor health.

(a)

(b)

Figure 82: (a) Mothers Poor Self-rate health (b) Unplanned Pregnancy Unplanned pregnancy was included as a measure of less than ideal health behaviour. The visual pattern for high aggregated rates of unplanned pregnancy is similar to that for poor self-rated health. In the ecological EFA both variables loaded on Factor 1 (“disadvantaged and powerless”). This distribution is visually similar to Factor One Anderson-Rubin scores (Figure 116).

266


(a)

(b)

Figure 83: Suburb health behaviours (a) Not breastfeeding (b) Mothers smoking Aggregated rates of not breast feeding was included as a measure of poor health promoting behaviours in a community. The influences on breast feeding rates are complex and may include personal, cultural, community-norm, and service access factors. The chloropleth map (Figure 83(a)) shows two main clusters of not breastfeeding with one in the north west and the other in the southern suburbs. It is notable that the north eastern suburbs show relatively high rates of breastfeeding. By contrast suburbs with high rates of smoking mothers (Figure 83(b)) are mainly in the south of the study area. In both cases the rates are raw with no Bayesian smoothing.

267


7.4.3 Factors and Causal Inference 7.4.3.1 Exploratory Factor Analysis The full exploratory factor analysis is at Appendix K. Below is the Pattern Matrix. Interpretation is in Section 7.5 – Theory Generation.

Factor 1

2

SOLE PARENT %

.912

HOME RENTED %

.818

PUBLIC ACCOMODATION %

.777

UNPLANNED PREG %

.709

CLASS 3 %

.584

POOR %

.567

CRIME RATE

.546

VOLUNTEER %

3

4

5

-.360

-.437

.927

ENTROPY

-.784

-.307

NO VOLUNTEER %

-.651

.305

SCHOOL YEAR 8 OR LESS %

-.567

IRSD DECILE

-.468

-.406

.528

BREAST FEEDING %

.907

SMOKING %

.899

NO REGRET LEAVING %

.795

APPARTMENT % SINGLE HOUSE %

-.825 -.427

.728

HOUSE VACANCY RATE

-.637

NURSE VISIT RATE

.405

HOME VISIT RATE POOR HEALTH %

.767 .535

.304

.363

NOT SUPPORT %

-.679

NO PRACTICAL SUPPORT %

-.589

DENSITY

-.452

DIFFERENT ADDRESS 5 YEARS MALY INDEX

Table 26: Factor Analysis Oblique Rotation Output – Pattern Matrix

268

6

-.493 .349

.395 .308


7.4.3.2 Assumptions of Causal Inference It was clear from the bivariate (Appendix G) and exploratory factor analysis (Appendix H) that the dataset contains a large number of measured indicator variables that are associated both with suburbs with high rates of maternal depressive symptoms and with each other. Some of the measured variables will be in the causal path, and some will be confounders. The EFA also suggests that there are a number of latent (unmeasured) variables related to emerging concepts such as “depressed community”, social capital, social cohesion and access to services. Adjustment for variables that may be confounders is necessary as part of EDA and ESDA to ensure against spurious findings. But before entering variables into a regression model it is important to make explicit assumptions of causal inference. As discussed earlier (Chapter 5), care should be taken to not control for variables that are descents (effects) of the outcome understudy (Greenland, Pearl et al. 1999). The analysis here will use the “Back-Door Criterion” for selecting covariates for adjustment. In this case I believe that the aggregated variable “poor self-reported health” may be a descendant (effects) of living in a suburb of high rates of mothers with depressive symptoms. That is that EDS and selfreported health may be measures of the same process. As discussed in Chapter 5, I postulate that “not breastfeeding” is also an “effect” of maternal depression. At the time of undertaking this analysis I did not consider any of the variables to be distal to any others, or alternatively I did not consider any of the variables to be exclusively mediators on the causal path. I delimited mediation analysis and did not undertake any analysis to test these assumptions. The limitations of this approach are discussed later in this Chapter and in Chapter 10.

269


7.4.4 Likelihood Linear Regression Results Below are regression coefficients for aggregated EDS >9 and EDS > 12. The full results are at Appendix H. For EDS > 9 the forward and backward regressions resulted in identical coefficients.

(Constant) Entropy

Unstand ardized Coefficients Std. B Error -1.39E-015 .085

Nurse Visit Rate % No Support Network % Apartment Dwelling

Standar dized Coefficie nts Beta

95% Confidence Interval for B t .000

Sig 1.000

Lower -.170

Upper .170

Colinearity Statistics Toleranc e VIF

.353

.117

.353

3.028

.003

.122

.585

.542

1.844

.320

.099

.320

3.225

.002

.123

.517

.748

1.336

.382

.104

.382

3.664

.000

.175

.589

.679

1.473

-.251

.100

-.251

-2.501

.014

-.450

-.052

.734

1.361

a Dependent Variable: EDS9 %

Table 27: EDS9 Coefficients

The results of forward and backward regression were different for EDS > 12 (Tables28 and 29) suggesting the presence of suppressor effects. Manual entering and removal of variables was undertaken (see Table 30).

Forward Regression M o d el

1

Unstand ardized Coefficients Std. B Error (Constant) % No Support Network

-3.24E-016

.094

.336

.095

Standa rdized Coeffic ients Beta

.336

95% Confidence Interval for B t

Lower

Upper

1.000

-.187

.187

3.555

.001

.149

.524


Table 28: EDS > 12 Forward Regression Coefficients

270

Sig

.000

Collinearity Statistics Tolera nce VIF

1.000

1.000


Backward Regression Unstand ardised Coefficients Std. B Error (Constant)

1.18E-015

.087

Entropy

.318

.137

% No Volunteer Nurse Visit Rate % Smoking %No Regret Leaving

-.260 .324 -.387

.112 .102 .151

.328

%Class 3 %School Year 8 or Less % Home Owned

Standar dized Coeffici ents Beta


Sig

Lower

Upper

Colinearity Statistics Tolerance

VIF

.000

1.000

-.172

.172

.318

2.316

.023

.045

.591

.403

2.480

-.260 .324 -.387

-2.332 3.188 -2.562

.022 .002 .012

-.482 .122 -.688

-.039 .526 -.087

.610 .734 .333

1.639 1.363 3.007

.160

.328

2.047

.044

.010

.646

.296

3.377

.710

.199

.710

3.563

.001

.314

1.106

.191

5.231

-.357

.160

-.357

-2.230

.028

-.676

-.039

.296

3.382

.437

.167

.437

2.619

.010

.106

.768

.273

3.661


Table 29: EDS > 12 Backward Regression Coefficients

Manual Regression Unstand ardised Coefficients Std. B Error (Constant) Entropy Nurse Visit rate %Home Rented % Class 3 %School Year 8 or Less %No Support Network

4.26E-017

.088

.326

.133

.292 -.282 .517

Standar dized Coeffici ents Beta


Sig

Lower

Upper

Colinearity Statistics Tolerance

VIF

.000

1.000

-.174

.174

.326

2.442

.016

.061

.591

.438

2.282

.103 .141 .181

.292 -.282 .517

2.835 -1.995 2.858

.006 .049 .005

.088 -.562 .158

.497 -.001 .877

.735 .391 .238

1.361 2.559 4.203

-.426

.163

-.426

-2.613

.010

-.750

-.102

.293

3.414

.290

.117

.290

2.476

.015

.057

.523

.567

1.763


Table 30: EDS > 12 Manual Regression Coefficients

271


7.4.5 Spatial Linear Regression (Bayesian) I first fitted a log-normal model followed by models with spatial and uncorrelated variance. Table 31 shows the DIC values. It can be seen that the models with only the spatial random effects were the better fit. EDS > 9 Variable

EDS > 12

Dbar

Dhat

Pd

DIC

Dbar

Dhat

Intercept Only

600.733

599.732

1.001

601.734

501.027

500.046

0.981

502.008

Spatial Only

541.880

518.256

23.623

565.503

451.146

429.189

21.957

473.102

Uncorrelated Only

552.426

524.141

28.285

580.712

465.919

444.106

21.814

487.733

BYM (a+s+u)

541.487

515.302

567.673

452.057

428.169

23.887

475.944

26.185

Pd

DIC

Table 31: Comparison of DIC for Models with random effects and no covariates Since it was not practical or desirable to consider every possible permutation of candidate factors, I started with the two random effects and confirmed that the BYM models consistently were a poorer fit than models with only the spatial random. alpha + bX + s Variable

alpha + bX + u + s

Dbar

Dhat

Pd

DIC

Dbar

Null

542.063

518.659

23.404

565.468

Density

543.599

525.281

18.318

561.917

542.143

514.461

27.682

569.826

Entropy

546.948

529.578

17.37

564.317

539.487

514.277

25.210

564.698

%No Volunt.

542.229

516.882

25.348

567.577

542.308

507.375

34.933

577.240

%Yes Volunt.

546.289

530.828

15.460

561.749

541.240

516.556

24.683

565.923

Nurse Visit

541.846

516.416

25.430

567.276

540.087

508.468

31.619

571.706

% Poor

547.615

529.094

18.522

566.137

541.187

513.946

27.240

568.427

ICE

542.882

522.087

20.796

563.678

542.253

513.811

28.442

570.695

539.864

Dhat 505.149

Pd 34.715

DIC 574.579

Table 32: Comparison of DIC for selected EDS >9 Bivariate models with random effects I then fitted every candidate variable using the spatial random effects model. The resulting coefficients, parameters and DIC for EDS >9 and EDS >12 Bivariate models are at Appendix I respectively. Significance was assessed as nonzero regression coefficients at the 95% credible interval.

272


7.4.5.1 Best Fitting Spatial Models – EDS >9 The EDS > 9 model with the lowest DIC and no non-significant coefficients included: % No Support, Entropy Index, %living in apartments and %smoking. Models that included %no regret leaving and %lone parent families also performed well. The further addition of covariates resulted in an increase in the DIC in all instances and usually made the other covariates non-significant. Colinearity was avoided by not entering variables that were known to be highly correlated from previous non-spatial analysis (correlation matrix, EFA and OLS linear regression). Models

Pd

DIC

No Covariate CAR Model

23.404

565.468

No Support

15.907

560.012

No Support + Entropy

13.012

558.326

No Support + Entropy + %Apartment

12.373

554.417

No Support + Entropy + %Apartment + nurse visit rate

10.449

555.714

No Support + Entropy + %Apartment + %No regret leaving

10.146

554.072

9.296

553.803

No Support + Entropy + %Apartment + %Smoking

10.543

552.535

No Support + Entropy + %Apartment + %Smoking (BYM)

12.340

553.200

No Support + Entropy + %Apartment + %One Parent Families

Table 33: EDS >9 best fitting Spatial CAR regression models The factor analysis was utilised to inform the analysis ensuring that candidate variables were added from each of the factors previously identified. Interestingly the final models had one covariate from four of the six identified factors. If one parent family and nurse visiting are included then there were significant covariates from each of the factors identified in the group level EFA. The findings suggest that variables loading on Factors 2, 4 and 6 are proximal and that variables loading on Factors 1, 3 and 5 may be distal and possibly mediated through Factors 2 or 6. Moderation and mediation analysis was not undertaken. The detailed results of the final aggregated EDS >9 model that includes %no support, Entropy Index, %living in an apartment and %smoking is reported at Appendix J.

273


7.4.5.2 Best Fitting Models – EDS >12 The EDS > 12 models with the lowest DIC and no non-significant coefficients included: % No Support, Entropy Index, and either %no regret leaving suburb or %smoking. There were no other models with three covariates where the coefficients were significant. As above, colinearity was avoided by not entering variables that were known to be highly correlated from previous analysis (correlation matrix, EFA and OLS linear regression). In addition the autocorrelation trace was examined in the WinBUGS monitor. Models No Covariate No Support No Support + Entropy No Support + Entropy + Smoking No Support + Entropy + No Regret Leaving Suburb

Pd

DIC 21.481 11.433 10.352 7.997 9.208

473.277 466.878 462.743 460.240 459.622

Table 34: EDS >12 best fitting Spatial CAR regression models As previously the factor analysis was utilised to inform the analysis. The final models had one covariate from three of the six identified factors. Interestingly there were no variables from Factor 1 representing the possible latent variable of “disadvantage”. This finding suggests that variables loading on Factor 1 may be distal to variables from Factor 2, 3 and 6. Moderation and mediation analysis was not undertaken.

274


7.4.5.3 Final Reported EDS > 12 Spatial As above I have elected to report here in detail the best fitting of the aggregated EDS >12 model namely that including the covariates:% No Support, Entropy Index and “%No Regret Leaving”. I have used map decomposition of the full BYM model to enable visualisation of the full range of residuals. # Spatial model EDS12 for map decomposition full GYM model { for (i in 1:N) { EDS12_OBSER[i] ~ dpois(mu[i]); log(mu[i]) 12 in the final BYM model. The clustering of EDS > 12 can be seen in the northern suburbs in a similar pattern to that observed previously (SMR -Figure 63, Smoothed RR – Figure 64 and SatSCAN Clusters – Figure 65). Figures 91 to 93 show the RR for areas in the north are strongly driven by the covariates “% No Support”, and Entropy. The covariate % No Regret Leaving is also making a contribution to the RR in several areas of the map. As previously the maps of residuals are strongly dominated by the spatial component exp(s[i]) as shown in Figure 94. The unexplained spatial residual is strongest in the southern suburbs. The implication is that covariates that would eliminate the spatial residual have not been included in the model and possibly have not been identified among the candidate variables.

279


Figure 89: Decomposition of Final Model EDS >12 Bayesian Regression The scales in Figure 89 are manually kept the same to assist comparison. The scales in Figures 90 to 95 use the preset Quantile classifications in ArcView 9.3.

280


Figure 90: Map of the Relative Risk of EDS > 12 – Final Model (BYM)

281


Figure 91: Contribution by covariate “%No Support” – EDS >12 Final Model (BYM)

282


Figure 92: Contribution by covariate Entropy Index – EDS > 12 Final Model (BYM)

283


Figure 93: Contribution by covariate “%No Regret” – EDS >12 Final Model (BYM)

284


Figure 94: Contribution by spatially unexplained components – EDS > 12 (BYM)

285


Figure 95: Comtribution by unstructured unexplained components – EDS >12 (BYM)

286


7.4.6 Spatial Multilevel Logistic Regression 7.4.6.1 Main Effects Models The Spatial Multilevel Bayesian Logistic Regression Main Effects Model described (Figures 62 and 99) was fitted for individual EDS >9 and EDS >12. Table 38 shows the DIC and pD values for fitted individual level covariates and suburb level CAR. EDS >9 Models

pD

DIC

a + FINSIT*b

2.055

9999

FINSIT+SPNET

3.006

9801

FINSIT+SPNET+SPEMO+

3.988

9716

FINSIT+SPNET+SPEMO +SPRAC

5.031

9696

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP

5.903

9546

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP + BDEMAND

7.009

9521

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP + BDEMAND + BCONTENT

7.935

9503

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP + BDEMAND + BCONTENT + HEALTH

8.829

9358

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP + BDEMAND + BCONTENT + HEALTH + COB

9.99

9341

FINSIT+SPNET+SPEMO +SPRAC + BTSLEEP + BDEMAND + BCONTENT + HEALTH + COB + suburb CAR

32.991

9322

pD

DIC

EDS >12 Models a + FINSIT*b

2.067

5656

FINSIT+SPNET

3.113

5505

FINSIT+SPNET+SPEMO+

3.971

5438

FINSIT+SPNET+SPEMO +BTSLEEP

5.032

5345

FINSIT+SPNET+SPEMO +BTSLEEP + BDEMAND

5.904

5325

FINSIT+SPNET+SPEMO + BTSLEEP + BDEMAND + HEALTH

6.068

5230

FINSIT+SPNET+SPEMO + BTSLEEP + BDEMAND + HEALTH + suburb CAR]

8.897

5215

FINSIT-Financial Situation, SPNET-Support Network, SPEMO-Emotional Support, SPRAC-Practical Support, BTSLEEPBaby Trouble Sleeping, BDEMAND, Baby Demanding, BCONTENT-Baby not content, HEALTH-Maternal Self Report Health, COB-Country of Birth, suburb CAR - the CAR random intercept

Table 38: EDS > 12 Logistic Regression Individual and Multilevel Models As predicted by the earlier logistic regression (Chapter 5), country of birth was not significant in the EDS >12 models. The fitting of the CAR random intercept improved both EDS >9 and EDS > 12 models indicating the existence of between-area variability.

287


Table 96 chloropleths show the Relative Risks of the two CAR models in Table 38. To assist analysis the relevant ecological CAR models are also shown. The differences between the two sets of chloropleths are the result of controlling for individual level covariates.

Figure 96: Bayesian CAR Models

288


The pD and DIC for multilevel models with covariates at the suburb level are at Table 39.

Individ. + CAR Public Accom. Change Add. 1yr Change Add. 5yr High Apartments Breast Feeding Violent Crime Pop Density Entropy Ind. Apartments Owned Dwelling Rented Dwellings ICE IRSD IRSD Decile MALY Index No Regret Leav. No Social Support Class 1 Class 3 Sole Parent Unplanned Preg. Poor Families Poor Schooling Simpson Index Smoking rates UHV Rate Unemploy. Rate Vacant Dwelling Nurse Visit Rate No Volunteer Volunteer Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6

EDS > 9 pD DIC 33.411 9322.00 35.811 9323.03 35.405 9322.85 34.746 9323.22 23.927 9325.57 35.934 9323.87 34.474 9324.15 32.728 9325.03 36.034 9322.89 27.665 9323.22 33.770 9323.65 34.041 9322.94 29.528 9324.38 30.038 9326.21 31.747 9323.27 27.698 9326.27 36.687 9322.86 22.420 9327.09 33.049 9323.17 31.342 9323.97 33.400 9325.20 34.802 9323.16 34.602 9322.97 29.192 9322.65 28.700 9322.84 31.438 9323.60 32.517 9320.69 * 28.700 9322.84 31.343 9321.51 33.301 9320.86 * 35.114 9323.28 32.794 9323.70 34.123 9324.27 35.795 9322.85 34.221 9322.76 30.160 9323.30 33.660 9321.78 * 26.101 9327.81

EDS > 12 pD DIC 12.117 5215.23 12.781 5216.98 13.796 5215.84 12.328 5216.23 9.786 5215.09 14.101 5216.09 15.989 5215.94 12.894 5214.23 * 14.708 5216.88 9.775 5215.05 16.564 5215.62 10.74 5216.17 9.956 5213.91 * 12.659 5214.09 * 16.023 5215.01 10.53 5214.40 * 14.738 5216.81 11.226 5211.69 * 16.772 5214.74 * 15.281 5215.18 15.513 5216.00 12.499 5215.06 13.108 5214.69 * 13.329 5211.65 * 13.173 5216.14 12.783 5215.78 10.619 5214.77 * 12.93 5215.01 14.971 5215.40 13.927 5214.98 * 13.52 5217.27 16.549 5215.24 13.050 5216.88 14.356

5216.52

13.515

5216.88

12.448

5215.96

14.589

5215.88 5209.26*

11.670

Comment

- sign, not significant

+sign, not significant +sign, not significant +sign, not significant - sign, significant +sign, not significant

- sign, not significant - sign, significant



Not significant Not significant Not significant Not significant - sign, not significant + sign, significant

* DIC decreased by more that 1.

Table 39: Main Effect Multilevel Models with suburb level fixed effects

289


7.4.6.2 Born in Australia Stratified Multilevel Analysis The Multilevel Bayesian Model described was fitted to the stratified sample of mothers born in Australia (n= 7366). Table 40 shows the DIC values for individual level logistic models with covariates and for a model with suburb level CAR. EDS > 9 Models

pD

DIC

a + FINSIT*b

2.049

4427

A FINSIT+SPNET

3.062

4368

FINSIT+SPNET+SPEMO+

4.041

4311


4.879

4306


6.096

4205


6.926

4187


8.071

4186


9.096

4112

+ Suburb CAR

10.389

4111

EDS > 12 Models

pD

DIC

a + FINSIT*b

2.098

2374

A FINSIT+SPNET

3.05

2327

FINSIT+SPNET+SPEMO+

3.911

2292


4.889

2293


6.215

2239


6.838

2215


8.142

2215


8.824

2132

+ Suburb CAR

10.926

2131

FINSIT-Financial Situation, SPNET-Support Network, SPEMO-Emotional Support, SPRAC-Practical Support, BTSLEEPBaby Trouble Sleeping, BDEMAND, Baby Demanding, BCONTENT-Baby not content, HEALTH-Maternal Self Report Health, COB-Country of Birth, Suburb CAR – CAR random intercept

Table 40: Born in Australia Logistic Regression Individual and Multilevel Models The notable finding from the Born in Australia multilevel model is failure for the DIC to decrease when the random effects CAR term was added indicating that there is no between suburb variability. As might be expected none of the ecological observed or latent variables reduced the DIC.

290


Figure 97 chloropleths show the Relative Risks of the two CAR models in Table 40. To assist analysis the relevant ecological CAR models are also shown. The differences between the two sets of chloropleths are the result of controlling for individual level covariates.

Figure 97: Bayesian CAR Models – Australian Born Mothers

291


The pD and DIC for the multilevel models with covariates at the suburb level at Table 41.

Individual + CAR Public Accommodation. Change Add. 1yr Change Add. 5yr High Apartments Breast Feeding Violent Crime Pop Density Entropy Ind. Apartments Owned Dwelling Rented Dwellings ICE IRSD IRSD Decile MALY Index No Regret Leaving. No Social Support Class 1 Class 3 Sole Parent Unplanned Pregnancy. Poor Families Poor Schooling Simpson Index Smoking rates UHV Rate Unemployment Rate Vacant Dwelling Nurse Visit Rate No Volunteer Volunteer Factor 1 Factor 2 Factor 3 Factor 4 Factor 5 Factor 6

EDS > 12

EDS > 9 pD

DIC

pD

DIC

10.389

4111.95

10.926

2131.17

14.816

4113.88

8.622

2132.71

12.143

4111.68

10.577

2132.90

13.396

4111.97

9.421

2133.29

15.086

4113.85

10.926

2132.78

15.76

4113.28

9.251

2132.71

17.995

4113.62

10.699

2133.16

10.845

4111.63

11.32

2132.55

12.691

4113.79

10.856

2132.76

13.696

4113.45

10.932

2132.79

15.423

4113.87

9.601

2132.58

15.832

4113.81

9.335

2132.59

16.801

4113.13

10.66

2132.84

12.979

4113.97

11.259

2133.44

14.458

4113.37

12.497

2132.02

17.137

4112.57

10.082

2131.55

19.89

4111.55

10.941

2133.15

15.763

4113.56

9.579

2132.66

14.906

4113.82

10.967

2133.15

15.818

4114.11

9.747

2133.00

14.558

4113.43

9.33

2133.04

14.531

4113.90

9.088

2133.08

13.847

4113.51

10.522

2132.57

16.951

4112.68

10.383

2132.67

16.936

4113.32

12.766

2132.99

16.105

4113.53

8.989

2132.82

14.933

4113.08

9.321

2131.95

12.548

4113.62

12.244

2133.37

19.566

4113.45

10.556

2132.91

14.114

4113.44

11.133

2132.42

12.526

4113.95

11.437

2133.39

16.232

4114.41

10.605

2132.76

12.811

4113.50

12.988

2134.70

17.353

4113.03

12.723

2134.45

17.774

4112.61

11.338

2134.50

15.432

4113.85

12.134

2134.66

12.147

4113.88

11.595

2134.98

14.308

4114.11

11.383

2134.40


Table 41: Born In Australia Multilevel Models with suburb level fixed effects

292


7.4.6.3 Not Born in Australia Stratified Multilevel Analysis The Multilevel Bayesian Model described was fitted to the stratified sample of mothers Not Born in Australia (n= 7256). Table 42 shows the DIC values for individual level logistic models with covariates and for a model with suburb level CAR. EDS > 9 Models

pD

DIC

a + FINSIT*b

1.923

5496

A FINSIT+SPNET

3.007

5406

FINSIT+SPNET+SPEMO+

4.071

5375


4.977

5361


6.251

5300`


6.806

5293


7.909

5278


8.950

5221

+ Suburb CAR

26.386

5200

pD

DIC

1.887

3249

EDS > 12 Models a + FINSIT*b A FINSIT+SPNET

3.047

3173

FINSIT+SPNET+SPEMO+

4.040

3142


5.043

3131


6.154

3093


6.75

3091


8.048

3091


9.172

3062

+ Suburb CAR

18.498

3053

FINSIT-Financial Situation, SPNET-Support Network, SPEMO-Emotional Support, SPRAC-Practical Support, BTSLEEPBaby Trouble Sleeping, BDEMAND, Baby Demanding, BCONTENT-Baby not content, HEALTH-Maternal Self Report Health, COB-Country of Birth, Suburb CAR – CAR random intercept

Table 42: Not Born in Australia Logistic Regression Individual and Multilevel Models The notable finding from the Not Born in Australia multilevel model is a large decrease in the DIC when the random effects CAR term was added indicating that there is between suburb variability. The ecological observed or latent variables that further reduced the DIC were mainly associated with Factor 6 (see Table 43).

293


Figure 98 chloropleths show the Relative Risks of the two CAR models in Table 42. To assist analysis the relevant ecological CAR models are also shown. The differences between the two sets of chloropleths are the result of controlling for individual level covariates.

Figure 98: Bayesian CAR Models – Mothers Not Born in Australia

294


The pD and DIC for multilevel models with covariates at the suburb level are in Table 43.

Individual + CAR Public Accomm. Change Add. 1yr Change Add. 5yr High Apartments Breast Feeding Violent Crime Pop Density Entropy Ind. Apartments Owned Dwelling Rented Dwellings ICE IRSD IRSD Decile MALY Index No Regret Leaving. No Social Support Class 1 Class 3 Sole Parent Unplanned Preg. Poor Families Poor Schooling Simpson Index Smoking rates UHV Rate Unemployment Vacant Dwelling Nurse Visit Rate No Volunteer Volunteer Factor One Factor Two Factor Three Factor Four Factor Five Factor Six

EDS > 9 pD 26.386 29.071 29.142 29.256 18.162 29.604 25.238 18.485 25.166 20.226 25.875 25.987 23.249 24.295 25.429 21.032 27.981 12.394 27.803 23.944 27.48 25.446 19.807 23.333 24.578 24.852 25.934 24.19 20.593 26.87 26.456 25.617 28.681 28.107 26.573 22.655 27.289 22.098

DIC 5200.37 5200.13 5200.38 5200.75 5200.28 5199.80 5200.67 5201.53 5200.41 5199.61 5200.31 5200.10 5198.63* 5198.34* 5199.28 5197.48* 5200.36 5198.44* 5199.25 5197.74* 5201.89 5199.56 5204.08 5198.69* 5200.87 5202.32 5197.00* 5198.93* 5197.70* 5196.81* 5199.43 5198.00* 5200.35 5199.86 5200.59 5199.39 5198.09* 5198.81*

NS NS NS -ve Sig NS NS -ve Sig NS -ve Sig NS NS +ve Sig +ve Sig NS +ve Sig NS -ve Sig NS -ve Sig NS NS -ve -ve Sig NS NS NS -ve Sig -ve Sig NS NS NS NS NS NS NS NS NS

EDS > 12 pD 18.498 17.851 18.62 17.562 17.578 18.195 18.449 11.55 7.828 18.925 17.886 17.167 17.186 15.3 17.121 14.829 19.126 12.233 17.412 15.338 18.576 16.349 16.52 10.987 16.972 14.628 18.486 15.949 19.019 19.512 17.777 19.514 19.316 18.767 19.031 18.830 18.130 13.247

DIC 3053.22 3055.29 3053.89 3054.14 3055.01 3053.40 3054.56 3048.18* 3054.46 3054.30 3054.02 3055.23 3053.36 3054.44 3054.29 3051.69* 3055.31 3049.77* 3054.36 3053.08 3055.11 3052.67 3053.83 3050.76* 3056.20 3054.77 3054.65 3053.20 3053.92 3054.73 3054.87 3055.66 3053.94 3055.34 3055.18 3054.36 3053.76 3046.56*

NS NS NS NS NS NS -ve Sig NS NS NS NS NS NS NS +ve Sig NS -ve Sig Ns NS NS NS NS -ve Sig NS NS NS NS NS NS NS NS NS NS NS NS NS + sig


Table 43: Not Born In Australia Multilevel Models with suburb level fixed effects

295


7.4.6.4 Final Reported EDS > 12 Spatial Multilevel Not Born in Australia with Factor 6 fixed effect As above, I have elected to report here in detail, the best fitting of the aggregated EDS >12 Multilevel model namely that including Factor Six as a covariate. I have also used map decomposition enable visualisation of the full range of residuals.

# Multilevel Spatial model EDS12 for map decomposition Factor Six fixed effect model{ for (i in 1:Nwomen) { IEDS12[i] ~ dbern(p[i]); logit(p[i]) 12 Multilevel Regression – Factor Six alpha[10] chains 1:2 0.4 0.3 0.2 0.1 0.0 3001

4000

5000

6000

i terat ion

alpha[1] chains 2:1

alpha[10] chains 2:1 0.4 0.3 0.2 0.1 0.0

-2.4 -2.6 -2.8 -3.0 -3.2 5850

5900 iteration

5950

5850

5900

5950

iteration

Figure 101: Sample Trace and History Plots the Posterior Distribution – EDS >12 Multilevel Regression – Factor Six

299


OR.alpha[9] chains 1:2


1.0 0.5 0.0 -0.5 -1.0

1.0 0.5 0.0 -0.5 -1.0 0

20

40

0

20

lag

40 lag

Figure 102: Sample Autocorrelation Graphs – EDS > 12 Multilevel Regression OR.alpha[9] chains 1:2


1.5

1.0

1.0

0.5

0.5 0.0

0.0 3051

3500

4000

3051

start-iteration

3500

4000

start-iteration

Figure 103: Sample Gelman-Rubin Convergence Plots – EDS >12 Multilevel Regression

Map Decomposition As previously described I used “map decomposition” to visualise the results of the final model. This allowed high and low areas of relative risk to be identified as well as an assessment of the contribution made by the covariates, and spatially structured random effects (Note: no unstructured effects were possible). To assist interpretation I have included Multilevel Model with no suburb level fixed effect (Figure 104(a)). Clustering of EDS > 12 can be seen in the southern suburbs in a similar pattern to that observed previously for Non-Australian born mothers (Smoothed RR – Figure 98). Figure 104(b) shows the final model and Figure 104(c) the RR contribution from Factor Six. The unexplained spatial residual (Figure 104(d)) is strongest in the southern suburbs. The implication is that covariates that would eliminate the spatial residual have not been included in the model and possibly have not been identified among the candidate variables.

300


Figure 104: Decomposition of EDS > 12 Multilevel Regression with Factor Six as Suburb level fixed effect The scales in Figure 104 are manually kept the same to assist comparison. The scales in Figures 105 to 107 use the preset Quantile classifications in ArcView 9.3.

301


Figure 105: Map of the Relative Risk of EDS >12 – Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect

302


Figure 106: Contribution of Factor Six - EDS >12 Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect

303


Figure 107: Contribution of spatially unexplained components - EDS >12 – Final Model Spatial Multilevel Regression with Factor Six as Suburb Level Fixed Effect

304


7.4.6.5 Final Reported EDS > 12 Spatial Multilevel Not Born in Australia with No Support fixed effect As above I have also elected to report here in detail the EDS >12 Multilevel model with “No Support” as the Suburb level fixed effect. I have also provided map decomposition enable visualisation which will contribute to interpretation. The WinBUGS code was as per Figure 99 with substitution of NSPT for FSIX. The final model was run using two chains simultaneously. Convergence was monitored by visual examination of the trace plots of the samples for each chain, autocorrelation graphs, and the Gelman-Rubin convergence statistic. Convergence was judged to have occurred by 3000 iterations. Those samples were discarded as burn in and each chain was run for a further 3000 samples with acceptable MC errors – Table 48 below. node Intercept Financial Situation Social Support Emotional Support Practical Support Baby Trouble Sleeping Baby Demanding Baby Not Content Self Reported Health No Support Network

mean ‐2.697 0.376 0.236 0.459 0.369 0.185 0.082 0.078 0.330 ‐0.137

sd 0.072 0.056 0.043 0.126 0.119 0.059 0.056 0.052 0.054 0.044

MC error

2.50%

0.00172 0.00001 0.00001 0.00212 0.00217 0.00128 0.00116 0.00001 0.00001 0.00121

‐2.841 0.267 0.154 0.216 0.140 0.067 ‐0.029 ‐0.024 0.224 ‐0.225

median 97.50% ‐2.695 0.375 0.238 0.461 0.370 0.185 0.083 0.078 0.329 ‐0.138

‐2.558 0.485 0.321 0.706 0.600 0.297 0.190 0.180 0.437 ‐0.050

Table 48: Coefficient Monte Carlo Error – EDS > 12 Spatial Multilevel Regression

305


Table 49 shows the odds ratio for the EDS > 12 Multilevel model. The odds of EDS >12 were increased in suburbs with increased loading of No Support Network. node

mean

Financial Situation Social Support Emotional Support Practical Support Baby Trouble Sleeping Baby Demanding Baby Not Content Self Reported Health No Support Network

1.459 1.270 1.596 1.456 1.205 1.087 1.083 1.393 0.872

sd 0.08158 0.05471 0.20220 0.17370 0.07107 0.06079 0.05657 0.07634 0.03882

MC error

2.50%

0.00129 0.00001 0.00338 0.00322 0.00154 0.00126 0.00101 0.00119 0.00113

1.306 1.167 1.242 1.149 1.070 0.972 0.976 1.251 0.799

median 97.50% 1.456 1.269 1.585 1.447 1.204 1.086 1.081 1.390 0.871

1.624 1.379 2.026 1.823 1.346 1.210 1.197 1.548 0.951

Table 49: Odds Ratio – EDS > 12 Spatial Multilevel Regression Table 50 shows the precision and standard deviation of the reported model. node Precision Standard Deviation

mean 198.4 0.1403

sd 450.6 0.07199

MC error 33.36 0.006247

2.50%

median

97.50%

12.86 0.0277

52.86 0.1376

1304 0.279

Table 50 : Precision and Standard Deviation – EDS > 12 Multilevel with No Support

Diagnostics As previously for the sensitivity analysis I compared the results of fitting the final model using a more informative hyper-prior of Gamma (0.1, 0.1) and the non-informative hyperprior Gamma (0.001, 0.001). The results of the two non-informative hyper-priors are almost identical. Compared to the non-informative hyper-priors, the informative hyper-prior has a poorer model fit (larger DIC) (see Table 51 below).

No Support as Suburb Level Fixed Effect

Intercept (2.5%, 97.5% DIC

Hyper-priors (variance) of the random effects Gamma Gamma Gamma (0.1, 0.1) (0.5, 0.0005) (0.001, 0.001) -2.700 -2.697 -2.697 (-2.858, -2.546) (-2.841, -2.558) (-2.836 -2.546) 3051.56 3049.05 3048.58

Table 51 : Sensitivity to Choice of Hyper-priors on Random Effects – EDS > 12 Multilevel with No Support

306


OR.alpha[2] chains 1:2 sample: 6000 6.0

OR.alpha[3] chains 1:2 sample: 6000 8.0 6.0 4.0 2.0 0.0

4.0 2.0 0.0 1.2

1.4

1.6

1.8

1.0

OR.alpha[4] chains 1:2 sample: 6000

1.4

OR.alpha[5] chains 1:2 sample: 6000

3.0

3.0

2.0

2.0

1.0

1.0

0.0

1.2

0.0 0.5

1.0

1.5

2.0

2.5

0.5

OR.alpha[6] chains 1:2 sample: 6000 6.0

1.0

1.5

2.0


4.0 2.0 0.0 0.8

1.0

1.2

1.4

0.8


1.0

1.2

OR.alpha[9] chains 1:2 sample: 6000 6.0 4.0 2.0 0.0

0.8

1.0

1.2

1.0

1.2

1.4

1.6


0.8

0.9

1.0

Figure 108: Density Plot of the Posterior Distribution – EDS >12 Multilevel Regression – No Support Figure 108 displays the density plot of the posterior distribution of the coefficients. It is notable that the density plot OR. Alpha[10] is less than 1.0 but does not cross 1.0. This indicates that the aggregated variable “no support network” decreases rates of depression for mothers not born in Australia. The variable is significant in that the density plot does not cross 1.0. Figure 107 shows the trace plots and a sample history plot. Figure 108 shows the autocorrelation graphs and Figure 109 the Gelman-Rubin convergence plots for the checking of convergence. These plots indicate that the chains converged and there was no significant colinearity.

307


OR.alpha[10] chains 2:1 1.0 0.9 0.8 0.7 5850

5900

5950

iteration

alpha[10] chains 1:2 0.1 2.77556E-17 -0.1 -0.2 -0.3 3001

4000

5000

6000

iteration

Figure 109: Sample Density and History Plot of the Posterior Distribution – EDS >12 Multilevel Regression – No Support

OR.alpha[10] chains 1:2 1.0 0.5 0.0 -0.5 -1.0 0

20

40 lag

Figure 110: Sample Autocorrelation Graph – EDS > 12 Multilevel Regression

alpha[10] chains 1:2 1.0 0.5 0.0 3051

3500

4000

start-iteration

Figure 111: Gelman-Rubin Convergence Plot – EDS >12 Multilevel Regression

308


Map Decomposition As previously described I used “map decomposition” to visualise the results of the final model. This allowed high and low areas of relative risk to be identified as well as an assessment of the contribution made by the covariates, and spatially structured random effects (No unstructured effects were possible). To assist interpretation I have included Multilevel Model with no suburb level fixed effect (Figure 112(a)). Clustering of EDS > 12 can be seen in the southern suburbs in a similar pattern to that observed previously for Non-Australian born mothers (Smoothed RR – Figure 98). Figure 112(b) shows the final model and Figure 112(c) the RR contribution from No Support Network. The unexplained spatial residual (Figure 112(d)) is strongest in the southern suburbs.

309


Figure 112: Decomposition of EDS > 12 Multilevel Regression with “No Support” as Suburb level fixed effect The scales in Figure 112 are manually kept the same to assist comparison. The scales in Figures 113 to 115 use the preset Quantile classifications in ArcView 9.3.

310


Figure 113: Map of the Relative Risk of EDS >12 – Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect

311


Figure 114: Contribution of Factor Six - EDS >12 Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect

312


Figure 115: Contribution of spatially unexplained components - EDS >12 – Final Model Spatial Multilevel Regression with “No Support” as Suburb Level Fixed Effect

313


7.5

Theory Generation

In this section I will draw on the ESDA and EFA findings of this Chapter to generate theoretical concepts that can then be developed further in the next Chapter – Theory Construction. As observed previously factor analysis assists in this theory generation process. I will thus use the factor analysis as the basis for theory generation and will interpolate the findings from the other components of the ESDA (i.e. visualisation, correlation and bivariate analysis, ORL regression, and Bayesian spatial regression and spatial multilevel regression). The discussion here is intentionally limited in scope with further detailed analysis in Chapter 8 – Theory Construction. As noted earlier, for theoretical EFA where there is expected to be correlation between factors the most appropriate method of extraction is “common factor analysis” using oblique rotation. In this situation both the structural and pattern matrices are used to interpret and describe the factors. A full description of the EFA method and findings is at Appendix K. Examination of the Factor Correlation Matrix (Table 52) reveals moderate correlations between Factors1, 3, and 6. This correlation is important from a theoretical perspective and may be material to the theory building. In particular Factor 1 is negatively correlated with Factor 6 which was found to be an important factor in the multilevel regression studies.

Factor 1

1

2

3

4

6

-.089

.404

-.218

-.020

-.365

2

-.089

1.000

-.064

.070

.181

.257

3

.404

-.064

1.000

-.105

-.089

-.148

4

-.218

.070

-.105

1.000

-.014

.229

5

-.020

.181

-.089

-.014

1.000

-.066

6

-.365

.257

-.148

.229

-.066

1.000

Extraction Method: Principal Axis Factoring. Rotation Method: Oblimin with Kaiser Normalization.

Table 52: Factor Correlation Matrix

314

5

1.000


Factor 1

2

3

4

SOLE PARENT %

.965

.442

HOME RENTED %

.888

.397


.843

.511

POOR %

.816

-.338

CLASS 3 %

.779

-.542

UNPLANNED PREG %

.755

CRIME RATE

.575

POOR HEALTH %

.458

VOLUNTEER %

-.329

ENTROPY SCHOOL YEAR 8 OR LESS % IRSD DECILE

-.527

-.321

.517

-.476

-.582

.511

-.318

-.502

.316

-.440

.301

.334

-.437

.945 -.341

.392

-.717

-.317

-.653

.663 -.603

.558 .313

.368

.904

BREAST FEEDING %

.332

.883

NO REGRET LEAVING %

.385

.832

APPARTMENT % -.591

HOUSE VACANCY RATE

-.637

-.357

SMOKING %

.312

NURSE VISIT RATE

6 -.385

-.805

NO VOLUNTEER %

SINGLE HOUSE %

5

-.860

-.319

.832

.379

-.658

.538

.830

HOME VISIT RATE

.523

MALY INDEX NOT SUPPORT %

.324


.327

DENSITY

-.389

-.385

-.796 -.616

-.599

DIFFERENT ADDRESS 5 YEARS

-.611 .334

.389

Extraction Method: Principal Axis Factoring. Rotation Method: Oblimin with Kaiser Normalization.

Table 53: Factor Analysis Oblique Rotation Output – Structure Matrix

315


Factor 1

2

SOLE PARENT %

.912

HOME RENTED %

.818


.777

UNPLANNED PREG %

.709

CLASS 3 %

.584

POOR %

.567

CRIME RATE

.546

VOLUNTEER %

3

4

5

-.360

-.437

.927

ENTROPY

-.784

-.307

NO VOLUNTEER %

-.651

.305


-.567

IRSD DECILE

-.468

-.406

.528

BREAST FEEDING %

.907

SMOKING %

.899

NO REGRET LEAVING %

.795

APPARTMENT % SINGLE HOUSE %

-.825 -.427

.728

HOUSE VACANCY RATE

-.637

NURSE VISIT RATE

.405

HOME VISIT RATE POOR HEALTH %

6

.767 .535

.304

.363

NOT SUPPORT %

-.679


-.589

DENSITY

-.452


-.493 .349

MALY INDEX

.395 .308

Table 54: Factor Analysis Oblique Rotation Output – Pattern Matrix

Factor One “Disadvantaged and Powerless” Pattern Matrix Factor 1 includes a number of variables that represent suburbs with disadvantage and low status with the principal loading being a high percentage of sole parents (0.9612), rental accommodation (0.818), high percentage of maternal accommodation in public housing (0.777), higher rates of families living in poverty (0.567), higher rates of blue collar workers (0.584), higher rates of unplanned pregnancy (0.709), communities with violent crime (0.546) and less significantly a low IRSD decile (-0.468).

316


The Structure Matrix Factor 1 contributes to conceptualisation of this factor and includes low volunteer rates (-0.329), low schooling (0.392), smoking (0.368), no breast feeding (0.332) and non regret leaving (0.385), no practical support (0.327), no support network (0.324) and a negative loading on single house (-0.591). Examination of both the pattern and structure matrices suggest that there is a latent variable related to disadvantage and powerlessness which is correlated with low social cohesion, unhealthy behaviour, apartment housing and lack of social support networks (social capital).

(a)

(b)

Figure 116: (a) Factor One (b) Factor One-Blue Factor One was spatially dominant in the built-up urban suburbs of Fairfield, Liverpool and Campbelltown and visually correlated with the distribution of family poverty, ICE, unemployment, one parent families and low social class. There was also some visual correlation with aggregated “no regret leaving” and “no social support networks”. It is notable that [latent] Factor One and the observed variables loading on Factor One were not significant in either the main effect ecological or multilevel regressions. ICE and IRSD did reduce the DIC and were significant in the Not Born in Australia EDS > 9 multilevel model. As discussed later, these findings may have been the result of control for financial difficulties at the individual level. Alternatively Factor One, and associated

317


observed variables, may have been on the causal path but distal to more proximal factors such as Factor Six.

Factor Two “Social Cohesion” Factor Two includes a mixture of variables that are conceptually linked to various aspects of cohesion and integration. The principal loadings are high rates of volunteer activity (0.927), a lack of “no volunteer activity” (-0.651), a lack of poor education (-0.567). There is also a loading of variables indicating higher social class and advantage with a high IRSD Decile (0.528), lack of blue collar class (-0.437) and lower population density (0.452). There is a negative loading of high entropy (0.784) which has been used here to measure population diversity. Volunteer activity is negatively correlated with diversity. (Separate analysis found that suburbs with high rates of volunteer activity are negatively correlated with high rates of population not born in Australia). Analysis of the Structure Matrix reveals a similar set of loaded variable with the addition of negative loadings to no support (-0.389), access to nurses (0.538), Class 3 (-0.542) and poverty (-0.338). This suggests that Factor Two (social cohesion) correlated with social support and access to nurses, and negatively correlated with poverty and low class. Factor Two was visually dominant in southern suburbs where there were higher rates of Australian born mothers, volunteerism, affluence and greater years of schooling. Factor Two was also correlated with Factor Six. Factor Two did not reduce the DIC and was not significant in the multilevel variables. But the related observed variables of poor schooling, volunteerism, nurse visit rate, no support network, class 3 and family poverty either reduced the DIC and/or were significant in the main effect and Mothers Not Born in Australia multilevel models. Entropy and No support were both significant in the ecological regressions.

318


(a)

(b)

Figure 117: (a) Factor Two Red (b) Factor Two Green On balance, Factor Two (and its loaded observed variables) was associated with depressive symptoms particularly among mothers not born in Australia. But as with Factor One, the limited importance in the regression studies may have been due to control for social support networks at the individual level and/or its distal position on any causal path.

Factor Three “Health Behaviours” The Pattern Matrix Factor Three includes three aggregated variables that could be described as representing unhealthy norms of behaviour and attitude. They are smoking (0.899), not breastfeeding (0.907) and having no regret at leaving the suburb (0.795). The variable “no regret at leaving the suburb” has been used in previous studies as an indicator of social capital (Health 2006, p 400). Here the variable is correlated with community norms of smoking and not breastfeeding. The Structure Matrix contains the additional loadings (mainly from Factor One) of sole parenthood (0.442), rental accommodation, (0.397), poverty (0.517), public accommodation (0.511), low class (0.511), unplanned pregnancy (0.316), poor self-report of health (0.301) and low IRSD Decile (-0.357). This suggests that suburbs with poor “health behaviours” are also likely to be disadvantaged and powerless.

319


Factor Three and associated observed variables were visually associated with the distribution of depression and both smoking and “no regret leaving” were significant in the ecological regression but not the multilevel regressions.

(a)

(b)

Figure 118: (a) Factor Three Red (b) Factor Three Green Based on these findings, Factor Three might not be on a causal pathway to maternal depression. Factor Three is, however, strongly correlated with Factor One and may represent a set of aggregated outcome variables arising from living in disadvantaged and powerless circumstances.

Factor Four “Housing Quality” The Pattern Matrix Factor Five has three significant variables related to housing including apartment living (-0.825), housing vacancy rates (-0.637) which are both negatively correlated with percentage of single housing in a suburb (0.728). Also correlated in the Pattern Matrix are rental accommodation (-0.36), entropy (-0.307) and low volunteer activity (0.305). The Structure Matrix also loads with poverty (-0.476), class 3 (-0.318), low schooling (0.317), and no support (-0.385).

320


(a)

(b)

Figure 119: (a) Factor Four Red (b) Factor Four Green Apartment living was significant in the ecological regression but Factor Four and its associated observed variables were significant in the multilevel regressions. Housing may be important but the analyses reported here do not support a proximal association with depressive symptoms.

Factor Five “Access to Services” The Pattern Matrix loadings for Factor Five are nurse visits in the first year of life (0.767), first home visits (0.535), self-reported poor health (0.363), and different address in the last five years (0.349). The Structure Matrix does not contribute any loadings to this factor suggesting that this Factor is not correlated with the other factors.

321


(a)

(b)

Figure 120: (a) Factor Five Red (b) Factor Five Green Nursing visiting was important in the ecological regressions and reduced some of the multilevel DIC although not significantly. Access to services may, therefore, be playing a role in relation to maternal depressive symptoms.

Factor Six “Support Networks or Social Capital” Two of the aggregated support variables loaded on Factor Six in the Pattern Matrix. Those are no practical support (-0.589) and no support (-0.679). I elected not use emotional support in the analysis as it aggregated emotional support does not plausibly represent a group level construct. Also loading on this factor are the variables of “different address five years previously” (0.395), low schooling (-0.406), Maly Index of diversity (0.308), and density of population (-0.493). The Structure Matrix also loads sole parent (-0.385), rental accommodation (-0.321), poverty (-0.582), class 3 (-0.502), unplanned pregnancy (-0.440), poor self reported health (-0.437), low decile (0.558), apartment living (-0.319), low percentage of single houses (0.379). It is clear that the Support Network factor shares variance with the factors “disadvantage and powerless” and “housing quality”. A latent factor “support” had also been previously identified at the individual level by Categorical PCA factor analysis and Exploratory Specification Search (Chapter 5).

322


(a)

(b)

Figure 121: (a) Factor Six Red (b) Factor Six Green In this study Factor Six and its associated observed variables were visually associated with the distribution of depressive symptoms. Support measures were also negatively correlated with measures of disadvantaged. No Social Support was significant in the ecological regression and together with Factor Six was significant in the main effect and “not born in Australia” multilevel regressions. The strength of these effects suggest that Factor Six and its associated observed variables are proximal in the causal pathway. Paradoxically “no social support networks” was protective at the ecological level and the counterfactual “social support networks” was promoting of depressive symptoms among those mothers “not born in Australia”. These findings will be explored further in the next Chapter.

323


7.6

Conclusion

The purpose of this Chapter was to describe the quantitative exploration of ecological level factors that might be associated with the phenomenon of perinatal depression, and perinatal adversity. The bivariate exploration and visualisation of spatial distributions identified association of depressive symptoms with ethnic segregation, no social support, volunteerism, IRSD, population density, schooling, occupational class, unemployment and ICE. The six key phenomenon identified by the factor analysis included community disadvantage and powerlessness, social cohesion, health damaging behaviours, housing, access to services and social support networks. Also important was ethnic diversity and segregation. Stratified analysis by “mothers country of birth” found that there was limited spatial clustering of mothers born Australia. By contrast mothers not born in Australia were spatially clustered. The exploratory spatial regression identified entropy, no social support network, percentage of apartments, smoking behaviour, one parent families, nurse visiting rates and “no regret leaving” as being associated with aggregated depressive symptoms. The multilevel studies found no ecological effects for mothers born in Australia. The Main Effect and “Mother not born in Australia” multilevel studies found improved model fit for population density, ICE, IRSD, Maly Index, no social support network, occupational Class, poor families, poor schooling, nursing visiting rates, unemployment, vacant dwellings and volunteerism. The most consistent multilevel finding was improved model fit for latent variable Factor Six and associated observed variables such as “no social support”. The findings were paradoxical in that “no social support” at the ecological level protected mothers not born in Australia from depressive symptoms. The findings from this Chapter are discussed further in Chapter 8 – Theory Construction. Critical analysis of the studies is reported in Chapter 10 – Conclusion, Limitations and Implications.

324

Part D: Theory Construction and Discussion

Part D: Theory Construction Discussion and Recommendations Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







325

Part D: Theory Construction and Discussion

326


Chapter 8: Theory Construction Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







327


8.1

Introduction

In this Chapter I will focus on the conceptual development of a Theory of Maternal

Depression and Context. As I have taken an “emergent” approach to theory building, it is necessary to detect phenomena and describe concepts through the interpretation of persons involved as the first stage in “explanatory research based on critical realism” (Danermark, Ekstrom et al. 2002; Haig 2005). As recommended by Danermark and colleagues (2002, p 109), I have used both qualitative and quantitative methods to describe the events, situations and concepts that I intend to study (Chapters 3 to 7). The Explanatory Theory Building process began during the “Exploration Phase” of this study and I made explicit in each chapter the elements of “theory generation” undertaken through abductive methods such as exploratory factor analysis, focused or substantive coding (Haig 2005; Haig 2005) and situational analysis. The Chapter will briefly summarise the philosophical, and methodological of the Explanatory Phase of Theory building using a critical realist approach followed by a method section that will elaborate on the critical realist explanatory research approach I have taken. The following sections will be broadly based on the components of the emerging Theory of Maternal Depression and Context.

328


8.2

Aims

The aim of this chapter is to: 1. To explain and conceptualise a framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression. 2. To describe future operationalisation and confirmation. The specific objectives are to: 1. Integrate the analysis 2. Undertake abductive triangulation 3. Compare theories 4. Identify the best explanation 5. Develop testable propositions 6. Modify the conceptual framework Specific analysis questions for Theory Construction Phase are: 1. How do individual and family level factors influence the phenomenon of maternal depression during the perinatal period? 2. How do group level factors influence the phenomenon of maternal depression during the perinatal period?

329


8.3

Methods

8.3.1 Theoretical Approaches In Chapter Two I discussed the philosophical and methodological approaches used in this study. I have adopted critical realist ontological and epistemological positions and utilised a Concurrent Triangulation Mixed Method design. The critical realist approach to theory building enabled the study to embrace a postpositivist approach to theory building relevant to my social epidemiology and population health background. Critical realism argues that the social world is an “open system” and that causal powers can produce different outcomes depending on the context or spatialtemporal relationships with other objects. I contend, therefore, that critical realism is an appropriate epistemology for the generation of causal explanations in social epidemiology. The concurrent triangulation design provided for integration at various stages of the research process including analysis question, data gathering, data analysis and this final stage of data interpretation and theory construction. Some of the methods described in this section have also been employed to some extent during the Emergent Phase including comparative analysis, analytic resolution, abduction, theoretical redescription, retroduction, and modelling of causal networks. In this section I will briefly describe the methods used in this Chapter and how they will be applied in this Theory Construction Phase. Those methods include: 1. Defining Stratified Levels 2. Building a Conceptual Framework 3. Analytic Resolution 4. Abductive reasoning 5. Comparative analysis (triangulation) 6. Retroduction 7. Postulate and Proposition Development 8. Comparison and Assessment of Theories 9. Modelling.

330


8.3.2 Conceptual Frameworks A conceptual framework explains, either graphically or in narrative form the key factors, constructs or variables and the presumed relationships among them (Miles and Huberman 1994, p 18). The graphical form of a conceptual framework is similar to “concept mapping” (Maxwell 2005) and the “integrative diagram” (Strauss 1987, p 170). Miles and Huberman (Miles and Huberman 1994, p 22) advise that conceptual frameworks are best done graphically rather than in text “having to get the entire framework on a single page obliges you to specify the bins that hold the discrete phenomena, to map the likely relationships, to divide variables that are conceptually or functionally distinct, and to work with all the information at once”. They also argue for many iterations and the use of prior theorising and empirical research. The “frameworks can be rudimentary or elaborate, theory-driven or commonsensical, descriptive or causal”. Based on prior knowledge a Perinatal Conceptual Framework was illustrated in Chapter 2. Subsequent theory generation occurred during the Emergent Phase and entailed the development of conceptual networks and models in both the quantitative and qualitative studies. The Emerging Conceptual Framework for Maternal Stress (Chapter 4 and 6) included the more significant emerging concepts. Given the ability of qualitative (intensive) studies to identify possible causal explanation, the findings from the qualitative studies were used in this Chapter to guide the analysis. Further modifications of these conceptual frameworks were constructed to progressively include findings from the analysis in this Chapter. Causal networks are closely related to conceptual frameworks/maps. A causal network is intended to illustrate causal relationships. In an emergent study as here it is the intention to move from a “conceptual framework” towards a “causal network”. In principal a full causal network could have been drawn at end of each empirical chapter. Miles and Huberman (1994, p158) drew attention to the risks of constructing an early “causal map” which is then used “to interpret all the phenomena of interest [and] a final “coherent” picture emerges before the individual parts of the picture have been carefully studied individually and in combination. ….. The better alternative is to save full causal network drawing and analysis for late in the project, making it perhaps the last analytic exercise”

331


8.3.3 Stratified Levels A hallmark of critical realist analysis is the ontological assumption that reality consists of hierarchically ordered levels where a lower level creates the conditions for a higher level. The higher level is not, however, determined by the lower level and has its own “generative mechanisms”. The existence of these level specific generative mechanisms is what constitutes or defines a level. The implication of this stratification is that it is not possible to reduce the causes of what occurs on one level to those on another level (whether lower or higher). In their study of social justice and disability Danermark and Gellerstedt (2004, p 350) argue that:

“Mechanisms are working on different levels and on each level such mechanisms that cannot be reduced to another level are at work, i.e. critical realism emphasizes a non-reductionist perspective (see Danermark et al., 2002). This implies that injustices to disabled people can be understood neither as generated by solely cultural mechanisms (cultural reductionism) nor by socio-economical mechanisms (economic reductionism) or by biological mechanisms (biological reductionism). In sum, only by taking different levels, mechanisms and contexts into account, can disability as phenomenon be analytically approached”. (Danermark and Gellerstedt 2004, p 350)

Source: (Danermark and Gellerstedt 2004, p 350)

Table 55: Analytical Levels in Disability Research

332


The above approach is useful as an analytical framework but “in reality levels are entwined and [the] mechanisms could be supporting each other or counteracting each other, and the outcome in a specific context is the result of a very complex interplay between levels and mechanisms” (Danermark and Gellerstedt 2004, p 350). If research focuses on one level (or two as in this study) this approach ensures that there is awareness of the existence and importance of other levels influencing the phenomenon and thus is sensitive to context. The multi-level approach described here is different to the statistical multi-level regression analysis of individual and neighbourhood spatial units in Chapter 7. That analysis contributed to phenomenon detection and subsequent theory generation, but has limited applicability to the abductive reasoning necessary for theory construction. Spatially determined multi-level regression studies are also constrained by the limits of regression methods (Oakes 2004; Mingers 2006) and the selected areal units of analysis. The situational analysis method utilised in the qualitative studies has some applicability to the critical realist view of a layered reality. Clarke (2005, p 72) argued that “everything in the situation both constitutes and affects most everything else in the situation in some way”. This view is consistent with the “entwined” and complex interplay between levels and mechanisms described by Danermark and Gellerstedt (2004). In this Chapter I will use the approach described by Danermark and Gellerstedt (2004) to conceptualise of mechanisms working at various levels. I will focus this analysis on mechanisms operating at the psychosocial and social levels while maintaining awareness of the existence and importance of the other levels.

8.3.4 Analytic resolution Analytic resolution is the second stage of explanatory research as proposed by Danermark and colleagues (2002, p 110). In this stage the authors argue that the “composite and complex” are separated and dissolved “by distinguishing the various components, aspects or dimensions”. This is necessary because “it is never possible to study anything in all its different components” and “therefore we must in practice confine ourselves to studying certain components and not others”. Analytic resolution was previously undertaken in both the qualitative and quantitative chapters as part of procedures such as coding of categories, situational analysis and factor analysis. Further analytic resolution was undertaken during the analysis in this

333


chapter. The process will contribute to the identification of the best theoretical explanation. A risk of this process is the loss of detail regarding the complexity of the processes under study. To partially address this the analytic process involved regular checking back to the empirical findings in both qualitative and quantitative chapters. This resulted in some codes and variables being “reinstated” with greater theoretical significance. I have taken “analytical resolution” to also give permission for the analysis here to be confined to the study of certain components of the emerging theory and not others. I have consequently elected to limit the abductive and retroductive explanatory analysis in this Chapter to mechanisms operating principally at the social level. This is based on the following analysis:

The phenomena of postnatal depressive symptoms is the empirical manifestation of a great number of cooperative and counteractive mechanisms – for example social isolation, loss of control of life, social exclusion, and infant temperament. These mechanisms in specific contexts together constitute one or more unobserved structures operating at biological, psychological, psychosocial or social levels. In this way the studied phenomenon of postnatal depressive symptoms can be described as being a result of several causal mechanisms and these causal mechanisms are located partly in the family structure and partly in the community structure both at the social level. It is, therefore, at the social level that external causal mechanisms arise (emergent powers) and under certain circumstances their effect results in postnatal depression. Each such part can then be related to different ideas and causal theories. (adapted from Danermark et al (2002), p 112)

334


8.3.5 Abductive Reasoning Abductive reasoning was introduced in Chapter 2 as the reinterpretation and recontextualisation of individual phenomena within a conceptual framework or a set of ideas. It is about being able to understand something in a new way by observing and interpreting this something in a new conceptual framework. Modell (2009, p 213) observes that abduction “does not move directly from empirical observations to theoretical inferences, as is the case in purely inductive research, but relies heavily on theories as mediators for deriving explanations”. Pierce (1960, p 117) described the formal logic of abductive reasoning as follows

“A surprising fact, C, is observed But if A were true, C would be a matter of course Hence, there is reason to expect that A is true”. Eco’s typology of abduction includes overcoded, undercoded and creative types of abduction (Eco 1984) cited by Danermark (2002, p 93). According to Eco, overcoded abduction is a mode of inference consisting of spontaneous interpretations based on cultural and social prejudging. Thus all observations involve some form of interpretive abductive process being a precondition for the observed phenomenon having any meaning at all. This interpretive abduction occurred naturally in the previous empirical chapters resulting, for example, in the initial codes in the qualitative chapters. By contrast undercoded abduction is where we choose between a number of possible frames of interpretation or theories. This process of abductive inference began in the empirical chapters as part of the second phase of coding and subsequent theory generation and will continue in this chapter. Here we interpret particular phenomena as part of general structures. The third type of abduction, creative abduction, is characterised by being unique and innovative and moving to a frame of interpretation that nobody has used before, or which “at least opposes conventional interpretations”. I have approached this process in three phases. In the first I have sought to recontexualise or redescribe the phenomenon identified within one of the more abstract concepts emerging from the Emergent Phase. This abductive inference was imbedded within the theory generation processes in Chapters 4, 5, 6 and 7. The second phase used in this Chapter was to recontexualise maternal depression through the lens of theories arising from literature, key informants and the previous theory generation. Through analytical resolution I rejected theories at non-social levels and focused on: stress process

335


theory, social exclusion, social isolation, social capital and ethnic diversity. Finally abduction was undertaken as part of the Comparison of Theories stage. The following summarises the process:

A redescription of each part is carried out using the respective theories. These descriptions make it possible to find a number of possible underlying causes. Plausible explanatory models are discussed. Several possible theories are identified at different stratum. I then reject theories at non-social levels (particularly biological theories). In critical realist terms we can say that I am looking for the emergent powers at the stratum where my study object is namely the social level. In relation to the respective theory the question is then asked ‘what in different social structures is it that can produce maternal stress and depression” (adapted from Danermark et al (2002), p 112)

8.3.6 Comparative Analysis (triangulation) As stated in Section 2.8.4 I have used an integration of methods, data collection and analysis as proposed by Yin (2006) and Woolley (2009). Comparative analysis was used during the Emergent Phase and in this way the two arms of the study remained integrated. Given that both the qualitative and quantitative arms of the study were emergent it was possible to use findings from the qualitative study to inform the quantitative study (i.e. selection of variables for study, interpretation of EFA). Similarly quantitative findings were entered as memos into the qualitative analysis. In this stage of comparative analysis findings from intensive (case-orientated) and extensive (variable-orientated) study designs will be compared. The intensive qualitative studies provide causal explanations of possible mechanisms while the extensive quantitative studies assist with distinguishing regularities, patterns and features of the population groups. During this phase of comparative analysis the relevant literature will be reviewed in more depth and this source of information will be treated as a third source of information for the comparative analysis. Attention will be paid to convergence and divergence of findings. Divergence of findings will be given particular attention as it is here that “new” knowledge or understanding may be elicited through abductive and retroductive reasoning. Thus the comparative analysis was undertaken using both the abductive and retroductive processes as described by Danermark and colleagues (2002, p 88-106).

336


The comparative analysis used here is also known as “triangulation”. Critical realist critiques of triangulation have been provided by Downward and Mearman (2007), Modell (2009), Danermark and colleagues (2002, p 152), and Kelle (2001). Modell (2009, p 208) argues that critical realism counters many criticisms of triangulation “by re-conceptualising it as firmly grounded in abductive reasoning. This provides a foundation for maintaining researchers’ sensitivity to context-specific variations in meaning in efforts to derive theoryrelated explanations”. Downward and Mearman (2007, p 77) argue that mixed-methods triangulation can be considered a “manifestation of retroduction, the logic of inference espoused by critical realism”. Kelle observed that there were three different understandings of the triangulation metaphor, “triangulation as mutual validation, triangulation as the integration of different perspectives on the investigated phenomenon and triangulation in its original trigonometrical meaning” (Kelle 2001, p 1). In a consideration of the micro- and macrolevels of sociological analysis Kelle suggests that an “understanding of triangulation in its original trigonometrical sense may be helpful in gaining a deeper insight into theoretical aspects of method integration” (Kelle 2001, p 1). Comparative analysis, or triangulation, will be used in this Chapter in its original trigonometrical sense, pulling together intensive and extensive findings, and micro and macro analysis as the basis for abductive and retroductive search for generative mechanisms.

8.3.7 Retroduction Retroduction is a process where we move from a description and analysis of concrete phenomena to reconstruct the basic conditions for these phenomena to be what they are. In this way thought operations and counterfactual thinking are used to argue toward counterfactual and transfactual conditions. “With counterfactuals, the antecedents need not be instantiated; with transfactuals the consequents need not be realised” (Fleetwood 2010, p 10). In this type of analysis we look for the necessary conditions to make the phenomenon possible. The critical realist analysis of causal inference was discussed in Section 2.4.3. In this view structures, or entities, may have causal powers or mechanisms that have a tendency to produce events or outcomes. Contextual conditions play an important role in the realist understanding of causality because causal powers may only result in an event

337


occurring under certain conditions. Thus the outcome patterns are also determined by context (Pawson 2006). The following summarises the retroductive process:

Identify and describe several fundamental generative mechanisms that can explain the phenomenon. What is fundamentally constitutive for the structures and relations (X) highlighted in stage 3 [Abductive/redescription]? How is X possible? What properties must exist for X to be what X is? What causal mechanisms are related to X? (adapted from Danermark et al (2002), p 112) The methods for determining what structures and mechanisms make up the conditions for X to be possible are unclear. Danermark and colleagues (2002) identify six strategies which might be used. They are: counterfactual thinking, thought experiments, social experiments, studying pathological circumstances and extreme cases, and comparison of different cases. I have limited the retroductive analysis here to counterfactual thinking, thought experiments and comparative analysis of the qualitative and quantitative studies. Divergent or paradoxical findings provide excellent material for retroductive analysis.

8.3.8 Postulates and Propositions Development Theoretical statements can be expressed as postulates, propositions or hypotheses. There is a logical order in which these terms are used here moving from the general to the specific. In this Chapter I have focused on the development of proposition statements leaving hypothesis development to a future Confirmatory Phase. A postulate is “a proposition that is accepted as true in order to provide a basis for logical reasoning” (Cambridge Dictionary). I will thus use the term here as a theoretical expression from which the subsequent propositions are reasoned. A proposition is defined by Hellevik (1984) as a statement about the relationship between variables. From a hypothetico-deductive perspective Dubin (1969, p 205) states that propositions are often expressed as “if a then b ” deductive statements. Dubin further observes that:

338


“A proposition is a truth statement about a model when the model is fully specified in its units, laws of interaction, boundary, and system states. Any truth statement that can be made about such as system is a proposition of the system” The hypothetico-deductive approach to theory building requires a “closed” system. To achieve this Dubin argues that boundaries and system states must be defined. The approach also requires that the theory be tested within the empirical world with “things observable” (Dubin 1969, p 205). By contrast critical realism views reality as an “open system” were causative processes are always contextually determined. Smith (2008, p 5) states that critical realist explanation includes:

“the structure that underlies the generative mechanisms (structure of X), the outcome that these mechanisms tend to produce (Y), and finally the elements of context that trigger or inhibit the firing of these generative mechanism (C). Any explanation must include all three of these elements” Thus theoretical propositions are about how “mechanisms (M) are fired in contexts (C) to produce outcomes (O)” (Pawson and Tiley 1997, p 85). Explanation cannot begin without the identification of the generative mechanism and their underlying structures. Causal relationships only occur when the generative mechanism comes into operation. Sometimes different mechanisms produce the same outcome. The contextual conditions determine whether the generative mechanism(s) will come into play and nature of the outcome. The contextual conditions include other mechanisms that may either trigger or counteract the causal mechanism. This can be illustrated the following mechanism (M), context (C), outcome (O) (MCO) model: Conditions (other mechanisms)

Mechanisms

Outcome

X

Y

Z Source: (Danermark 2002)

Figure 122: Graphical representation of critical realist MCO propositions

339


I will use the above graphical representation of a critical realist proposition to summarise the findings of the comparative analysis (triangulation), and abductive and retroductive reasoning. Where possible I will seek to identify those mechanisms that I consider necessary and those that I consider conditional.

8.3.9 Comparison and Assessment of Theories Critical realist methodologist Danermark and colleagues (2002), and Haig (2005) both identify a stage in explanatory research and theory construction where comparison and assessment of the identified theories and abstractions is undertaken. The purpose here is not to test or confirm the theoretical propositions but rather to use further abductive reasoning to identify the best explanation. Danermark and colleagues (2002, p 110) suggest that in this stage one elaborates and estimates the relative explanatory power of the mechanisms and structures identified by previous abduction and retroduction. The result may be one theory that “describes the necessary conditions for what is to be explained”, or alternatively theories may be complementary “as they focus on partly different but nevertheless necessary conditions”. One theory is then said to have more explanatory power than another theory about the same subject matter if it can predict and otherwise account for all the facts that the second one does, but also explains the causes of other facts which the second one does not. Haig (2005, p 380) proposes the use of “inference to the best explanation, which accepts a theory when it is judged to provide a better explanation of the evidence than its rivals do”. Haig argues for the use of Thagard’s (1992) formulation of inference to the best explanation (theory of explanatory coherence (TEC)) , which identifies, and systematically uses, a number of evaluative principals and criteria in a way that has been shown to produce reliable judgments of best explanation in science. Thagard’s seven principles are: symmetry, explanation, analogy, data priority, contradiction, competition, and acceptability. The determination of the explanatory coherence of a theory is made in terms of three criteria: consilience, simplicity, and analogy (Thagard 1988). The criterion of consilience, or explanatory breadth, is the most important criterion for choosing the best explanation. It captures the idea that a theory is more explanatorily coherent than its rivals if it explains a greater range of facts. The criterion of simplicity is that preference should be given to theories that make fewer special or ad hoc assumptions. Finally, analogy is an important criterion of inference to the best explanation because it can improve the explanation offered by a theory. Explanations are judged

340


more coherent if they are supported by analogy to theories that scientists already find credible. Within TEC, each of the three criteria of explanatory breadth, simplicity, and analogy are embedded in one or more of the seven principles. Thagard (1992; 2000) formulated these principles in both formal and informal terms. They are stated here informally in his words as follows (Thagard 2000, p 96-97):

“Symmetry: Explanatory coherence is a symmetric relation, unlike, say, conditional probability. That is, two propositions p and q cohere with each other equally. Explanation: (a) A hypothesis coheres with what it explains, which can either be evidence or another hypothesis. (b) Hypotheses that together explain some other proposition cohere with each other. (c) The more hypotheses it takes to explain something, the lower the degree of coherence. Analogy: Similar hypotheses that explain similar pieces of evidence cohere. Data Priority: Propositions that describe the results of observations have a degree of acceptability on their own. Contradiction: Contradictory propositions are incoherent with each other. Competition: If p and q both explain a proposition, and if p and q are not explanatorily connected, then p and q are incoherent with each other (p and q are explanatorily connected if one explains the other or if together they explain something). Acceptance: The acceptability of a proposition in a system of propositions depends on its coherence with them.” The Bradford Hill aspects of association are commonly used approach to assessing causal claims in epidemiology. Hill suggested that the following aspects of an association be considered in attempting to distinguish causal from non-causal associations.

341


The following restatement of Hills aspects of association is adapted from Thygesen and colleagues (Thygesen, Andersen et al. 2005):

Strength: if an observed effect of A on E is strong it could implicate causality, Consistency: if an observed effect of A on E is shown in different situations it could indicate causality Specificity: it is preferable if the causal association is between one specific cause A and one specific effect E Temporality: in order for A to cause E, A must precede E in time Biological gradient: a higher exposure to factor A may result in a higher incidence of the disease E, or a higher exposure to factor A may result in a higher biological effect Plausibility: the association ought to be biologically plausible Coherence: the association should not conflict with the generally known facts of the biology of the disease Experimental evidence: whether the effect E occurs when the factor A is present and absent, respectively, while no other factors vary. Other factors X and Y in the full cause are important for the occurrence of E. In the case Y is present or X is absent the prevalence of E is independent of A. The experiment will only support the causal claim of A on E where X is present and Y is absent Analogy: we should be ready to accept a slighter but similar evidence with a similar risk factor and a similar affect. Ward (2009) critically reviewed the role of Hill’s “criteria” and argued that they are not appropriate for either deductive or inductive inferences but that they have an important role to play in abductive inferences to the best explanation. This conclusion clearly places the commonly used Bradford Hill “criteria” within a realist epistemology and therefore appropriate for assessment of the theories constructed here. Inference to Best Explanation is an abductive mode of inference considered by some philosophers of science to be an alternative to the hypothetico-deductive and Bayesian probabilistic approaches to evaluation of theory (Thagard 1978). In the following analysis the Thagard and Hill principles of best explanation will be used as part of the abductive process to identify the best explanation. The application of the principles will be summarised following each proposition.

342


8.3.10

Modelling

The terms “conceptual frameworks”, “theories” and “models” are often used in interchangeably and together. As stated in Chapter Two, I have used the typology proposed by Carpiano and Daley (2006) where the different levels of abstraction move from the broadest level of conceptualisation (framework) to the more focused (model). A model is the narrowest in focus and is used to make specific assumptions about a limited set of parameters and variables. A model may draw on several theories and when presented as a diagram a conceptual model may provide a visualisation of proposed causal linkages (Carpiano and Daley 2006). Diagrams are used to as visual representation of categories and their relationships. Many grounded theorists treat creating visual images of theory as an intrinsic part of grounded theory methods (Charmaz 2006, p117). Such methods are consistent with critical realist qualitative and quantitative methodology. Miles and Huberman (1994, p222-238) give an extensive account of the use of causal chains, networks and models including the use of factor analysis and modelling of variable relationships. Borsboom and colleagues (2003) examined the theoretical status of latent variables as used in modern test theory models. They argued that a consistent interpretation of such models requires realist ontology for latent variables. Borsboom (2008) also argues that observed variables are rare in psychology and that psychological variables should be considered latent until proven otherwise. Using the Latent Variable Theory proposed by Borsboom, variables used in this study should also be considered latent (i.e. representations of unobserved structures and mechanisms) until proven as observed. Models will be used in this Chapter to visually represent proposed: 1.

Causal flow

2.

Structures, mechanisms and outcomes

3.

Causal connections between variables (observed or latent).

Note: Formal Structural Equation Modelling (SEM) will not be discussed or illustrated.

343


8.4

Setting the Scene

8.4.1 Tentative Conceptual Framework

Figure 123: Tentative Conceptual Framework Stress emerged as a cause of both antenatal and postnatal depression with causes including: loss of control, lost expectation, isolation, economic hardship, marginalisation and infant temperament. The literature reviews identified empirical studies that suggested that antenatal stress and depression resulted in altered infant temperament which in turn was a cause of postnatal stress and depression. Other empirical studies indicated that antenatal maternal stress had effects on the foetus that potentially affected the life course. Similarly there were studies indicating a causal link between postnatal depression, poor maternal-infant attachment and detrimental impacts on child development. Support emerged as a protective factor against social marginalisation, isolation, stress and loss of control. That support was in turn influenced by family, social networks or social capital, social cohesion and social services. The quantitative studies supported the above emerging causal network but also indicated that ethnic diversity played a complex role in relation to social support. The macro role of globalisation was identified as impacting on economic and social marginalisation and maternal expectations.

344


8.4.2 Analytical Levels The above conceptual framework identifies structures and mechanisms operating at different levels. Examples of possible structures, mechanisms, contexts and outcomes are shown below based on the findings from the Emergent Theory Building. Levels

Example of Structures

Example of Mechanisms

Example of Contexts

Example of negative Outcomes

Global economic

Multinational companies

Exploitation, Profit

Labour market

Unemployment, migration

Cultural

Ethnic community

Segregation

Migration

Bonding networks

Social

Neighbourhood Relationships social capital

Social networks

Isolation

Social

Family

Emotional support

Absent partner

Isolation

Psychological

Self

Relation to self

Isolation

Feeling overwhelmed and alone

Psychological

Mind

Stress

Overwhelmed and alone

Depression, reduced motivation, anxiety

Biological

Body

Neurobiological

Reduced motivation

Hypoactivity of motivation areas

Table 56: Analytical Levels of Depression and Context

345


8.4.3 Abductive Theoretical Interpretations Abductive reasoning is central to the critical realist explanatory method. As discussed above one approach to abductive reasoning is to recontexualise or redescribe the phenomenon within a number of frames of interpretation or theories. The theoretical and conceptual frameworks selected for the analysis in this Chapter were chosen after considering a wide range of relevant social epidemiological, social, community psychology, public health, perinatal and postnatal depression theories including: social isolation and social exclusion theories, the stress process model (Pearlin, 1989), social network and social capital theories, theories of globalisation, psychosocial and social support theories, life-course theory, eco-social theory (Krieger, 2001), materialist and neomaterial theories, transactional theory, the ecological model for development (Bronfenbrenner, 1979), acculturation theory, integrated perinatal models, and theoretical models of neighbourhood effects. I have limited the analysis here to explaining causal mechanisms at the social level and have selected candidate theoretical and conceptual frameworks with the most promise for identifying the best explanation. Recontextualisation and redescription will be undertaken using the following theories or frames of interpretation: 1. Stress Process Model 2. Social Isolation Theory 3. Social Exclusion Theory 4. Social Services 5. Social Capital Theory 6. Acculturation Theory 7. Global-Economic Level Mechanisms. An overview of other theoretical frameworks explored will be provided in Section 8.12.

346


8.5

Stress Process Theory

8.5.1 Introduction Stress was identified as a phenomenon causing depression in both the qualitative studies and literature review. The stress process model was first described by Pearlin and colleagues (1981). The model is concerned with explaining ways in which social structure influence mental health with a focus on the connection between disadvantaged social status and psychopathology. In this section redescription and recontextualisation of perinatal stress, and depression, will be undertaken within the conceptual framework provided by the “Stress Process Model”. I will intentionally limit analysis to the social levels and will not discuss the role of infant temperament. According to the stress theory stressors occur either because of psychological characteristics of individuals or because of environmental factors over which the person has little control. The effects of these stressors may be to increase psychopathology, in increase psychological growth, or to lead to no permanent change in the individual (Sandler, Braver et al. 2000). Baron and Kenny (1986) proposed that these stressful experiences can be viewed from a causal perspective with different types of stress affecting each other in complex ways to lead to the development of psychological and physical health problems. Stress Theory is considered particularly applicable to the development of a theoretical model of postnatal depression and will be discussed further below.

8.5.2 Postulate I postulate that: 1. Antenatal Stress is a psychological level cause of biological changes to the pregnant mother and unborn foetus 2. Stress is a psychological level cause of antenatal and postnatal depression.

347


8.5.3 Comparative Analysis Triangulation Stress was identified early by experts interviewed as a cause of both antenatal and postnatal depression. While the term “stress” was only mentioned by one mother, I interpreted the coded category “loss of control” as expressing a similar concept. The phase two focused coding related a number of sub-codes to the concept “Stress” and it subsequently became a “core” category for the qualitative analysis. The Individual Level Qualitative Theory Generation (Section 4.6) confirmed stress as a strong contender as the proximal cause of depression. The quantitative studies were not able to directly measure stress as a variable. In the Individual Level Quantitative Theory Generation (Section 5.5) I abductively inferred that the latent variable F1 in all CFA Exploratory Specification Searches may be representative of stress or marginalisation. The loading of self-reported health on this latent variable gives support to the idea that this represents stress or something similar. Perinatal Depression Literature The subsequent literature review reported in Section 3.6 confirmed the central role that stress plays as a causative mechanism in the aetiology of perinatal depression and probably other adverse perinatal outcomes. As reported in that section there are many different types of stressful events that potentially have an effect on postnatal depression (marital relationship, housing, finances, unemployment, bereavement), and the influence of any stressor is related to the woman’s perception and appraisal of the event (Stein, Cooper et al. 1989; O'Hara 1995; Zelkowitz and Milet 1995; Lane, Keville et al. 1997). The NHMRC Systematic Review (NHMRC 2000) found evidence has been found linking postnatal depression with significant stresses and increased recent negative life events. Recent reviews of research at the neurobiological and psychological levels support the role that stress plays in the genesis of depression (Kinderman 2005; Stone, Lin et al. 2008).

348


The Stress Process Literature The stress process can be summarised as follows. An individuals’ location in the social structure has an associated inequality in resources, status, and power that differentially expose them to stressors which can damage their physical and/or psychological health. The damage may be moderated or lessened by their psychological resources and coping strategies which are socially patterned in ways that can leave members of disadvantaged groups more vulnerable to the harmful effects of stress (Pearlin 1989). Two major pathways linking social structure with stress are exclusion from full participation in the social system and participation that fails to provide the expected returns (Aneshensel 1992). Social and economic class, race and ethnicity, gender and age may result in the unequal distribution of resources and opportunities. Thus low status may itself be a source of stress (Pearlin 1989). Aneshensel (1992, p 18) argues that the “structural perspective on social causation understands stress both as a consequence of location in the social system and as a determinant of some outcome, most typically psychological distress. … Location in the social system influences the risk of encountering stressors, which in turn influences the chances of becoming emotionally distressed”. Hinkel (1987) notes that societies prescribe a variety of forms of behaviour, conditions and relationships that are proper for its members with sanctions for stepping outside these prescriptions acting as a form of stress. This may have implications for certain ethnic and cultural groups. Three types of moderators of the stress process are coping strategies, personal resources and social support. Coping strategies are behavioural or psychological state changes that people can make themselves. To make these coping strategies individuals need personal resources which may be either personal or social. Personal resources include notions of self-esteem and self-mastery. Social support has been consistently shown to have a strong buffering effect on stress outcomes. Social support literature is discussed in more detail below.

349


Source: (Turner 2010, p 5)

Figure 124: Modified Stress Process Model Abductive and Retroductive Analysis Recontexualisation of perinatal depression within a Stress Process Model provides an explanation of the causes of perinatal depression. External social or environmental and internal biological or psychological stressor mechanisms are able to “trigger” a causative mechanism called Stress at the psychological and biological levels that result in the phenomenon of depression. Stress is defined here as a “necessary” mechanism. The conditional “stressors” mechanisms may be event stressor such as poor health, childbirth experience, loss of status, loss of financial income, loss of control, unmet expectations, irritable infant and partner behaviour. Chronic stressors may also contribute and include: status strains (i.e. race, sole parent, poverty, and religion), role strains (i.e. wife, motherhood, and daughter), ambient strains (i.e. neighbourhood crime, services, physical decay), and quotidian strains (i.e. loneliness, daily hassle of sole parenthood). The Stress Process Model is able to explain the observed association of financial stress and infant temperament with depressive symptoms. Consideration of the transfactual makes it clear that stressor mechanisms may not result in depression if moderator mechanisms are active. Some of those moderators will be the counterfactual of the causative mechanisms identified above. Moderator mechanisms that have been described include coping resources, personal resources and social support. The Stress Process Model is thus able to explain the observed protective effect of practical, emotional and social support on depressive symptoms.

350


The Stress Process Model also explains that there may be more than one Stress

Outcome. These may be depression, anxiety, or drug/alcohol abuse. I consider here that “self-reported poor health” may be an observed variable associated with Stress Outcome.

8.5.4 Propositions A critical realist model of the mechanism stress is below. Conditions (other mechanisms)     

Isolation Loneliness Expectations Loss of control Support

Psychological

Outcomes

mechanisms



Depressive Symptoms

Stress

Anxiety Self Reported Poor Health

Figure 125: Critical Realist Model of the Stress Proposition Based on the above analysis the following propositions are made: 

Stress is the mechanism that causes depression when certain personal characteristics and contextual conditions exist



Stress is positively associated with depression when there is a past medical history of depression, difficult infant temperament including crying, financial stress, loss of expectations, a feeling of “loss of control”, isolation, and lack of support.

351


Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Assessment Hill’s aspects of association Strength Expectation and lack of support both have strong associations with depression Consistency The role of stress as a cause of depression has been found in different situations. The role of expectations and lack of support has also been found Specificity No specificity identified Temporality No temporality demonstrated in this study Biological Higher the causes of stress the higher the observed depression gradient Plausibility The association between stress and depression is biologically plausible Coherence The association is coherent with what is know Experimental Interventions that provide support have been demonstrated to reduce depression evidence Analogy There is an analogy between the effect of loss of expectation and loss of support. Both result in a similar effect. Thagard’s Principles Symmetry There is symmetry between stress causing depression and support preventing depression Explanation The stress proposition a) coheres with evidence on depression, b) evidence on role of support, isolation, loss of control, and c) is single proposition. Analogy Stress causing depression is coherent with stress causing anxiety and physiological changes to H-P axis Data priority Proposition describes observation re isolation, expectations, support. Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The stress proposition is coherent with the overall system of propositions Thagard’s Criteria Consilience The central role of stress as an explanation for depression explains the largest range of facts Simplicity Stress as a necessary cause of depression is a simple explanation Analogy

Table 57: Stress Proposition - Inference to Best Explanation In summary the stress proposition is an explanation of a “necessary” cause of depression. It explains much of the observed and known evidence but is insufficient on its own. The stress proposition could be used as the basis for building a more complete explanation.

352


8.6

Social Isolation Theory

8.6.1 Introduction Isolation and loneliness were identified as phenomenon associated with postnatal depression. In this section I have used abductive logic to redescribe and recontextualise the phenomena of isolation and loneliness associated with postnatal depression within the theoretical concept of social isolation. The central purpose in this section is to ascribe meaning to the phenomenon of postnatal depression within a social isolation conceptual framework.

8.6.2 Postulate I postulate that: 1. Maternal isolation is a psychological level mechanism causing maternal stress and postnatal depression 2. Social isolation is a social level mechanism causing maternal isolation.

8.6.3 Comparative Analysis Triangulation The possibility of social isolation as a “generative” mechanism of the phenomenon of postnatal depression emerged earlier in the exploratory phase. The literature review (Chapter 3) and individual level quantitative study (Chapter 5) both provided empirical evidence that lack of social support was a risk factor for postnatal depression. The qualitative study (Chapter 4) identified marginalisation and isolation as mechanisms that could result in stress and depression. Examination of the social worlds map drew attention to the situation that may be experienced by most mothers. Buried in the voices of the mothers were stories of “being alone” with the crying infant, with an absent partner, mother or other support friend”. At the group level (Chapter 6) social marginalisation and isolation emerged in relation to social support. One mothers group drew attention to the role of language and culture noting that “may be some people don’t speak English, we have a special language group … if they don’t know they just isolate…” This link between isolation and culture I will explore later. Social isolation was also linked to access to transport, financial situation, social networks and “bonding” and ”bridging” forms of social capital.

353


The group level quantitative study (Chapter 7) did not include specific measures of marginalisation and social isolation but did include the counterfactual of social support (i.e. no social support). The ecological studies (linear regression and spatial) found an association of aggregated postnatal depression with apartment living, entropy index, percent of one parent families, nurse visiting rates, smoking rates and aggregated “no social support”. When controlling for individual level factors the multilevel study paradoxically found protective associations with high density, poor social support, and poor schooling. The stratified multilevel studies found that these associations were strongest for mothers not born in Australia. Abductive analysis in Section 7.5 (Theory Generation) using the findings from the Factor Analysis suggested that weak bonding networks and high ethnic diversity protected mothers in those communities from depression. Recontextualisation of these multilevel findings within the social isolation conceptual framework suggests that weak bonding networks and high suburb ethnic diversity protect mothers from social isolation. The converse logic is that strong bonding networks in suburbs with low ethnic diversity contribute to social isolation. Postnatal Depression and Social Isolation Much of the relevant research on isolation and loneliness has been undertaken through qualitative studies. In a metasynthesis of 18 qualitative studies Beck (2002) identified four overarching themes: (a) incongruity between expectations and the reality of motherhood, (b) spiraling downward, (c) pervasive loss, and (d) making gains. For the theme “spiraling downward” Beck exported the sub-concepts of anxiety, overwhelmed, obsessive thinking, cognitive impairment, isolation and loneliness, guilt, and contemplating harming oneself. In her metasynthesis Beck (2002) found that mothers soon began a downward spiral of postpartum depression as their feelings worsened. All of the 18 studies included in her metasynthesis included aspects of this downward spiral. The feelings included: depression, sadness, anger, guilt, being overwhelmed, thought of harming themselves, anxiety and loneliness. Beck describes post-partum women as being enveloped in “unbearable loneliness due to the discomfort they felt being around others and their belief that no one else really understood what they were experiencing” (Beck 2002, p 464). Beck noted that “as mothers silenced themselves and withdrew socially they felt a profound sense of isolation and loneliness”. Beck (2002) also found that different cultural contexts influenced the

354


degrees of isolation and loneliness. Hmong mothers living in the United States received a high level of family support for a 30 day protected period and did not experience the isolation and loneliness that Jordanian mothers living in Australia suffered (Beck 2002: 464). There have been few quantitative studies that have separately assessed the role of “isolation” or “social isolation”. Birkeland et al (2005) in their study of adolescent motherhood and postpartum depression identified social isolation, maternal self-efficacy and weight/shape disturbance as being significantly predictive of depression. In their study social isolation had been a subscale of the Parental Stress Index (Abidin 1995). Hatton and colleagues (2007) in a study of antenatal depression among high risk women included a specific question on social isolation together with EDS, the Sarason Life Events Survey and Multidimensional Scale of Perceived Social Support. Social isolation (r = 0.636, p 12 the Index of Concentrated Extremes (ICE), Index of Relative Social Disadvantage (IRSD), % Poor and Factor 1 improved the model but were not significant. In the stratified multilevel analysis measures of social exclusion did not improve the model for mothers born in Australia but for those mothers not born in Australia ICE, IRSD, % Occupational Class 3 and % unemployed improved the model for EDS >9. Paradoxically a high percentage of occupation class 3 and unemployment protected against depression. None of the measures of social exclusion were significant in the non born in Australia EDS > 12 analysis. The ecological studies (linear regression and spatial) did not find a strong association of aggregated postnatal depression with measures of social exclusion. The exception was apartment living. Similarly when controlling for individual level factors (including financial difficulties) the multilevel studies only found a significant group level effect in the “not born in Australia” EDS > 9 study. The quantitative findings suggest that social exclusion is operating predominantly at the individual family level with the exception of mothers not born in Australia where it is impacting on less severe depressive symptomatology (EDS > 9). For mothers not born in Australia living in communities with extremes of wealth is detrimental to mental health, while living in communities with a high percentage of labourers, clerical workers and unemployed is protective. An alternative explanation is that social exclusion is operating distal to social support in a causal chain. Consequently both the spatial and multilevel regressions will control for social exclusion and the only significant associations will be with the more proximal social support measures.

360


Aneshensel (2009, p 387) in a discussion of the stress process model notes that “more commonly used analytic techniques typically combine all antecedent variables into one equation, so that only the direct effects of each variable on the outcome of interest are estimated. Of particular interest are multivariate analyses that find that distal factors, such as income, are not statistically significant when more proximal factors, such as social support, are taken into consideration”. Such findings can be interpreted to mean that the distal factors are unimportant to the health outcome. Mediational analysis or structural equation modelling may be necessary to fully account for the impact of social exclusion on depressive symptoms. Statistical examination of this possibility is beyond the scope of this study but the possibility will be considered in the analysis below. Recontextualisation of these multilevel findings within the social exclusion conceptual framework suggests that for migrant mothers, social relations (and possible bridging networks) are stronger in working class communities and protect against maternal depression. The converse logic is that, for migrant mothers, there is more social exclusion than might be expected, in suburbs with extremes between rich and poor. Interestingly these findings were not apparent for mothers born in Australia. Perinatal Depression literature As discussed in Section 3.4 women of low socio-economic status have consistently been found to have higher rates of antenatal and postnatal depression (Zajicek 1981; Cox, Connor et al. 1982; O'Hara, Neunaber et al. 1984; Gotlib, Whiffen et al. 1991; O'Hara 1995; O'Hara and Swain 1996; Seguin, Potvin et al. 1999; Beck 2001; Bennett, Einarson et al. 2004). Beck did not find SES to be significant in her first meta-analysis (Beck 1996) but added both SES and unplanned pregnancy in her update (Beck 2001). The relationship between SES and postpartum depression, in the 2001 published metaanalysis, was in the range of a small effect size (0.19 ~ 0.22). There is increasing evidence for contextual effects of area-level economic deprivation on mental health (Weich, Blanchard et al. 2002; Weich, Twigg et al. 2003; Skapinakis, Lewis et al. 2005; Fone and Dunstan 2006; Fone, Dunstan et al. 2007; Fone, Lloyd et al. 2007). A qualitative study of pathways from neighbourhoods to mental well-being found that neighbourhood affordability, negative community factors including crime and vandalism,

361


and social makeup including unemployment and poverty, were felt to be associated with poor mental well being (O'Campo, Salmon et al. 2009). At the neighbourhood level, Nager and colleagues (2006) found that Swedish women living in poor neighbourhoods were at significant risk of first hospital admissions for postpartum psychosis than women in the wealthiest neighbourhoods. A Cox Regression Analysis was used that did not control for income at the individual level. Thus it is not clear whether the effect was the result of low income at the individual level or neighbourhood level. While postpartum psychosis is a different clinical condition to the depressive symptoms analysed in this study, the study by Nager is the only other study I have found that considers the impact of neighbourhood level poverty on postnatal depressive conditions. Social exclusion literature The definition of social exclusion remains contested but there is a common “understanding that social exclusion is not only about material poverty and lack of material resources, but also about the processes by which some individuals and groups become marginalised in society” (Millar 2007, p 2). While there is an overlap with the concept of social isolation, it is used here primarily in relation to access to resources. Millar cites five key factors of social exclusion proposed by Room (1995), three main elements proposed by Atkinson (1998) and a consensus five attributes proposed by Tsakloglou and Papadopoulos (2002). The consensus proposed by Tsakloglou and Papadopoulos included:

1. “Multidimensional: across a wide range of indicators of living standards, including neighbourhood and community resources 2. Dynamic: it relates not just to the current situation but also to prospects for the future 3. Relative: it implies exclusion in a particular society at a particular time 4. Agency: it lies beyond the narrow responsibility of the individual 5. Relational: meaning a major discontinuity with the rest of society.” (Tsakloglou and Papadopoulos (2002) cited by Millar (2007, p 3) In the above framework of indicators Tsakloglou and Papadopoulos (2002) grouped the indicators into measures of income poverty, living conditions, necessities of life and social relations. The measures of social relations included meeting friends, talking to neighbours and membership of clubs or groups. It is worth noting here that the inclusion of measures

362


of social relations will result in an overlap, or correlation, with measures of social isolation, social cohesion, social support and social capital that are discussed separately. Millar (2007, p 6) notes that measures such as social isolation can be difficult to interpret “because the way we engage with our neighbours, or with political institutions, will reflect cultural norms, as well as national, regional and ethnic identities”. Saunders (2003) undertook a study using the 1998/99 Household Expenditure Survey (HES) to assess the level of social exclusion in Australia. Saunders studied the following major forms of social exclusion: social interaction, domestic deprivation and extreme consumption hardship. He found that the major form of social exclusion was lack of social interaction which he measured at twice the level of domestic deprivation and four times the level of extreme consumption hardship. Saunders also found that sole parents were the most excluded group. In a comparative study between Australia and Britain, Sanders and Adelman (2005) found that poverty was higher in Britain but levels of material deprivation and social exclusion were higher in Australia, with sole parents the most affected in both countries. In those studies (Saunders 2003; Saunders and Adelman 2005) lack of social interaction was measured as: “cannot afford a week’s holiday away from home once a year”, “cannot afford a night out once a fortnight”, and “cannot afford to have friends/family over for a meal once a month”. The finding by Saunders that lack of social interaction was the major form of social exclusion in the Australian setting is consistent with the emerging findings from the study. Social exclusion may be impacting on mothers through social isolation and social support rather than material deprivation. Abductive and Retroduction Analysis Redescribing poverty and marginalisation within a social exclusion conceptual framework provides explanation for a possible causative mechanism. Social exclusion is a contested notion about the process by which some individuals and groups become marginalised from society (Millar 2007). I have abstracted that social exclusion may impact on mothers through social isolation and stress. While there may be many causes of social isolation social exclusion is clearly an important social level mechanism. Structures within the social level that may generate social exclusion as a causative mechanism include: global markets, big business, political structures, dominant culture, government agencies, religious organisations, and capitalist structures. Candidate conditional mechanisms may include: income support policy, ethnic segregation or diversity, racial tolerance, religious tolerance, public transport, and social cohesion.

363


As noted above, recontextualisation of paradoxical multilevel findings within the social exclusion conceptual framework suggests that for migrant mothers, social relations (and possible bridging networks) are stronger in working class communities and protect against maternal depression. The converse logic was that, for migrant mothers, there is more social exclusion than might be expected, in suburbs with extremes between rich and poor. Interestingly these findings were not apparent for mothers born in Australia. Based on the triangulation, literature review and retroductive thought analysis I propose that the following conditional mechanisms may influence the tendency of social exclusion to cause social isolation, stress and depression: 1. weak bridging networks or social cohesion 2. segregation of social and ethnic groups and strong bonding networks 3. poor access to services including home visiting nursing services.

8.7.3 Proposition A critical realist model of the social level mechanism of social isolation is below: Conditions (other mechanisms)    

Weak bridging networks Weak social cohesion Segregation of ethnic minority Poor access to services.

Social level

Outcomes

mechanisms Social exclusion

Stress Social isolation

Figure 127: Critical Realist Model of Social Exclusion Proposition Proposition: Social exclusion is a social level mechanism that causes maternal stress and isolation when certain personal characteristics and contextual conditions exist

364


Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Application Hill’s aspects of association Strength Financial stress has a strong assoc with depression at the individual level but ecological measures of poverty and social exclusion had weak association Consistency The poor income has been found as a cause of depression in large number of studies. Specificity No specificity identified Temporality No temporality demonstrated in this study Biological Higher the lack of financial stress the higher the observed depression gradient Plausibility The association between social exclusion and stress is biologically plausible Coherence The association is coherent with what is know Experimental No experimental evidence identified evidence Analogy There is an analogy between the effect of social exclusion and isolation. Both result in a similar effect. Thagard’s Principles Symmetry There is symmetry between social exclusion causing depression and social support and practical support preventing depression Explanation The social exclusion proposition a) coheres with evidence on depression, b) coheres with other propositions and c) is a single proposition. Analogy Social exclusion causing stress is coherent with isolation causing depression mediated through stress Data priority The proposition describes the observation re financial stress and depression Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The social exclusion proposition is coherent with the overall system of propositions Thagard’s Criteria Consilience Social exclusion explains a range of facts but not all known facts Simplicity Social exclusion is not sufficient to cause of depression. Not the most simple explanation Analogy Social exclusion causing stress in mothers is analogous to social exclusion causing poor self report health

Table 59: Social Exclusion proposition – Inference to Best Explanation In summary the social exclusion proposition is not the best explanation. It explains some of the observed and known evidence but is insufficient on its own.

365


8.8

Social Services

8.8.1 Introduction Access to services was identified by mothers as a phenomenon associated with protecting mothers from postnatal depression. The buffering impact of delivery of social services on maternal (and infant) psycho-social and psychological processes is reviewed by Hutchinson and colleagues (2007). In this section I have used abductive logic to redescribe and recontexualise the phenomena of access to services associated with postnatal depression within the policy level concept of delivery of social services. The central purpose in this section is to ascribe meaning to the phenomenon of postnatal depression within a social services conceptual framework. Postulate I postulate that delivery of social services is a policy level mechanism causing reduced social exclusion, social isolation and stress.

8.8.2 Comparative Analysis Triangulation The possibility that social services may play a protective role in the phenomenon of postnatal depression emerged in the individual level qualitative study (Chapter 4) where mothers spoke of the supportive role of midwives, nurses, mother’s groups and phone calls. The literature review (Chapter 3) did not identify lack of social services as a risk factor for postnatal depression. There is evidence that service interventions can play a role in treating postnatal depression (Dennis 2005; Morrell, Warner et al. 2009). The individual level quantitative study (Chapter 5) did not include variables that measured levels of social service delivery. Any additional nurse visits would have occurred after the EDS was measured. Group level theoretical and empirical literature (Chapter 6) suggests that the social service environment may play an important role in influencing perinatal outcomes (Culhane and Elo 2005). The group level qualitative study (Chapter 6) identified “access to services” and “supportive social policy” as important themes. The situational analysis drew these together into a combined services planning, delivery and social policy arena. The group level quantitative study (Chapter 7) included two measures of early childhood nursing support. Analytical findings were mixed. The Exploratory Factor Analysis loaded both the variables on Factor 5 together with % poor health and % different address in 5

366


years. Spatial visualisation suggested an association between nursing visiting rates and aggregated EDS but the bivariate analysis, while statistically significant, had low R Squared. Bayesian bivariate analysis was not significant. In the ecological linear regressions “Nurse Visit Rate %” was significant for both EDS >9 and EDS > 12 but did not contribute to improving the Bayesian ecological spatial models. In the Bayesian multilevel spatial studies nursing visiting rate improved, but not significantly, the EDS > 9 overall and not-born in Australia studies. Factor 5 also improved the same models but was not significant. The quantitative studies had a limited set of variables for measuring service access. Visualisation showed marked spatial differences in nursing service delivery which correlated with the distribution of aggregated EDS> 9 and EDS >12. The analytical studies also suggested a possible contribution to both the aggregated and individual level EDS rates. Recontextualisation of these multilevel findings within the social services policy conceptual framework suggests that services policy and resourcing may contribute to protecting against maternal depression through social and psycho-social level mechanisms. Perinatal Depression Literature There is limited evidence that service level interventions will prevent the risk of postnatal depression. Dennis (2005) undertook a systematic review of all published and unpublished randomised controlled trials of preventive psychosocial and psychological interventions in which the primary or secondary aim was a reduction in the risk of postnatal depression. All the trials recruited pregnant women or new mothers less than six weeks postpartum. Of the fifteen trials assessed, the only intervention to have a clear preventative effect was intensive postpartum support provided by a health professional. It was notable that individually based interventions were more effective than group based interventions. The efficacy of home visiting for postnatal depression has recently been confirmed (Morrell, Warner et al. 2009). Social support networks were protective in this study suggesting that antenatal interventions that promote friendship groups may be beneficial. The role of antenatal groups in preventing postnatal depression has not yet been confirmed (Austin 2003). More recent studies have, however, demonstrated the effectiveness of support-focused interventions by social services. Dennis and colleagues (2009) undertook a multisite

367


randomised controlled trial where the intervention was proactive individualised telephone based peer (mother to mother) support, provided by a volunteer recruited from the community who had previously experienced and recovered from self reported postnatal depression and who had attended a four hour training session. Relationship-based community health worker support through phone and face to face contact was demonstrated, in a randomised control trial, to alleviate depressive symptoms in Medicaid eligible women compared to controls. The effect was greater for women at higher risk (Roman, Gardiner et al. 2009). Broader social service interventions that are often described as “joined-up Government” may also have benefits. “Joined up” interagency approaches to improving child and family health, such as UK Sure Start initiative have had mixed results (Rutter 2006). The recent study by Mulhuish and colleagues (2008) however, found promising outcomes not identified in earlier evaluations. While these outcomes were among children, the interventions were family focused and many commenced during pregnancy and are thus relevant to this analysis.

368


Abductive and Retroduction Analysis The redescription of maternal depression within a services conceptual framework focuses analysis on the service structures that may generate causative mechanisms. That analysis identifies a complex set of structures and mechanisms that is beyond the scope of this study. For example, service delivery arises from political, public service, health service, business and economic structures. Simplistically social service mechanisms may provide interventions that reduce social exclusion, social isolation, stress and depression. Those mechanisms arise at macro and meso levels but may impact directly on biological and psychological levels. For illustration I have included the following conditional mechanisms in the model. 1. Political support for social service interventions 2. Public Service Policy level support for social service interventions 3. Operation level support for effective evidence-based social service interventions.

8.8.3 Proposition Conditions (other mechanisms)   

.Political Support for Social Service Interventions Public Service Policy Support Operational Support for effective evidence-based interventions

Social Service mechanisms

Social Service Interventions

Outcome

Reduced Stress Increased support

Figure 128: Critical Realist Model of Social Service Proposition Proposition: Social Services Delivery is a policy level mechanism that protects against maternal stress and isolation when certain other personal characteristics and contextual conditions exist.

369


Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Application Hill’s aspects of association Strength Social services were identified by QUAL but has a weak association with depression at the ecological level. No individual level data available. Consistency The Social services have not been well studied as protective also intervention studies show some effect Specificity No specificity identified Temporality No temporality demonstrated in this study Biological Limited information available gradient Plausibility The association between Social Services and stress is plausible Coherence The association is coherent with what is know Experimental There is experimental evidence that service intervention can reduce depression evidence Analogy There is an analogy between of social service support with social support and practical providing buffering Thagard’s Principles Symmetry There is symmetry between social services buffering stress and social support and practical support preventing depression Explanation The social service proposition a) coheres with evidence on depression, b) coheres with other propositions and c) is not a single proposition. Analogy Social services buffering stress is coherent with social support buffering stress Data priority The proposition describes the observation re ecological association. Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The social service proposition is coherent with the overall system of propositions Thagard’s Criteria Consilience Social service proposition explains a limited range of known facts Simplicity Social service proposition is not sufficient to protect from depression. Not the most simple explanation Analogy Social service buffering stress in mothers is analogous to social support buffering stress and depression

Table 60: Social Service proposition – Inference to Best Explanation In summary the social service proposition is not the best explanation. It explains some of the observed and known evidence but is insufficient on its own.

370


8.9

Social Capital Theory

8.9.1 Introduction Maternal emotional, practical and social support networks were identified as phenomenon protecting mothers from postnatal depression. In Chapters 4, 5 and this section I have used abductive logic to redescribe and recontextualise the phenomena of support associated with postnatal depression within the psycho-social and social level concepts of social networks and social capital. The central purpose in this section is to ascribe meaning to the phenomenon of postnatal depression within a social capital conceptual framework. Kawachi and colleagues (2008, p 3) argue that social capital has both individual and group attributes and that it can be conceptualised as both social cohesion and as resources embedded in networks. The idea that social capital has both individual and group attributes is consistent with the findings in this study which has found network-based resources to be important at both the individual and group levels. The authors also observe that there is currently contention as to whether social capital should be conceptualised as social cohesion or as resources imbedded in social networks. Moore and colleagues (2006, p 729) p 729 advanced the argument that network approaches to social capital were lost when the concept was translated into public health. The authors argue that social capital was originally viewed within public health as a psychosocial mechanism operating at an ecological level that might mediate the income inequality-health pathway. This led to a dominance of what the authors call a “communitarian” approach to social capital with “disproportionate attention to normative and associational properties of places”. It is argued that network approaches to social capital should be recovered in order to enable the full translation and conceptualisation of social capital in public health [and epidemiology]. Hulse and Stone (2007), writing from an Australian policy perspective, reviewed the use of the terms social cohesion, social capital and social exclusion across North American, European and Australasian jurisdictions. In their account they identify at least three dimensions of social cohesion: 1)

Social relations of everyday life, family and social relationships, networks, and voluntary social processes (social capital)

371


2)

Reduction of differences, cleavages and inequalities between groups of people and people living in different geographical areas (social exclusion)

3)

A distinct cultural dimension referring to “ties that bond people together with a sense of common purpose, shared identity and common values (social cohesion).

This debate is well articulated by Carpiano (2006; 2008) who compares two theories of social capital as advanced by Putnam (1993; 1995; 2000) and Bourdieu (1986; 1992). Carpiano (2008) argues for the separation of what he calls the: 1) structural antecedent factors (i.e. socio-economic conditions, residential stability, income inequality), 2) social cohesion (connectedness and values), 3) social capital (social support, social leverage, information social control, neighbourhood organisation participation), and 4) social capital outcomes.

Source: (Carpiano 2006, p 169)

Figure 129: Conceptual model of neighbourhood social capital processes Of relevance to the analysis in this Chapter are the bonding and bridging forms of social capital. Kawachi and colleagues (2008, p 5) observe that “regardless of whether one subscribes to the social cohesion school of social capital or the network school, consensus now exists about the importance of distinguishing between bonding and bridging social capital”. Bonding social capital refers to resources that can be accessed

372


within social groups whose members are alike in terms of their social identity. The term “bridging capital” is used to describe the process whereby resources are accessed by individuals and groups through their connections that cross class, race, cultural and other boundaries of social identity. As discussed elsewhere in this Chapter, bonding capital may have detrimental effects and a key to improving health may be increasing access to resources outside of immediate social milieu (Kawachi, Subramanian et al. 2008). Postulate I postulate that: 1. Social capital is a social level mechanism protecting mothers from stress 2. Social capital is a social level mechanism protecting mothers from economic, and social exclusion.

8.9.2 Comparative Analysis Triangulation of findings The possibility of social capital as a [generative mechanism] of the phenomenon of postnatal depression emerged as the concept of “support and nurturing” of mothers in the individual level qualitative study (Chapter 4) and the subsequent group level qualitative study (Chapter 7). The individual level literature review (Chapters 3) and quantitative study (Chapters 5) provided empirical evidence that social relationships are protective for maternal postnatal depression. The group level findings were paradoxical and suggest that ethnic diversity is playing a significant role. The individual level qualitative study (Chapter 4) identified “support for mothers” as a mechanism that could protect mothers from stress and depression. Also identified were partner support, family support, support of a “mum type” person, social networks, and mothers groups. The group level qualitative study (Chapter 6) identified social capital as a strong theme and the situational analysis confirmed the importance of the linked concepts of social support networks, social cohesion and social capital in protecting mothers from depression. The individual level logistic regression (Chapter 5) identified measures of emotional support, practical support and social support as being significantly associated with EDS > 9 and EDS > 12. The initial exploratory Categorical Principal Component Analysis loaded these variables on the same dimension as “self report health” and “country of birth”. In the four factor specification search, Factor 3 loaded: social support, emotional support and

373


practical support, and Factor 2 loaded: not born in Australia and social support. The emerging model (Figure 44) proposed that an individual level latent variable related to support networks was protective while latent variables related to ethnicity and social isolation was causative of maternal depression. The group level quantitative study (Chapter 7) included a number of indicators of social capital. In relation to Social Capital, the Exploratory Factor Analysis loaded the following variables on Factor 2: Volunteerism, Low lack of schooling, high entropy, low percentage of Class 3, High IRSD, low density, high nurse visiting rate; and Factor 6: low lack of schooling, low % no social support, low % no practical support, low density, high different address in past five years and high Maly index. Both these latent variables included measures of ethnic diversity. Ecological visualisation and bivariate analysis found associations between most measures of social networks, social cohesion and social capital and the aggregated and spatial distribution of postnatal depression. Bayesian bivariate analysis was consistent except for measures of change of address. In both the ecological likelihood linear and Bayesian spatial regressions “% No Support” and “Entropy” were significant for EDS >9 and EDS > 12. The Bayesian multi-level spatial analysis found for EDS > 12 that density, the Maly Index and Factor 2 improved the model but were not significant. Low % no support, low % poor schooling and Factor 6 improved the model and were significant. Factor 6 significantly improved the Bayesian multi-level model reducing the DIC from 5215 to 5201. In the stratified multilevel analysis measures of social capital did not improve the model for mothers born in Australia but for those mothers not born in Australia ICE, Maly Index, % no support and % poor schooling improved the model for EDS >9 and EDS > 12. These findings are paradoxical suggesting that low group level social capital may be protective when controlling for individual level measures of support. The ecological studies (linear regression and spatial) found a strong association of aggregated postnatal depression with aggregated measures of social support. When controlling for individual level factors (including social support) the multilevel studies found a significant protective group level effect of “no social support”. In the stratified studies this finding was not found for mothers born in Australia. For mothers not born in Australia this finding was associated with a group level protective effect from increased neighbourhood ethnic diversity (Maly Index).

374


The quantitative findings suggest that social capital is operating predominantly at the individual family level with the exception of mothers not born in Australia. For mothers not born in Australia living in communities with high levels of social capital (social networks) is detrimental to mental health. Recontextualisation of these multilevel findings within the social capital conceptual framework suggests that for migrant mothers, social relations (and possible bridging networks) are stronger in ethnically diverse communities and protect against maternal depression. The converse logic is that, for migrant mothers, there is more social support than might be expected, in suburbs with low aggregated social support and less social support in suburbs with high aggregated social support. If aggregated social support represents strong bonding networks then these findings suggest that migrant mothers may be socially isolated in communities where they are the minority. This is supported by the high rates of depressive symptoms among mothers not born in Australia (Figure 98) present in the Southern suburbs were the indices of ethnic diversity are low. Perinatal Depression Literature There is limited literature in relation to perinatal depression and social capital of the “social cohesion school”. There is, however, extensive empirical evidence for the role of the social support network component of social capital as described by Carpiano (2008). Social support has also been found to be protective of a wide range of other adverse perinatal outcomes(Orr 2004). Social support is a multidimensional concept. Sources of support can be a spouse, relatives, friends, or associates. There are also different types of social support, e.g.,

informational support (where advice and guidance is given), instrumental support (practical help in terms of material aid or assistance with tasks), and emotional support (expressions of caring and esteem). Emotional support can be defined as relationships that make the individual feel loved, appreciated, and valued. Instrumental support has been described as tangible assistance and includes support available to the individual from members of his or her family or others to help with specific, concrete needs, such as lending money, giving a ride, helping with childcare etc (Orr 2004). Social support and its counterfactual, lack of social support, have consistently been found to be associated with perinatal depressive symptoms (Zajicek 1981; Cox, Connor et al. 1982; O'Hara, Neunaber et al. 1984; Gotlib, Whiffen et al. 1991; O'Hara 1995; O'Hara and

375


Swain 1996; Seguin, Potvin et al. 1999; Beck 2001; Bennett, Einarson et al. 2004). In her most recent systematic review of predictors of postnatal depression Beck (2001) identified 27 studies that examined social support. The relationship between social support and postpartum depression had a moderate effect size. Studies have consistently shown a negative correlation between postpartum depression and emotional and instrumental support during pregnancy (Menaghann 1990; Richman, Raskin et al. 1991; Beck 1996; O'Hara and Swain 1996; Seguin, Potvin et al. 1999; Ritter, Hobfoll et al. 2000). Two recent studies have found that perceived social isolation (or lack of social support) during pregnancy was a strong risk factor for depressive symptoms postpartum (Seguin, Potvin et al. 1999; Forman, Videbech et al. 2000). These findings suggest that women who do not receive good social support during pregnancy are more likely to develop postpartum depression. This concept was confirmed in a recent study that argued that receiving informational support from a large number of social network members was protective against postpartum depression (Seguin, Potvin et al. 1999). It should be noted, however, that researchers have consistently found differences between depressed women’s perceptions of social support, and the amount of support they objectively received (Logsdon, Birkimer et al. 2000). These differences may be accounted for, in part, by the fact that depressed individuals tend to view everything more negatively, including the level of support they perceive. Immigrants can face increased stressors related to discrimination or the stress of adjusting to a new culture. Social support might be particularly relevant in that context. Surkan and others (2006) found that social support, and social networks were independently related to depressive symptomatology in a multiethnic sample of women having recently given birth. I will explore this further in the next section. Ritter (2000) cites House et al. (1988) as noting that social support may have a more positive direct effect on health than stress has a negative effect. They also note that social support has been found to limit the negative psychological effects of chronic life stress. In the context of pregnancy, social support has been found to have a positive effect on psychological well-being even among low income women. Ritter and colleagues (2000, p 577) argue that “because of the relative lack of financial resources, low-income pregnant women may be particularly reliant on social support, because they may become less likely to work and have greater needs for emotional and emotional assistance”. They note, however, that this positive effect may not apply to inner-city or ethnic minority women

376


where “women's social resources may already be stretched owing to chronic stressful conditions and may no longer be available during stressful periods” (Ritter, Hobfoll et al. 2000, p 577). Ritter and colleagues examined the prospective influence of stress, self-esteem and social support on postnatal depressive symptoms of 191 inner-city women. They found that stress was related to increased depression and that greater income and social support were related to decreased depressions. They found no differences between African American women and European women in terms of depression or how they were impacted by stress and psychosocial resources (Ritter, Hobfoll et al. 2000). Social Capital and Mental Health Literature The impact of social capital on mental health has been recently reviewed by Almedom and Glandon (2005; 2008). Those reviews found that social capital could be both an “asset and a liability with respect to the mental health of those in receipt of and those providing services and other interventions”. The review included both individual level and ecological level studies and none of the studies reviewed were in relation to perinatal mental health. De Silva and colleagues (2005) in a systematic review of 21 studies suggested that individual and ecological studies of social capital may measure different aspects of the social environment. They concluded that current evidence is inadequate to inform the development of specific social capital intervention to combat mental illness. Their review included 7 ecological (6 multilevel) studies none of which related specifically to perinatal maternal mental health. With respect to the ecological studies there was no clear pattern of association between ecological social capital and mental illness. In particular there was no clear evidence of an inverse relationship between the level of social capital and mental illness. In a study of neighbourhood deterioration and depression Kruger and colleagues (2007) found that the effects of residential deterioration were mediated by social contact, social capital and fear of crime. The study was not multi-level and defined “neighbourhood” as the geographic buffer around each individual respondent’s home. The authors undertook a multi-level analysis at Census tract level and found low intra-tract variance with most variance occurring at the individual level. Recent multi-level studies lend some support to the hypothesis that ecological level social capital may protect against mental health disorders. A UK multilevel study by Fone (2007, p 343) found that “small-area level income deprivation and low levels of social cohesion

377


were independently associated with poor mental health” and that “high levels of social cohesion attenuated the association between mental health and income deprivation”. In another UK multilevel study Stafford and colleagues (Stafford, De Silva et al. 2008) found no evidence of a main effect of social capital on mental health. For people living in deprived circumstances, however, an association between social capital (contact amongst local friends) and lower reporting of common mental disorders was found. Of relevance to the findings of this study was the finding that elements of bonding social capital were associated with higher reporting of common mental disorders. This finding was only in deprived households. Bonding capital may represent an important survival mechanism for residents of disadvantaged communities (Kawachi, Subramanian et al. 2008). Kawachi and colleagues cite Carol Stack’s (1974) ethnographic study of a poor African-American community which revealed high levels of mutual support from kinship networks as the primary mechanism for “getting by”. Kawachi and colleagues (2008, p 6) observe that such “bonding capital often extracts a cost to the providers of support in terms of the mental and financial strain of caring for others in need”. Kawachi and colleagues cite other studies that also (Caughy, O'Campo et al. 2003; Ziersch and Baum 2004) suggest that strong bonding ties within disadvantaged communities may be a detriment to the health of residents. The authors conclude that “the emerging picture from these studies seems to be that bonding capital within disadvantaged communities may be a health liability rather than a force for health promotion that it is often assumed to be. The key to improving health therefore appears to lie in residents’ ability to access resources outside their immediate social milieu, i.e. access to bridging social capital” (Kawachi, Subramanian et al. 2008). Abductive and Retroduction Analysis The redescription of maternal depression within a social capital conceptual framework focuses analysis on social network structures that may generate causative mechanisms. Those mechanisms may be protective or detrimental depending upon the context. In this study emotional and practical support and social networks all had protective effects at the individual level. Put simplistically these modes of support may reduce the effects of social exclusion, social isolation, stress and depression. The social capital mechanisms may arise at macro and meso levels but are impacting directly on the individual psychological level.

378


The outcome of this social capital is usually a reduction in stress but in some circumstances there may be an increase in stress. The forms of social capital are complex and cannot be reduced to a single construct. I have therefore considered that bonding, bridging and linking social ties should be considered as separate social capital mechanisms. The contextual conditions are poorly understood but based on the findings in this study and the above literature I have included the following conditional mechanisms in the model: 1. Class and social position 2. Poverty and social exclusion 3. Crime and community safety 4. Neighbourhood decay 5. Population density 6. Social service support.

8.9.3 Proposition A critical realist model of the social capital mechanism is below. Conditions (other mechanisms)      

Class/position Poverty and social exclusion Crime and community safety Neighbourhood decay Population density Social service support

Social Capital

Outcome

mechanisms Social networks

Reduced Stress Increased support

Figure 130: Critical Realist Model of Social Capital Proposition Based on the above analysis the following proposition is made: Social capital (social networks) is a social level mechanism that protects against maternal stress and isolation when certain other personal characteristics and contextual conditions exist.

379


Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Application Hill’s aspects of association Strength Social Capital (social networks) was identified by QUAL and QUANT as associated with depression at the individual and ecological levels. Consistency Social Capital (social networks) and social support has been previously identified as associated with maternal depression Specificity Postnatal depression is strongly linked to lack of support Temporality No temporality demonstrated in this study Biological There was a gradient at the individual and ecological level gradient Plausibility The association between social networks with stress is plausible Coherence The association is coherent with what is know Experimental There is experimental evidence that home visiting and other support reduced evidence maternal depression Analogy There is an analogy between of social network support and the benefits of emotional and practical support all reducing stress and depression Thagard’s Principles Symmetry There is symmetry between social capital reducing stress and social isolation causing stress Explanation The social capital propositions a) coheres with evidence on depression, b) coheres with other propositions and c) is not a single proposition. Analogy There is an analogy between of social network support and the benefits of emotional and practical support all reducing stress and depression Data priority The proposition describes the data observations. Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The social capital propositions are coherent with the overall system of propositions Thagard’s Criteria Consilience Social capital propositions explain a limited range of known facts Simplicity Social capital propositions are not sufficient to explain depression. Not the most simple explanation Analogy There is an analogy between of social network support and the benefits of emotional and practical support all reducing stress and depression

Table 61: Social Capital proposition – Inference to Best Explanation In summary the social capital proposition is not the best explanation. It explains some of the observed and known evidence but is insufficient on its own.

380


8.10 Acculturation Theory 8.10.1

Introduction

The phenomenon of postnatal depression was associated at an individual and ecological level with mothers not being born in Australia and ethnic segregation respectively. A number of abstract or conceptual frameworks might be used to redescribe or recontexualise these findings including: racism, racial or ethnic segregation and diversity, cultural theory, migration and acculturation theory. Colonisation and waves of migration over 200 years provide a historical backdrop to this study. It is within this context that I have elected to ascribe meaning to these phenomena within an acculturation2 conceptual framework. I might have used segregation as a conceptual framework but considered Acculturation included segregation and would allow for a broader recontextualisation of the phenomena identified. The earlier discussion of stress, social exclusion, social isolation, and social capital are also relevant here. An eco-cultural framework described by Berry and colleagues (2002) attempted to link cultural and ecological factors to psychological outcomes. The authors described the antecedent mechanisms as related to population-level characteristics such as the ecological and socio-political context. The outcomes described were observable behaviours and psychological characteristics and the intervening processes were assumed to be genetic transmission, cultural transmission, ecological influences and acculturation (Van de Vijver, Van Hemert et al. 2008). Acculturation has a broad range of meanings including the socialization of children to the norms of their own culture. Here I will use it to mean the modification of the culture of a group or individual as a result of contact with a different culture. A contribution from the field of acculturation research describes acculturation outcomes as including psychological well-being (e.g. distress and mood states), socio-cultural competence in ethnic culture and socio-cultural competence in the mainstream culture (Arends-Toth, Van de Vijver et al. 2006) cited by (Van de Vijver, Van Hemert et al. 2008).

2

Note: I am using “Acculturation” here as a conceptual or theoretical framework for

recontextualisation of phenomena described. I am not analysing the process “acculturation” which may, or may not be associated with postnatal depression.

381


Nauck (2008) observes that acculturation is by nature a multilevel issue and that the most widely accepted approach to acculturation in cross-cultural psychology is that of Berry (1997).

Source: Berry (1997, p 15) cited by Nauck (2008, p 381)

Figure 131: A framework of acculturation research (Berry 1997) The model proposes four individual level outcomes of the acculturation process. They are: integration, assimilation, separation and marginalisation. Nauck (2008) is critical of the above model as a multilevel framework arguing that a ”full-fledged multilevel study of acculturation … should include group dynamics between immigrants and the population majority”. He then proposes that social contexts influence individual level behaviour through: social structures that provide opportunities and constraints, social control within social networks, places of cultural transmission, migration goals and context specific identification (e.g. within ethnically homogenous peer groups). Thus of direct relevance to understanding acculturation are ecological level studies of racial and ethnic integration or segregation as undertaken here.

382


8.10.2

Comparative Analysis

Triangulation The possibility of acculturation as a [generative mechanism] of the phenomenon of postnatal depression emerge from the group level qualitative study (Chapter 6) and was supported by the findings of the logistic regression, CFA specification search and ecological studies. The interview guide was altered to integrate those quantitative findings and subsequently encouraged exploration of cultural and ethnic issues in the focus groups. The qualitative interviews and focus groups reported a belief that maternal depression was higher among migrant families. This was confirmed by the quantitative study (Chapter 5). In relation to unplanned pregnancy it was suggested that where a mother was not married different cultures may respond in different ways. Those responses were not thought to be ethnically determined and it was noted that cultural differences also existed within mainstream society. Practitioners interviewed identified a link between social cohesion and ethnic segregation. Divisions between Arabic and Vietnamese groups were noted with a tendency for large groups to become quite separate. By contrast in more diverse communities migrant women participated in multicultural groups because “we want to be part of our community”. By contrast in communities with large homogenous ethnic groups mothers did not want to meet with others because they didn’t like the way women from other cultures parented. “What I found startling was, what we find in Bankstown is there is a huge Vietnamese community and a huge Arabic community. Is it because they are so large that they end up becoming quite separate? Where as in the other two, Liverpool and Fairfield, does the salt and peppering of the multinational nature of it mean that there is a greater degree of integration?”

The segregation may not be a problem until there is a pregnancy and infant. One expert explained it as follows “They might be self-sufficient. Working in a restaurant they are liaising with Vietnamese that is fine. But once you have a pregnancy they will have to go out to access services [and] be part of the broader society for schooling and for everything. That is why the isolation is highlighted because they don’t know the system. They might have been fully integrated into Cabramatta and this pregnancy has to bring them out of their comfort zone.”

383


Also highlighted was the role that Health and Community services might be playing in creating segregation. One expert felt that services were so “fully into addressing the needs of particular religions, or specific groups, that we are creating segregation because it is an easy solution for us”. She gave the example of “Like say Arabic Muslim from Bankstown wouldn’t go to Lakemba. They are different. Although they are both Muslims they wouldn’t mix”. Mothers in the focus groups also highlighted the possible role of ethnic clusters. One mother, when asked why some women might get more depressed than others, wondered if it may be because there are clusters of people from different cultures living in those communities. She said that she knew that depression was more prevalent in mothers of a certain ethnic group and therefore if there was a high concentration of those people in the suburb there would be more depression. There appeared to be a contradiction in what would be best for mothers. One expert felt that if every Muslim woman “who identify let’s say as Suni” was sent to a particular establishment, then we would be creating a cluster or group that would not know anything except that. She felt it was a risk not to expose them to other groups. “So we are addressing this sense of belonging but then we are creating nuclease where this cluster of people, where they don’t know anything of the broader society”

A different expert noted that people from similar ethnic groups seek each other out. Many of those ethnic communities are a “supportive family orientated group” where the women are very supportive of each other. They may come from large families where there are many sisters, aunties and grandmothers and the women often do not work. “So in terms of parenting and having babies, for women from some of those ethnic backgrounds, the Arabic and African groups, are very supportive and very helpful.”

The concurrent group-level exploratory factor analysis and spatial exploratory data analysis suggested a possible interplay between social disadvantage, and ethnic integration or segregation. Indices of ethnic segregation were higher in those suburbs with social disadvantage. One expert had wondered about this relationship in the suburb of Miller. “I have often wondered in Miller the kind of interplay, racism is going up and people don’t think that multiculturalism makes life better. I’ve often wondered if that as an area becomes more depressed and disadvantaged and there’s less opportunities and lots of structural

384


problems, and less connectiveness. Whether that means people will go at each more and won’t be able to support each other. Where as if you’ve got those other things in place, the ethnic diversity comes into play in a good way”.

The individual level logistic regression study (Chapter 5) confirmed my earlier South West Sydney findings of an association between mother not being born in Australia and postnatal depressive symptoms (Eastwood 2005). The variable “country of birth” was also analysed using CatPCA and CFE specification search. In the CatPCA it loaded most strongly on dimension four together with self reported health, health of child, social support network and marital status. In the four factor measurement model specification search country of birth, social support network and practical support loaded on Factor 2 which I labelled “ethnic social marginalisation”. The group level quantitative study (Chapter 8) included a three of indicators of ethnic segregation and diversity – Entropy Segregation, Simpson Segregation and Maly Neighbourhood Diversity indices. In the exploratory factor analysis the Entropy Segregation Index loaded on Factor 2 together with low volunteerism, poor schooling, high density low class and low IRSD deciles. By contrast the Maly (Ethnic) Diversity Index loaded on Factor 6 with increased support network, increased practical support, better schooling, low density, and different address in the past 5 years. Ecological visualisation and bivariate analysis found associations between measures of segregation and postnatal depression. The Maly Index of Diversity was weakly associated. In both the ecological likelihood linear and Bayesian spatial multiple regressions “Entropy” remained significant in the models for EDS >9 and EDS > 12. The Bayesian multi-level spatial analysis found for EDS > 12 the density, the Maly Index and Factor 2 improved the model but were not significant. Low % no support, low % poor schooling and Factor 6 improved the model and were significant. Factor 6 significantly improved the Bayesian multi-level model reducing the DIC from 5215 to 5201. In the stratified multilevel analysis measures of segregation and diversity did not improve the model for mothers born in Australia but for those mothers not born in Australia the Maly Index improved the model for EDS >9 and EDS > 12. The ecological studies (linear regression and spatial) found a strong association of aggregated postnatal depression with aggregated measures segregation. When controlling for individual level factors (including not born in Australia) the multilevel

385


measures of segregation were no longer significant. In the stratified study of mothers not born in Australia there was a group level protective effect from increased neighbourhood ethnic diversity (Maly Index). The concomitant paradoxical association with low aggregated social networks was discussed previously. The quantitative findings suggest that ethnic segregation is operating predominantly at the individual family level. By contrast group level ethnic diversity is protective for mothers not born in Australia living in communities with high levels of social capital (social networks). Recontextualisation of these multilevel findings within the acculturation conceptual framework suggests that for migrant mothers, social relations (and possible bridging networks) are stronger in ethnically diverse communities and protect against maternal depression. The converse logic is that, for migrant mothers, there is more social support than might be expected, in suburbs with low aggregated social support. Ethnicity and Perinatal Depression Perinatal depression has been found to be more common among recent migrants to Australia (Williams and Carmichael 1985; Brown, Lumley et al. 1994; Brown and Lumley 2000; Lansakara, Brown et al. 2009). Small and colleagues found that the rates were high among Turkish women but relatively low among Vietnamese and Filipino women (Small, Lumley et al. 2003). Rates of postnatal depression among migrants have also been found to be high among Pacific Island mothers in Auckland, New Zealand (Abbott and Williams 2006), Canadian immigrants, asylum seekers and refugees (Sword, Watt et al. 2006; Stewart, Gagnon et al. 2008; Dennis, Hodnett et al. 2009), London ethnic minorities (Onozawa, Kumar et al. 2003) and Latinas or Hispanic US mothers (Beck, Froman et al. 2005; Diaz, Le et al. 2007). The significance of these findings is complicated by the wide international cross-cultural variation in postnatal depression and depressive symptoms. Halbreich and Karkum (2006) undertook a review of 143 studies from 40 countries and found a wide range in reported rates. The authors concluded that the variability might be due to cross-cultural variables, reporting style, differences in perception of mental health and its stigma, differences in socio-economic environments and biological vulnerability factors. Bina (2008) analysed 70 studies on culture and postnatal depression. Of those she explored in depth 14 studies that focused on the impact of cultural factors on postnatal depression. Researchers in eight of those studies concluded that cultural rituals alleviated postnatal depression and that a lack of cultural traditions leads to increased rates of

386


depression. The most common ritual identified by Bina was resting for a period after birth with support from the extended family (usually the mother or mother in law). Social support was again identified as important but specifically the review highlighted the importance of the woman’s perception of support (Bina 2008). Of direct relevant to this study are the findings of Stuchbery and colleagues (1998) who undertook a study of Vietnamese, Arabic and Anglo-Celtic mothers in South West Sydney specifically to examine which deficits in their social support network were associated with postnatal depression among mother of a non-English speaking background. In summary, for Anglo-Celtic women, low postnatal mood was associated with a perceived need for more emotional support from partners and mothers. For Vietnamese low mood was associated with poor quality relationship with the partner and a perceived need for more practical support from him. For Arabic women low mood was associated with a perceived need for more emotional support from partners. The authors noted the importance of postnatal rituals. In their study 64 percent of Vietnamese and 61 percent of Arabic women did not have their mothers with them. Beck cited Nahas and Amasheh (1999) who had explored the experiences of postnatal depression among Jordanian women living in Australia. Nahas and Amasheh had found that “Jordanian women were not supposed to be sad because in their culture this means that the women are not at all able to cope and are bad mothers. Living in a country that was not their homeland only accentuated these conflicts. Jordanian mothers living in Australia did not have strong family support they were used to in their own country. These depressed women experienced helplessness due to their inability to fulfil their traditional gender roles as wife and mother” ((Beck 2002) citing (Nahas and Amasheh 1999)). It is clear from this brief overview of literature on ethnic migration and postnatal depression or depressive symptoms that the elements of stress and support continue to play an important role. There is clearly a cultural context to both the perception and reality of these two important elements. Segregation and Integration Literature An important subset of research into neighbourhood effects is the study of racial, ethnic, and socioeconomic segregation. This is where research investigates whether the observed differences in outcomes are attributable to the fact that different sub-groups live in different social, physical and institutional environments (Reardon 2006).

387


There is extensive literature concerning racial segregation principally from the United States where African Americans are significantly more segregated than the white population or other racial/ethnic minorities. The effects of segregation on health have been thus examined principally in relation to the health status of US Blacks (AcevedoGarcia and Lochner 2003; Kramer and Hogue 2009). In their review of 15 such studies Acevedo-Garcia and Lochner (2003) authors found that no studies explored explanations of why and how residential segregation influences health, the testing of specific pathways and the development of multilevel models. They hypothesized that segregation had an indirect effect on health outcomes operating through multiple mediator variables (Acevedo-Garcia and Lochner 2003). Kramer and Hogue identified 39 studies that tested an association between segregation and health outcomes. The health effects were described as complex with isolation segregation associated with poor pregnancy outcomes and increased mortality for blacks. Several studies, however, reported health-protective effects of living in clustered black neighbourhoods after controlling for the effects of social and economic isolation. The majority of the segregation measures developed have focused on analysis of residential separation of whites and blacks in the US. Maly (2000) developed the Neighbourhood Diversity index which allows researchers to include multiple racial and ethnic categories while adhering to a focus on integration rather than segregation. These measures of segregation and integration have allowed researchers to study urban residential changes in ethnic integration both spatially and over time (Modarres 2004; McCulloch 2007). The importance of using measures of both segregation and integration are illustrated below. As noted in previous sections ethnic segregation may provide health advantages or conversely be detrimental. Consistent with this notion, in a small study of a disadvantaged minority community in Alabama, Mitchell and LaGory (2002) reported that high bonding social capital (measured by the strength of trust and associational ties with others of a similar racial and educational background as the respondent) was paradoxically associated with higher levels of mental distress. In the same study, however, individuals who reported social ties to others who were unlike them with respect to race and class (i.e. who had access to bridging capital) were less likely to report mental distress. To recontexualise this study within an Acculturation framework provides an alternative theoretical perspective. The four individual level outcomes of the acculturation process

388


are: integration, assimilation, separation and marginalisation. Within the Acculturation framework “bridging capital” is also “integration” and provides positive mental health benefits whereas “high bonding capital” is “segregation” and is associated with higher levels of mental distress. Abductive and Retroduction Analysis Redescribing ethnicity and perinatal depression within an acculturation conceptual framework provides explanation for a number of possible causative mechanisms including cultural traditions of postnatal support, marginalisation, societal prejudice and discrimination, social support, ethnic integration and segregation. Specific to this conceptual framework is a possible causative mechanism called acculturation. A recent study by Ayers and colleagues (2009) found that a composite measure of acculturation, among US Korean immigrant women, had an indirect protective relationship with depression mediated through immigration stress. The mechanism of acculturation was not associated with social support which, in their study, had a direct protective effect on depression. Structures within the global, political and social levels that may generate conditional causative mechanism include: global markets, big business, political structures, dominant culture, government agencies, religious organisations, and civil society structures. Candidate conditional mechanisms may include: immigration and refugee policy, skills shortages, societal prejudice and discrimination, religious tolerance, and social cohesion. As noted above, recontextualisation of paradoxical multilevel findings within the acculturation conceptual framework suggests that for migrant mothers, social relations (and possibly bridging networks) are stronger in working class communities and protect against maternal depression. The converse logic was that, for migrant mothers, there is more social exclusion than might be expected, in suburbs with extremes between rich and poor. I have previously examined the related causal and conditional mechanisms of stress, social isolation, social exclusion, social support networks and social services. I propose that the following conditional mechanisms may influence the tendency of acculturation to reduce social isolation, stress and depression: 1. Immigration policy (i.e. family members, promote tolerance) 2. Media policy and response toward migration and refugees 3. Global market and business approach to migration

389


4. Civil society response to migration 5. Settlement patterns (i.e. integration or segregation) 6. Linking social capital and social cohesion 7. Strong ethnic bonding networks 8. Strong bridging networks 9. Access to services including home visiting nursing services.

8.10.3

Proposition

A critical realist model of the ethnic migrant mechanisms is below. Conditions (other mechanisms)        



Immigration policy Media policy and response Global market and business approach Civil society response to migration Settlement patterns Linking social capital and social cohesion Strong ethnic bonding networks Strong bridging networks Access to services.

Ethnic Migrant

Outcomes

Mechanisms Acculturation

Reduced Stress

Cultural Practices

Increased

Integration

support

Figure 132: Critical Realist Model of Acculturation & Ethnic Integration Propositions Based on the above analysis the following propositions are made: 

Immigrant acculturation is a social level mechanism that protects against maternal stress when certain personal characteristics and contextual conditions exist



Ethnic segregation is a social level mechanism that increases maternal isolation when certain personal characteristics and contextual conditions exist



Ethnic integration is a social level mechanism that increases maternal support when certain personal characteristics and contextual conditions exist.

390


Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Application Hill’s aspects of association Strength Acculturation and ethnic mechanisms were identified by QUAL and QUANT as associated with depression at the individual and ecological levels. Consistency Ethnic and migrant factors have been previously identified as associated with maternal depression Specificity No specificity identified Temporality No temporality demonstrated in this study Biological There was a gradient at the ecological level gradient Plausibility The association between migration and ethnic segregation with stress is plausible Coherence The association is coherent with what is know Experimental No experimental evidence was identified evidence Analogy There is an analogy between of ethnic segregation and social exclusion both causing stress Thagard’s Principles Symmetry There is symmetry between ethnic segregation causing stress and the buffering of stress by social support and practical support thus preventing depression Explanation The ethnic and acculturation propositions a) coheres with evidence on depression, b) coheres with other propositions and c) is not a single proposition. Analogy There is an analogy between of ethnic segregation and social exclusion Data priority The proposition describes the data observations. Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The ethnic and acculturation propositions are coherent with the overall system of propositions Thagard’s Criteria Consilience Ethnic and acculturation propositions explains a limited range of known facts Simplicity Ethnic and acculturation propositions are not sufficient to explain depression. Not the most simple explanation Analogy There is an analogy between of ethnic segregation and social exclusion

Table 62: Social exclusion proposition – Inference to Best Explanation In summary the Ethnic and acculturation propositions are not the best explanation. It explains some of the observed and known evidence but is insufficient on its own.

391


8.11 Global-Economic Level Mechanisms 8.11.1

Introduction

The intention has been to limit this study to neighbourhood and community level economic, social and physical influences. As observed in Chapter 1 Clark challenges this perspective and argues “everything in the situation both constitutes and affects most everything else in the situation in some way … Here the macro/meso/micro distinctions dissolve in the presence/absence” (Clarke 2005). The critical realist ontology provides another perspective by claiming that reality is stratified and that each level can influence each other level. The levels are defined by the structures and their generative mechanisms. Table 56 (Section 8.4.2) described the following ontological levels as relevant to this study: biological, psychological, social, cultural and global economic. I have limited my analysis in this Chapter to the Social Level with some exploration of Psychological and Cultural Levels. I will provide brief comment here on a number of global-economic level mechanisms. The following have been identified by drawing on the qualitative findings of Chapters 4 and 6, and the retroductive analysis in the previous sections: 1. Migration 2. Media 3. Big Business and Power.

8.11.2

Migration

Migration has been a significant historical feature of the populating of South West Sydney. Following the Second World War those migrants originated predominantly from Europe. Since the 1970s there has been a significant migration of Indo-Chinese peoples and more recently migration of peoples from Eastern Europe and the Middle East. In this study approximately 45 percent of mothers were not born in Australia. I explored the role of migration, acculturation, and ethnic segregation in Section 8.10 Migration will be included as a Global-Economic Level mechanism in the conceptual framework that follows.

392


8.11.3

Media and Advertising

The role that media and advertising might play in relation to perinatal depression and maternal expectation emerged from focus groups and situational analysis in Chapter 4. “Huggies adverts” were used to describe what Beck (2002, p 458) called the “incongruity between expectations and the reality of motherhood”. Another impact that media and advertising may have on mothers relates to lifestyle “dreams”. These dreams and aspirations relate to aspirations that she may have regarding wealth, material goods, holidays and education for her children. The large shopping malls with their “glitter” were seen as also playing a role. This portrayal of mothers, babies, “motherhood” and lifestyle “dreams” in advertising may reflect existing cultural norms, or alternatively play a role in creating expectations that may be inaccurate. I have postulated that media and advertising constitute a generative mechanism that is influencing mother’s expectations and thus contributing to stress. “Big Media” in Australia plays an important role in relation to sport franchises. Emerging from the interviews was concern regarding the impact that this had had on the local community when the local Rugby League team merged with a team from outside the region. Rugby League in South West Sydney was reported to have declined with no obvious replacement sporting outlet for young people. The Malls with their “glitter” and the loss of the local football were both seen as contributing to the phenomenon of “depressed community”. I have postulated that media and advertising constitute a generative mechanism that contributing to a “depressed” neighbourhood context. Media and Advertising will thus be included as a Global-Economic Level mechanism in the conceptual framework that follows.

393


8.11.4

Big Business and Power

The situational analysis in Chapter 6 identified the strong influence of big business, media and the global economy on the situation for mothers and infants. The impact of the large shopping malls on local communities was seen as both having positive and negative impacts on mothers as it provided meeting places for mothers but also “depopulated” local shopping areas and parks. Big business was seen as having “power” to influence local politicians. One example given, related to the building of “fast food” outlet on land that community members wished to use as a community garden. The “McDonaldization” thesis advanced by Ritzer (2008, p 457) “is the process by which the principles of the fast-food restaurant are coming to dominate more and more sectors of society”. Ritzer describes the process as being delineated by efficiency, calculability, predictability, control through technology and “irrationality of rationality”. The later process inevitably leads to dehumanisation of jobs, settings and circumstances (Ritzer 2008, p 459) with impacts on local neighbourhoods and communities. During the course of this study, South West Sydney experienced the impact of global economic mechanisms with closure of large business and loss of employment. Corporate Business elected to move industry to other countries and jurisdictions. As a result the New South Wales Government budgets were affected leading to impacts on urban development, social services and maintenance of essential infrastructure. I have postulated that corporate business is a global-economic level structure with generative powers that can have significant impacts on the neighbourhood context. Corporate Business will be included as a Global-Economic Level mechanism in the conceptual framework that follows.

394


8.12 Review of other Theoretical Frameworks The theoretical and conceptual frameworks selected for the analysis in this Chapter were chosen after considering other relevant social epidemiological, social, community psychology, public health, perinatal and postnatal depression theories. Some are briefly discussed here.

8.12.1

“Black box” Debate

Cognisance was given to the “black box debate” of the 1990s and the proposal for a new era called “eco-epidemiology” in which broad explanatory theories of disease and health in populations were developed consistent with a systems approach (Pearce 1996; Susser and Susser 1996; Krieger 2001). The debate concerning the merits of “eco-epidemiology” versus “risk-factor (black-box) epidemiology continues (Greenland, Gago-Dominguez et al. 2004; Susser 2004). Underlying these perspectives are epistemological questions of scientific theory (Haack 2004; Mayo and Spanos 2004) including the discourse between inductivism, and deductive theory/refutationism.

8.12.2

Eco-social Theory

In relation to the theoretical basis of social epidemiology I reviewed the theories explored by Krieger (2001). Krieger critiqued three main theories used by practising social epidemiologists at that time namely: (1) psychosocial, (2) social production of disease and/or political economy of health, and (3) eco-social theory and related multi-level frameworks. Drawing on her earlier work (1994; 2000) Krieger argued for an eco-social construct that minimally included: 1. “embodiment, a concept referring to how we literally incorporate, biologically, the material and social world in which we live, from conception to death; a corollary is that no aspect of our biology can be understood absent knowledge of history and individual and societal ways of living. 2. pathways of embodiment, structured simultaneously by: (a) societal arrangements of power and property and contingent patterns of production, consumption, and reproduction, and (b) constraints and possibilities of our biology, as shaped by our species' evolutionary history, our ecological context, and individual histories, that is, trajectories of biological and social development;

395


3. cumulative interplay between exposure, susceptibility and resistance, expressed in pathways of embodiment, with each factor and its distribution conceptualized at multiple levels (individual, neighbourhood, regional or political jurisdiction, national, inter- or supra-national) and in multiple domains (e.g. home, work, school, other public settings), in relation to relevant ecological niches, and manifested in processes at multiple scales of time and space; 4. accountability and agency, expressed in pathways of and knowledge about embodiment, in relation to institutions (government, business and public sector), households and individuals, and also to accountability and agency of epidemiologists and other scientists for theories used and ignored to explain social inequalities in health; a corollary is that, given likely complementary causal explanations at different scales and levels, epidemiological studies should explicitly name and consider the benefits and limitations of their particular scale and level of analysis.”

8.12.3

Health Theories of Inequality

Also considered were theories used to explain health inequalities. Bartley (2004) discussed four of these. They were the: 1) Behavioural and Cultural, 2) Psychosocial, 3) Materialistic, and 4) Life-course theories. The Life-course theory has been used in this thesis to argue for the importance of early influences on the foetus and infant. Life-course theory will be incorporated as appropriate into the development of theory and conceptual models in this Chapter. The Psychosocial Theory is central to the development of theories in this Chapter but I have elected to focus the theory development process on two of its key elements namely stress and support. Krieger noted that the psychosocial theory “directs attention to endogenous biological responses of human interactions [and] focus is on responses to ‘stress’ and on stressed people in need of psychosocial resources. Comparatively less attention, theoretically and empirically, is accorded to: (1) who and what generates psychosocial insults and buffers, and (2) how their distribution – along with that of ubiquitous or non-ubiquitous pathogenic physical, chemical or biological agents – is shaped by social, political and economic policies” (Krieger 2001, p 670). The psychosocial theory is central to this analysis and is integrated into the theoretical framework described in Chapter 9. The Behavioural and Cultural theories discussed by Bartley were principally related to life style behaviours which would be of direct relevance to consideration of perinatal tobacco

396


and alcohol use. Of more relevance to this study was consideration of theories related to ethnic and cultural segregation, isolation and integration which were explored above. The Materialist and Neo-material Theories relate to existence of material causes of health and inequality which indicate that health is worse, and life expectancy lower, in people who have lower incomes (Bartley 2004). Recently “neo-materialist” theory has emerged which concentrates on public provision, or subsidisation of services and utilities such as education, housing, water and transport. The ‘Materialist’ and ‘Neo-materialist’ Theories, with their elements of essential needs, social participation and public provision of subsidies and services have not been sued in this study. I elected to use the closely linked theory of social exclusion but acknowledging that the concept of social exclusion extends beyond material causes of poor health.

8.12.4

Community Psychology Theories

The field of community psychology had much to offer the development of a theoretical approach to neighbourhood context and perinatal depression. Theoretical positions reviewed included: 1) Transactional and Ecological Theory, 2) Empowerment Theory, 3) Stress Theory, 4) Social Support Theory, and 5) Citizen participation. Transactional theory is a theory that emphasises the dynamic, reciprocal interactions between the infant and their context, with bidirectional influence being a fundamental element. A comprehensive ecological model of development (Bronfenbrenner 1979) has been linked to this transaction theoretical view. The resulting transactional – ecological (TE) theory is similar to the eco-epidemiological and Eco-social theories discussed earlier and supports the general approach being taken in this Study. Empowerment Theory can be considered as both a “value orientation for working in the community and as a theoretical concept for understanding the process and consequences of efforts to exert control and influence over decisions that affect one’s life, organisational functioning, and the quality of community life” (Zimmerman 2000, p 43). Empowerment theory is understood at multiple levels including individual, organisational and community levels.

397


Levels of analysis Individual

Organisational

Community

Process (“empowering”)

Outcome (“empowered”)

Learning decision-making skills

Sense of control

Managing resources

Critical awareness

Working with others

Participatory behaviours

Opportunities to participate in decision-making

Effectively compete for resources

Shared responsibilities

Networking with other organisations

Shared leadership

Policy influence

Access to resources

Organisational coalitions

Open government structure

Pluralistic leadership

Tolerance for diversity

Residents’ participatory skills

Source: (Zimmerman 2000, p 47)

Table 63: A Comparison of Empowering Processes and Empowered Outcomes across Levels of Analysis In his critical analysis of empowerment theory Zimmerman notes that it is difficult to measure empowerment and thus some have dismissed its usefulness. Zimmerman further notes that empowerment may be considered as equivalent to power and be linked to issues regarding the struggle for power, power relationships and efforts to exert control over, or influence community power structures (Zimmerman 2000). Empowerment theory has significant heuristic application to the development of theory in relation to neighbourhood context and perinatal depression. There is, however, difficulty in measuring empowerment at the various levels as described by Zimmerman. Stress Theory forms a central tenant of Psychosocial Theory and was used in the analysis in this Chapter. See Section 9.5 Social Support Theory forms a central tenant of Psychosocial Theory, Stress Process Theory, and Social Capital Theory. Barrera (1986) distinguishes between the concepts of social embeddedness, enacted support, and perceived social support. According to Barrera “social embeddedness” refers to the characteristics of the individual’s social network. “Enacted support” refers to the actual helping behaviours that are exchanged. Six different kinds of helping transactions are proposed. They are: advice and information, emotional support, physical assistance, recreation, and positive feedback. “Perceived support” refers to the individual’s evaluation of the quality of support provided by the social network, that is, the degree to which it is seen as available and helpful.

398


Citizen Participation Theory concerns the “process in which individuals take part in decision making in the institutions, programs, and environments that affect them” (Heller, Price et al. 1984) cited by Wandersman and Florin (2000). Community participation is an important component of social cohesion.

8.12.5

Theoretical Models of Neighbourhood Effects

Drawing on the work of Green and Ottoson (1999) and Wandersman and Nation (1998), Ellen and colleagues (2001) propose that neighbourhoods can influence health outcomes through four pathways: (1) neighbourhood institutions and resources (2) stresses in the physical environment (3) stresses in the social environment and (4) neighbourhood based network and norms. 1. Neighbourhood institutions and resources: Neighbourhoods clearly differ in their resources such as parks, libraries, access to healthy food, public transportation, access to health care facilities and so on. Thus, the distribution of those institutionalized resources will have consequence for maternal and infant health outcomes. This pathway suggests that collective investment in the quality and quantity of social and material resources would contribute to the outcomes of individual children. 2. Physical stresses in the neighbourhood environment: the most commonly discussed way in which neighbourhoods influence health is through the proximity of polluting factories and toxic waste sites, which may increase people’s chance of contracting cancer and other illness. Aging and poorly maintained environments – crumbling sidewalks, decaying stairwells, and dangerous playgrounds – are likely to increase the risk of accidents. These conditions are more likely to affect families living in low income neighbourhoods. 3. Social stresses in the neighbourhood environment: people’s health status can be directly affected by the social conditions in a neighbourhood. For instance, living in a neighbourhood with high rates of crime, a mother is more likely to be injured. Furthermore, there have been evidences that exposure to social conditions such as crime, violence, and noise can lead to a higher level of stress. Elevated level of stress in turn may result in many diseases and unhealthy behaviours like smoking. 4. Neighbourhood based social networks: neighbourhood social networks may shape health outcomes through transmitting norms about accepted behaviour,

399


communicating important information or providing social support. For instance, smoking or eating a high fat diet may be more socially acceptable in some neighbourhoods than in others or feeling of hopelessness and isolation are more widely spread among residents of poorer and less empowered communities. Macintyre and colleagues (1993) developed a framework that proposed that the following aspects of neighbourhoods might be health promoting or health damaging:

1. “Physical features of the environment shared by all residents in a locality (for example, air and water quality) 2. Availability of healthy environments at home, work, and play (for example, decent housing, secure and nonhazardous employment, safe play areas for children) 3. Services provided to support people in their daily lives (for example, education, transportation, street cleaning and lighting, and policing) 4. The socio-cultural features of a locality (for example, the political, economic,

ethnic, and religious history and the degree of community integration 5. The reputation of an area (for example, how the area is perceived by residents, service or amenity planners, and investors) (Macintyre, Maciver et al. 1993)”. (Macintyre and Ellaway 2003, p 33)

400


8.12.6

Perinatal Models

A framework proposed by Misra et al (2003) marries a life course perspective, incorporating forces that influence the health of women through successive stages of their lives and their reproductive cycles with a multiple determinants model (Figure 133). The perinatal health framework is an adaptation of the Evans and Stoddart (1990) model of health determinants, which while acknowledging the direct influence that biological, behavioural, environmental, and social factors have on health status, provides a framework for understanding the interrelations between such factors. The framework proposed by Misra et al (2003) is focused on the perinatal period with an emphasis on how factors relate to the preconception and inter-conception periods, and how multiple factors interact to influence perinatal outcomes.

Source: (Misra, Guyer et al. 2003) p 68 (Misra et al, 2003).

Figure 133: Perinatal health framework

401


Culhane and Elo (2005), reviewed the potential mechanisms through which neighbourhood context may influence perinatal outcomes. They outlined a conceptual framework that links neighbourhood context to adverse perinatal events highlighting important intervening variables along this pathway (Figure 134).

Source:(Culhane and Elo 2005, p S23)

Figure 134:Conceptual Framework of Neighbourhood Influence on Pregnancy Outcomes In the framework proposed by Culhane and Elo the neighbourhood conditions that are hypothesized to influence health, either directly or indirectly, are features of the neighbourhood’s social environment, service environment, and physical characteristics (Robert 1999). Social environment are referred to as the level of neighbourhood cohesion or disorganization, norms of reciprocity, civic participation, crime, socioeconomic composition, residential stability, and related attributes. These characteristics are hypothesised to influence health outcomes through a number of potential pathways that include availability of social support, adaptation of coping strategies, and exposure to chronic stress (Anderson, Sorlie et al. 1996; Diez Roux, Nieto et al. 1997; Taylor and Repetti 1997; Cubbin, LeClere et al. 2000; Buka, Brennan et al. 2003; Morenoff 2003).

402


Matthews and Meaney (2005) provide an extensive review of the scientific basis for the influence of adverse perinatal events on life course outcomes. They noted that adversity associated with poverty produces incomplete, dysfunctional families often rife with domestic violence, drug use, child abuse and neglect and that reproduction within this context results in maternal stress, increased risks for infection and thus preterm labour, perinatal deaths, birth insults, poor nutrition for mother and offspring and serious compromises in the quality of parent-child interactions. Matthews and Meaney (2005) also noted that environmental adversity can compromise the emotional well-being of the parent and thus influence the quality of parent-child relationships and that that relationship is strongly linked to maternal anxiety and depression.

Source: (Matthews and Meaney 2005, p 173)

Figure 135: Proposed major pathways of environmental adversity influencing development

403


8.12.7

Qualitative Studies of Postnatal Depression

As noted in Chapter 4, after completing the qualitative studies, I found that Beck had also undertaken extensive previous qualitative research on postnatal depression including a metasynthesis (Beck 1992; Beck 1993; Beck 2002). Her phenomenological study of postpartum depression indentified 11 themes that “described the essence of this experience”. Those themes dealt with unbearable loneliness, obsessive thoughts, loss of self, suffocating guilt, cognitive impairment, loss of previous interests and goals, uncontrollable anxiety, insecurity, loss of control of emotions, loss of all positive emotions, and contemplation of death (Beck 1992). Beck went on to develop a substantive theory of postpartum depression entitled “Teetering on the Edge” (Beck 1993, p 42). Beck found that “Loss of control emerged as the basic social psychological problem”. She described women with postpartum depression as trying to “cope with this problem through the following four-stage process: (a) encountering terror, (b) dying of self, (c) struggling to survive, and (d) regaining control”. In 2002 Beck published a metasynthesis of 18 qualitative studies on postpartum depression. Four overarching themes emerged from her metasynthesis that reflected four perspectives of postnatal depression: “(a) incongruity between expectations and the reality of motherhood, (b) spiralling downward, (c) pervasive loss, and (d) making gains (Beck 2002, p 453). I integrated this substantial body of previous qualitative research into the individual level qualitative theory generation (Section 4.6) and will utilise again here as part of the theory generation process.

404


8.13 Comparison of Theories 8.13.1

Introduction

Making a comparison between different theories and abstractions is Stage 5 of the critical realist “stages in explanatory research” as described by Danermark and colleagues (2002, p 110). “In this stage one elaborates and estimates the relative explanatory power of the mechanisms and structures which have been described by means of abduction and retroduction within the frame of stages 3 and 4”. As suggested by Danermark and colleagues I initially undertook this process as part of the abductive and retroduction abstraction in the previous sections.

8.13.2

Contribution from Analysis

The theories used for that analysis are complementary and focus on different mechanisms and context. The retroductive analysis across sections remained consistent. Areas of complexity and uncertainty, such as the role of bonding and bridging social capital, were similar for each of the theoretical perspectives and were consistent with findings from empirical studies. No one theoretical perspective was able to provide a complete explanation of neighbourhood context and the phenomenon of perinatal depression. Stress Process theory provided a strong foundation for building a conceptual framework of maternal depression, stress and neighbourhood context. Theories of social isolation, exclusion, and capital were able to strengthen the explanatory power of the emerging framework. Missing from analysis at the psychological and social levels was the explanatory power provide by a study of culture. The analysis of acculturation and ethnic segregation contributed significantly to explaining the study findings. But this analysis did not fully explore the role of mainstream Australian “culture”. The brief analysis of media and advertising argued that advertising was a mechanism that influenced maternal expectations of motherhood. I have elected to not explore theory related to media in any further depth.

8.13.3

Contribution from other theories

The “eco” theories critically examined by Krieger (2001) have played an important role in the development of social epidemiology theory. This layered approach has been conceptually used for the development of recent multi-level studies based on spatial or

405


areal units. This approach also forms the basis for the conditional matrix proposed by Corbin and Strauss (2008, p 94). The approach is, however, overly simplistic and as Clarke (Clarke 2005) observes “everything in the situation both constitutes and affects most everything else in the situation in some way”. Such a view is consistent with critical realism ontology where each strata may interact with layers above and below to produce new mechanisms, objects and events. Critical realism also places importance on temporal and spatial dimensions. The eco-social construct proposed by Krieger (2001) is similar to critical realism and draws on the concept of embodiment which incorporates the biological, material and social worlds. The psychosocial theory is central to the development of the conceptual framework developed here and is closely related to the stress process model proposed by Pearlin (Pearlin, Menaghan et al. 1981). As Krieger observes, however, the psychosocial theory does not explain who and what generates the psychosocial insults and buffers or how they are distributed. This is where critical realist with its examination of structures and mechanisms can contribute. The theoretical models of neighbourhood effects (Ellen, Mijanovich et al. 2001; Macintyre and Ellaway 2003) are directly relevant to this study. The emerging theory is consistent with that proposed by Ellen and colleagues with the identification of social stresses and the buffering effects of social networks. Neighbourhood or community institutions, resources and physical attributes emerged from the qualitative studies as important stressors or generative mechanisms. The perinatal models (Misra, Guyer et al. 2003; Culhane and Elo 2005; Matthews and Meaney 2005) reviewed above are consistent with the emerging conceptual framework. In particular, the model described by Matthews and Meaney (2005) goes a long way to describing the findings of this study. Finally Becks’ metasynthesis and model provides a explanation of the mechanisms and processes occurring at the psychological level and will be included in the conceptual framework that follows (Beck 2002).

406


8.14 Revised Conceptual Framework Globalisation Theories

Global-Economic Level Mechanisms (examples)

Social Isolation and Exclusion Theories Stress Process Model Developmental Origins of Health and Disease Theories

Media & Advertising Motherhood Lifestyle “dream” Sports Franchise “Glitter” of Malls

Migration War dislocation Economic Family groups Settlement Policy

Cultural Level Mechanisms Expectations of motherhood, wife, daughter Expectations of father, husband, son Australian Dream Acculturation Cultural practices

Social Level Buffers Social Support Social Networks Bridging Networks Bonding Networks Linking Networks Social Inclusion Trust, Safety Information Services Emotions support Practical support

Neighbourhood Context Social Level Stressors Class Social economic position Racism Income inequality Marginalisation Segregation Neighbourhood decay Neighbourhood crime

Psychological Level Stressors Chronic Colourism Discrimination Lifetime traumas Event Pregnancy, birth Expectations Overwhelmed Alone Loss

Context(C)

Social Capital, Network & Cohesion Theories

Corporate Business Employment Mega-Malls Urban Development McDonaldsism

Psychological Level Buffers Mastery Agency Sense of Control Emotional Reliance Mattering Expectation Confidence

Maternal Outcomes (O) Depression, anxiety Smoking, Alcohol Drugs Decreased breastfeeding Decreased responsiveness & poor attachment to infant

“Necessary” Mechanism (M) Stress

Developmental Trajectory

Figure 136: Conceptual Framework of Maternal Depression, Stress and Context

407


8.14 Conclusion The purpose of this Chapter was to construct a theory of maternal depression and context using a critical realist approach to explanatory theory building. The Chapter commenced with an overview of the philosophical and methodological approach to the Explanatory Phase of Theory building. The approach used included definition of stratified levels, analytical resolution, abductive reasoning, comparative analysis of quantitative and qualitative findings, retroduction, postulate and proposition development, and the review and comparison of relevant theories. A conceptual framework was described which included examples of mechanisms at psychological, social, cultural and global-economic levels. Stress was identified as a necessary mechanism that has the tendency to cause several outcomes including depression, anxiety, and health harming behaviours. The conceptual framework utilised the stress process model as a starting point and included conditional mechanisms identified through retroduction. In the next Chapter I will describe The Thesis, Theoretical Framework and related propositions and models.

408

Chapter 9: The Thesis

Chapter 9: The Thesis, Theoretical Framework, Propositions and Models Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







409


The Thesis

In the neighbourhood spatial context, in keeping with critical realist ontology, global-economic, social and cultural level generative powers trigger and condition maternal psychological and biological level stress mechanisms resulting in the phenomenon of maternal depression and alteration of the infants’ developmental trajectory

9.1

Introduction

The purpose of this Chapter is to present The Thesis, Theoretical Framework, Propositions and Models explaining neighbourhood context, maternal depression and the developmental origins of health and disease. The Chapter draws on both the theories and abstractions in Chapter 8 and the empirical findings and theory generation of earlier chapters.

9.2

Theoretical Framework Preamble

This theoretical framework takes a critical realist perspective of perinatal social context, depression and the developmental origins of health and disease (Bhaskar 1975; Layder 1993; Sayer 2000; Danermark, Ekstrom et al. 2002; Pawson 2006) and builds on the emerging literature on stress process (Pearlin, Menaghan et al. 1981; Hogue, Hoffman et al. 2001; Aneshensel 2009; Avison, Aneshensel et al. 2010), social isolation (Cacioppo and Hawkley 2003; Hortulanus and Machielse 2006), social exclusion (Bonner 2006; Hutchison, Abrams et al. 2007), social capital (Bourdieu 1986; Putnam 1993; Carpiano 2006; Kawachi, Subramanian et al. 2008), segregation (Acevedo-Garcia and Lochner 2003; Kramer and Hogue 2009), acculturation (Berry 1997; Nauck 2008), Globalisation (Florey, Galea et al. 2007), neighbourhood effects on health (Ellen, Mijanovich et al. 2001; Macintyre, Ellaway et al. 2002; Kawachi and Berkman 2003), perinatal adversity (Misra, Guyer et al. 2003; Culhane and Elo 2005; Matthews and Meaney 2005), and the developmental origins of health and disease (Barker 1992; Ben-Shlomo and Kuh 2002; Gluckman and Hanson 2006).

410


Critical Realist Meta-Theory I respond to the call for a realist approach to social epidemiological theory building (Muntaner 1999; Raphael 2006) and draw on the philosopher Bhaskar’s (1975) articulation of critical realism with its ontological stratification of reality. Such a stratified ontological perspective adds theoretical depth to the layered ecological models advanced by earlier social epidemiologists (Susser and Susser 1996; Lynch 2000; Krieger 2001; Kaplan 2004). The first stratified ontological domain consists of the empirical experienced reality, the actual or possible events if activated, and the real which comprises structures, mechanisms, powers, and agency of participants. The second domain of reality consists of hierarchically ordered levels. Each level has its own level specific generative mechanisms that constitute or define a level (Danermark 2002). Mechanisms, objects and events exist at each different strata which I have defined here as biological, psychological, social, cultural and global economic. The critical realist interpretation also makes a distinction between the real and actual with “generative” or causal powers that may, or may not, be activated depending upon other conditions or context, such as exist in the “open system” of the social world.

Developmental Origins of Health and Disease I take as my starting point the proposition that antenatal and postnatal maternal stress and depression adversely impact on the developmental origins of health and disease. We cannot yet be certain of the biological level mechanisms that alter the genotypic and phenotypic response to perinatal adversity but the triggering of genetic, neuroendocrine and physiological mechanisms by psychological and nutritional stress are regarded as strong contenders (Matthews and Meaney 2005; Gluckman and Hanson 2006; Meaney 2010). Maternal stress at the psychological level thus may trigger biological level mechanisms that alter her unborn infants’ developmental trajectory (Gluckman, Hanson et al. 2007; Meaney 2010). The role here of antenatal maternal depression is less clear. It is either an indicator and outcome of maternal psychological level stress, or alternatively, an independent trigger of biological level events as suggested by its impact on infant temperament (Davis, Glynn et al. 2007). Postnatal depression and anxiety have consistently been demonstrated to adversely impact on maternal-infant interaction and attachment (Beck 1995; Murray, Stanley et al. 1996; Martins and Gaffan 2000) and subsequent child cognitive, language, behavioural and psychological problems (Cogill, Caplan et al. 1986; Downey and Coyne 1990; Gelford and Teti 1990; Cummings and Davies 1994; Murray, Hipwell et al. 1996; Sohr-Preston and

411


Scaramella 2006). Postnatal depression may also be associated with subsequent maternal and childhood obesity (Herring, Rich-Edwards et al. 2008; Surkan, Kawachi et al. 2008), not breast fed (Henderson, Evans et al. 2003; Dennis and McQueen 2009), and maternal health harming behaviours such as tobacco and drug addiction (Rubio, Kraemer et al. 2008; Cinciripini, Blalock et al. 2010; Lancaster, Gold et al. 2010).

Stress and Depression That depression is caused by psychological stress is increasingly certain (Kinderman 2005; Stone, Lin et al. 2008). Less clear, in critical realist terms, is whether stress is a necessary condition and must be present for there to be depression. The phenomenon of perinatal depression has repeatedly been found to be associated with stressful life events (Beck 1996; O'Hara and Swain 1996; Wilson, Reid et al. 1996; NHMRC 2000; Beck 2001). I agree with Beck (1992; 1993; 2002) that depressed mothers experience an “incongruity between expectations and the reality of motherhood”, a sense of pervasive loss, and loneliness. I propose here that this psychological felt stress, be it internally or externally triggered, is a necessary condition for perinatal depression. But stress in itself is not a sufficient mechanism and the tendency of stress to cause depression is conditional on other personal characteristics and social and cultural context. Following Pearlin (1981) it is argued here that the generative power of stress is conditioned, or moderated, by personal resources such as coping and by social resources such as social support. Those personal resources such as mastery, agency, self-esteem, optimism, mattering, emotional reliance and personal predisposition may themselves be conditioned by social context. The notion of agency and mastery, from a critical realist perspective of agency (Archer 1995, p 175), means that mothers themselves have “generative powers” and may, or may not, be able to influence and control the environmental and experiential forces acting upon them. But the actual outcome will be the result of the interplay between what Archer (1995, p 175) calls structural, cultural and agential emergent properties of society, each with their own “generative mechanisms” and relative autonomy. Thus a mothers individual agency, and ability to “cope”, may be enabled or blocked by her social [structural] and cultural context.

The Triggering Events Drawing the Stress Process Model I argue that the triggers of perinatal depression are

event stressors such as poor health, childbirth experience, loss of status, loss of financial

412


income, loss of control, unmet expectations, irritable infant and partner behaviour.

Chronic stressors may also contribute and including: status strains (i.e. race, sole parent, poverty, and religion), role strains (i.e. wife, motherhood, and daughter), ambient strains (i.e. neighbourhood crime, services, physical decay of environs), and quotidian strains (i.e. loneliness, daily hassle of sole parenthood). In keeping with critical realist meta-theory these triggering mechanisms arise from structural, cultural and agential generative mechanisms including: income inequality, class structure, social exclusion, social isolation, access to social services, neighbourhood physical environments, gender roles, cultural expectations of motherhood, ethnic segregation and acculturation.

Contextual “Stressor” Mechanisms Aneshensel (2009) argues that the stress process model provides a means of explaining mental health disparities that arise from people’s disadvantaged placement within a social hierarchy. The emphasis here is on a generative mechanism called social stress that is generated by disadvantaged social status. Social stress is defined as a state of arousal that results from social demands that tax the ordinary adaptive capacity of the individual or from blocking of the means to attain personal goals. Drawing on social isolation (Hortulanus and Machielse 2006), materialist (Bartley 2004) and social exclusion (Millar 2007) theories, social stress arising from hierarchical placement is joined by theoretical explanations arguing that mechanisms arising from social and economic structures marginalise individuals and groups from social and material resources, and block attainment of their aspirations. Pregnant women and mothers may be unable to access neighbours, friends, family, services, transport, phone and their partners support. In this social context of exclusion the expectations and dreams of mothers may be shattered as they lose control and are overwhelmed with pervasive loss. The social structures generating social exclusion and social stress include occupational class structures, racism, and residential, ethnic, and educational segregation resulting in “depressed communities” with few amenities, decayed environments and perceived unsafe streets. Business, service and political structures contribute to the marginalisation of “depressed communities” through the centralisation of amenities, services and sporting opportunities. The impact on the lives of mothers is to be “stuck in the middle of nowhere … being couped up in a house with a kid 24/7”.

413


Contextual Buffering Mechanisms The buffering role that support plays in relation to both the stress process and perinatal depression are strong and consistently shown (Pearlin, Menaghan et al. 1981; Beck 1996; O'Hara and Swain 1996; Wilson, Reid et al. 1996; NHMRC 2000; Beck 2001; Aneshensel 2009). Studies have consistently shown a negative correlation between postpartum depression and emotional and instrumental support during pregnancy (Menaghann 1990; Richman, Raskin et al. 1991; Beck 1996; O'Hara and Swain 1996; Seguin, Potvin et al. 1999; Ritter, Hobfoll et al. 2000). The support mechanisms received by mothers are determined by her social and cultural context. Drawing on both Social Network (Crow 2004) and Social Capital (Kawachi, Subramanian et al. 2008) Theory it is evident that social structures can generate strong social bonds with a clear demarcation between ‘insiders’ and ‘outsiders’. This ‘bonding’ and ‘bridging’ social capital is generated from complex social structures and has been associated with protection from mental health disorders (Fone, Lloyd et al. 2007; Stafford, De Silva et al. 2008). But strong bonding ties within disadvantaged communities may also be a detriment to the health of residents (Kawachi, Subramanian et al. 2008). For migrant ethnic mothers access to support is critical. In this social context they must rely on both ethnic bonding capital and the bridging capital provided by supportive social structures. Drawing on Carpiano (2006) neighbourhood social capital is the resource inhered within the networks of neighbourhood residents, with social cohesion generating the necessary conditions for group members to access those resources. The structural antecedents of both social cohesion and social capital are the local, and surrounding, socioeconomic conditions, residential stability, and ethnic, family, gender, age, and social class composition. Together with policy level mechanisms such as social and economic interventions, conditions may be created that have a tendency to buffer social and cultural stressor mechanisms.

Cultural Mechanisms Critical realists define culture as a distinct ontological level (Archer 1995; Danermark and Gellerstedt 2004). This is consistent with current multilevel cultural and cross-cultural psychological theory (Adamopoulos 2008) which seeks to . Under this thesis separate and different mechanisms operate at the cultural level from those at the social or psychological levels. Drawing on Beck’s (2002) concept analysis of postnatal depression, culturally defined expectations of motherhood may be shattered by the reality of a new mothers lived experience. In keeping with Cultivation Theory and Social Expectation

414


Theory (McQuail 1994) global media will play an important role in the “cultivating and acculturation” process shaping all mothers beliefs and expectations of pregnancy, childbirth and parenting to a selective view of reality. The cultural differentialism paradigm of Cultural Theory (Ritzer 2008, p 453-463) would hold that mothers deep cultural practices and expectations will remain largely unaffected by globalisation and media. In juxtaposition cultural hybridization will be occurring within local South West Sydney neighbourhood settings producing new and distinctive cultural realities. Within these cultural fluxes migrant mothers’ family and personal cultural expectations will generate stress. As observed by Beck (2002) these conflicting expectations and experiences of motherhood lead mothers “down the path to becoming overwhelmed, perceiving themselves as failures as mothers, and bearing a suffocating burden of guilt”.

Global-Economic Mechanisms As observed by Clark “everything in the situation both constitutes and affects most everything else in the situation in some way … Here the macro/meso/micro distinctions dissolve in the presence/absence” (Clarke 2005). Globalization Theory (Ritzer 2008, p447-473) and conceptual frameworks of Globalization and Health (Labonte and Torgerson 2005; Galea 2007) describe the global and economic generative powers that will impact on neighbourhood context and the lived experiences of mothers and their families. In tension are two economic systems of globalisation – transnational capitalism and socialist globalization each with mechanisms capable of influencing migration, media, corporate business, social services, politics and policy. Drawing on Labonte and Torgerson (2005) the global context will condition, at the policy level, macro-economic, labour, food security, environmental protection, political power, public provision, and migration and refugee policies. Together, and separately from, the generative powers of corporate business, these policy level mechanisms will shape urbanisation, disparities, community capacity, and service access with resulting impacts on triggers, and conditioners, of maternal stress and depression.

Historical and Spatial Context The spatio-temporal situation of people and resources affect the nature of social phenomena (Sayer 2000, p 115). With global, policy, social, psychological and biological processes occurring over time, the historical, and life-course context, will influence or condition, the tendencies of causal mechanisms. Thus individual histories will unfold

415


inside the larger historical sweep of social [and cultural] evolution (Layder 1993, p 15). Though time has a single dimension and is irreversible, space has three dimensions and movement is reversible. Drawing on Bourdieu ((Bourdieu 1977) cited by Sayer (2000, p 117)) structures, objects, mechanisms and individuals can reversibly interact within space. Thus individuals that compose a neighbourhood influence that neighbourhood, and adjacent neighbourhood context, which in turn has a reciprocal temporal and spatial impact on those and other individuals who reside or move through the neighbourhood. Sayer (2000, p 127) argues that in some respects space or geography are vital, but in others, as illustrated by the ability of things and people to remain the same in different settings, it makes only small differences. The implication of Sayers Thesis is that spatial conditioning of maternal depression and stress will be in turn conditioned by generative properties of social, cultural and historical context.

Summary In the neighbourhood spatial and historical context, in keeping with critical realist ontology, global-economic, social and cultural level generative powers trigger and condition maternal psychological and biological level stress mechanisms resulting in the phenomenon of maternal depression and alteration of the infants’ developmental trajectory toward health and disease. In the following section I will make propositions based on the above Thesis and theoretical framework.

416


9.3

Theoretical Propositions

9.3.1 Introduction I will present here a limited set of propositions drawn from the analysis in Chapter 8, and the conceptual framework. The propositions are underpinned by the Theoretical Framework. The approach taken here will be in keeping with the post-positivist assumption of “impossibility of scientific proof” and the criteria of falsification.

Falsification: At least some of the major propositions should be empirically falsifiable. All useful theories suggest ways in which they may be subjected to empirical assessment. (Carpiano and Daley 2006, p 565) Thus the focus will be on testable propositions. As noted in Section 8.3.8 the formulation of the propositions will be “x causes y (in circumstances c)”. The explanation may also include comment on the structure that underlies the generative mechanism (structure of X).

Postulate (untested assumption): Maternal psychological stress is a necessary, but not sufficient, cause of depressive symptoms

417


9.3.2 Maternal Depression and Outcomes Proposition 1: Maternal psychological depressive mood (symptoms) tends to decrease maternal responsiveness to her infant in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services.

Proposition 2: Maternal psychological depressive mood (symptoms) tends to decrease breastfeeding of her infant in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services.

Proposition 3: Maternal psychological depressive mood (symptoms) tends to increase smoking in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services. Conditions (other mechanisms)     

Isolation Lack of emotional support Lack of practical support Limited social network Limited support services

Outcomes

Psychological mechanisms

Decreased Maternal Responsiveness to Infant

Depressive Mood

Decreased Breast Feeding Increased Smoking

Figure 137: MCO Model of Maternal Depression and Outcomes Practical Support

Maternal Depressive Mood

Failure to Respond to Infant

Figure 138: Moderated Model of Depression and Responsiveness

418


9.3.4 Psychological Stress and Depression Proposition 4: Maternal psychological stress triggered by “mismatched” expectation tends to cause (increase) depressive symptoms in the context of loneliness, isolation, lack of emotional support, lack of practical help, limited social network, limited support services, financial stress and poor health

Proposition 5: The maternal psychological depressive mood (symptoms) caused by stress tend to be less in the context of emotional support, practical support, large social networks and good health

Proposition 6: Maternal psychological stress triggered by loneliness tends to cause (increase) depressive symptoms in the context of sole parenthood, social isolation, lack of emotional support, lack of practical help, limited social network, limited support services, financial stress and poor health Conditions (other mechanisms) Trigger Mechanism 

Mismatched Expectations

      

Loneliness/isolation Lack of emotional support Lack of practical help Limited support network Limited support services Financial stress Poor health.

Outcome Stress

Depression

Necessary Mechanism Figure 139: MCO Model of Psychological Stress and Depression

Emotional Support

Maternal Expectations

Maternal Psychological Stress

Maternal Depressive Mood

Failure to Respond to Infant

Figure 140: Moderated Mediated Model of Expectation, Depression and Responsiveness

419


9.3.5 Social Stress and Buffers Proposition 7: Social level stressor mechanisms including class, position, racism, inequality, social isolation, segregation, crime and neighbourhood decay tend to cause an increase in psychological stress in the context of social level buffers including social networks (e.g. Bonding, bridging, linking networks), trust and community safety, social services, information and emotional and practical support


Social Stress mechanisms       

    

Social networks Trust and reciprocity Social services Emotional and practical support Information services

Class/position Racism Inequity Social isolation Segregation Crime Decay

Outcome

Psychological Stress

Figure 141: MCO Model of Social Stress, Psychological Stress and Buffers

Proposition 8: Social level stressor mechanisms including class, position, racism, inequality, segregation, crime and neighbourhood decay tend to cause a decrease in social level buffers in the context of global-economic mechanisms such as migration, Mega-Malls, Sports Franchise, and urban development.


Social Stress mechanisms

   

Migration Mega-Malls Sports Franchise Urban development.

Class/position Racism Inequity Social isolation Segregation Crime Decay

Figure 142: MCO Model of Social Stress and Social Buffers

420

Outcome Decreased Social support Community Trust Social services


Social Stress Class SES Position Racism Inequality Marginalisation Segregation Neigh. Crime Neigh. Decay

Global-Economic Employment Policy Mega-Malls Urban Development Migration

Social Buffers Social Support Bridging Networks Linking Networks Emotional Support Practical Support Services/Information

Social Isolation/ Loneliness


Depression

Responsiveness to Infant

Figure 143: Model of Social Isolation, Stress and Depression

421


9.3.6 Migrant Social Stress and Buffers The above mechanisms will apply differently to non-English speaking and recently-arrived migrant groups. The following is a proposition specifically addressing ethnic migrant mechanisms.

Proposition 9: Migrant related social level mechanisms including acculturation, cultural pregnancy, childbirth and parenting practices, and ethnic geographical integration tend to cause a decrease in social level stress in the context of immigration policy, media, global mechanisms, civil society response, forms of social capital, and access to services. Conditions (other mechanisms)        



Immigration policy Media policy and response Global market and business approach Civil society response to migration Settlement patterns Linking social capital and social cohesion Strong ethnic bonding networks Strong bridging networks Access to services.

Ethnic Migrant Mechanisms Acculturation Cultural Practices Integration

Figure 144: MCO Migrant Mechanisms

422

Outcomes Reduced Stress Increased support


9.3.7 Cultural Mechanisms Proposition 10: Expectations of mothers (and wives and daughters) is a cultural level mechanism that tends to increase maternal psychological stress in the context of strong bonding social networks, weak bridging networks, language barriers, poor access to services and information, and cultural practices.


Cultural mechanisms

    

Bonding social networks Bridging social networks Language barriers Cultural practices Access to services and information

Expectations of mothers, wives and daughters

Outcome Psychological stress

Figure 145: MCO Cultural Mechanisms – Expectations of Mothers

Proposition 11: Mothers expectation of motherhood is a cultural level mechanism that tends to increase maternal psychological stress in the context of media portraying of motherhood, financial stress, and difficult infant temperament, lack of practical and emotional support.


Cultural mechanisms

    

Media portrayal of motherhood Financial stress Infant temperament Emotional, practical, social support Access to services and information

Mothers expectation of motherhood

Outcome Psychological stress

Figure 146: MCO Cultural Mechanisms – Mothers Expectations

423


Social Stress Class SES Position Racism Inequality Marginalisation Segregation Neigh. Crime Neigh. Decay

Social Buffers Social Support Bridging Networks Linking Networks Emotional Support Practical Support Services/Information

Global-Economic Employment Policy Mega-Malls Urban Development Migration

Social Isolation/ Loneliness


Depression

Responsiveness to Infant

Figure 147: Model of Social Isolation and “Expectations Lost”

424

Expectations not met

“Expectations Lost””


Inference to Best Explanation Assessment of Inference to the Best Explanation for the above propositions, using Hills “aspects of association” and Thagard’s principles and criteria, is shown below. Criteria Application Hill’s aspects of association Strength Final models integrate strong associations of financial stress, lack of support, lost expectations and findings related to migrant mothers Consistency Final models consistent with earlier models and analysis Specificity No specificity identified Temporality No temporality demonstrated in this study Biological Gradients were demonstrated in individual and ecological studies gradient Plausibility The associations described are plausible Coherence The association is coherent with what is know Experimental No experimental evidence was identified evidence Analogy There are analogies between isolation and “expectations lost” causing stress Thagard’s Principles Symmetry There is symmetry in the final models depression Explanation The propositions a) coheres with evidence on depression, b) coheres with other propositions and c) are not a single proposition. Analogy There are analogies between isolation and “expectations lost” causing stress Data priority The propositions describe the QUANT and QUAL data observations. Contradiction There are no contradictory proposals Competition No competitive explanation identified where p and q were not explanatorily connected Acceptance The propositions are coherent with the overall system of propositions Thagard’s Criteria Consilience The propositions explain a significant range of known facts Simplicity The propositions are the most simple set of explanations Analogy There are analogies between the propositions here and The Stress Process Model

Table 64: Social exclusion proposition – Inference to Best Explanation In summary the set of ten propositions and the integrated models of social isolation and expectations lost provide the best explanation of the data in this study.

425


Conclusion The purpose of this Chapter was to present The Thesis, Theoretical Framework, Propositions and Models explaining neighbourhood context, maternal depression and the developmental origins of health and disease. The Chapter utilised both the theories and abstractions of Chapter 8 and the empirical findings and theory generation of the earlier chapters. Global, economic, social and cultural mechanisms were identified that explain maternal stress and depression within family and neighbourhood contexts. There is a complex intertwining of historical, spatial, cultural, material and relational elements that contribute to the experiences of loss and nurturing. Emerging are the centrality of social isolation and expectation lost as possible triggers of stress and depression not only for mothers but possibly also others who have their dreams shattered during life’s transitions. In these situations social and cultural context can either nurture and support or marginalise and isolate. In the next Chapter I will summarise these findings with reference to the Study objectives, examine the study limitations and strengths, make recommendations for research, policy and practice relevant to the rapidly growing, multiethnic and relatively disadvantaged communities of South West Sydney.

426

Chapter 10: Conclusion

Chapter 10: Conclusion, Limitations, and Implications Part A: Introduction & Methodology Chapter 1: Introduction Part B: Individual Level Exploratory Analysis







427


10.1 Introduction The previous Chapter presented The Thesis and Theoretical Framework, Propositions and Models developed and constructed using a critical realist Explanatory Theory Building

Method. An overall evaluation of the Theoretical Propositions (Theory) was made using criteria for Inference to the Best Explanation. This Chapter will summarise the study, discuss the limitations and strengths, and explore the potential implications for research, policy and practice. The Chapter will first summarise the findings with reference to the Study objectives. An examination of the study limitations and strengths will follow addressing aspects of epistemology, mixed method study design and quantitative and qualitative methodologies. The following sections on implications for research, policy and practice will have an applied focus with proposal that will be relevant to future research, policy and practice in the rapidly growing, multiethnic and relatively disadvantaged communities of South West Sydney.

10.2 Conclusion This study sought to develop a theoretical model of neighbourhood context and the developmental origins of health and disease using postnatal depression as a case study. Given the nature of social phenomenon that social epidemiology seeks to understand, and the inductive, abductive and deductive phases of theory building, a transdisciplinary and mixed method approach was used. The Aim was to: “utilise mixed methodology to build a conceptual framework, theory and model describing the mechanisms by which multilevel factors influence the developmental and life course outcomes with a focus on perinatal depression in South West Sydney”.

428


The study objectives were: 1. To explore and describe the individual level influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney (Section B) 2. To explore and describe the neighbourhood and community level economic, social, and physical influences of perinatal and infant health, and specifically perinatal depression, in South Western Sydney (Section C) 3. To explain and conceptualise a framework, theory and model describing the mechanisms by which multilevel factors influence developmental and life course outcomes with a focus on perinatal depression (Section D). Based on my prior philosophical position a critical realist approach to theory building was utilised. I described an Explanatory Theory Building Method that integrates transdisciplinary approaches to emergent, abstract and confirmatory phases of theory building. For the Emergent Phase I constructed the Perinatal Conceptual Framework (Figure 11) based on my prior knowledge and beliefs. My four research questions were taken from this conceptual framework. The first question was, “What is the impact of perinatal events on developmental and lifecourse outcomes with a focus on perinatal depression?” I identified a growing body of empirical evidence implicating adverse perinatal events in subsequent developmental and life-course outcomes. Previous research had tended to focus on the long-term outcomes of perinatal low birth weight and poor nutrition. I was able to identify a growing body of evidence that implicated maternal stress and depression as impacting on long-term developmental and life-course outcomes. The purpose of this first research question was to situate the importance of this Study and its subsequent theoretical propositions. I thus took as my starting point the proposition that antenatal and postnatal maternal stress and depression adversely impact on the developmental origins of health and disease. My second question was, “How does maternal [stress and] depression cause detrimental effects to the foetus and newborn infant?” The empirical and theoretical research evidence here has focused on in utero biological mechanisms and postnatal psychological mechanisms. There has been some research on the detrimental impact of maternal depression on the birth process itself. I identified increasing empirical evidence that

429


maternal stress at the psychological level may trigger biological level mechanisms that alter her unborn infants’ developmental trajectory (Gluckman, Hanson et al. 2007; Meaney 2010). The role of antenatal maternal depression was less clear. Postnatal depression and anxiety have, however, been demonstrated to adversely impact on mothers response to her infant, breastfeeding and health related behaviours. I subsequently made the following three critical realist (MCO) propositions. Proposition 1: Maternal psychological depressive mood (symptoms) tends to decrease maternal responsiveness to her infant in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services. Proposition 2: Maternal psychological depressive mood (symptoms) tends to decrease breastfeeding of her infant in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services. Proposition 3: Maternal psychological depressive mood (symptoms) tends to increase smoking in the context of isolation, lack of emotional support, lack of practical help, limited social network, and limited support services. The third research question was, “What, and how do, individual and family level factors influence the phenomenon of maternal depression during the perinatal period?” The perinatal (antenatal and postnatal) empirical research has mainly focused on the role of individual and family level factors. Psychosocial risk factors that have been implicated include history of mental illness, lack of social support, recent life stresses, personality variables and feelings about pregnancy or parenthood. This study found a relationship between maternal depressive symptoms and expectations, infant behaviour, self-reported health, financial stress, emotional, practical and social support. Stress, loneliness, expectations and loss were identified as potential mechanisms. I proposed that psychological felt stress, be it internally or externally triggered, is a necessary condition for perinatal depression. I argued that stress in itself is not a sufficient mechanism and the tendency of stress to cause depression is conditional on other personal characteristics and social and cultural context. Following Pearlin (1981), I argued that the generative power of stress is conditioned, or moderated, by personal resources such as coping and by social resources such as social support. Thus a mothers

430


individual agency, and ability to “cope”, will be enabled or blocked by her social [structural] and cultural context. I subsequently made the following three critical realist (MCO) propositions.

Proposition 4: Maternal psychological stress triggered by “mismatched” expectation tends to cause (increase) depressive symptoms in the context of loneliness, isolation, lack of emotional support, lack of practical help, limited social network, limited support services, financial stress and poor health

Proposition 5: The maternal psychological depressive mood (symptoms) caused by stress tend to be less in the context of emotional support, practical support, large social networks and good health

Proposition 6: Maternal psychological stress triggered by loneliness tends to cause (increase) depressive symptoms in the context of sole parenthood, social isolation, lack of emotional support, lack of practical help, limited social network, limited support services, financial stress and poor health. The fourth research question was, “What, and how do, group level factors influence the phenomenon of maternal depression during the perinatal period?” Social and physical environmental adversity and measures of social capital have been found to be associated with maternal stress and pregnancy and infant health outcomes including prematurity, lowbirth weight and infant mortality. Ecological risk factors that have been implicated in poor mental health include: environmental decay, social capital, economic position and segregation. This study found an ecological level relationship between maternal depressive symptoms and social networks, ethnic segregation, extremes of income, education and density. Those relationships were not always found when controlling for individual level variables suggesting that the observed spatial relationships were compositional in nature. The exception was the negative impact on non-Australian born mothers living in communities with strong social networks, higher education, low population density and ethnic segregation. Social stress, marginalisation, social capital and ethnic segregation were identified as potential mechanisms. Drawing on the Stress Process Model and critical realist meta-theory I argued that stress triggering mechanisms arise from structural, cultural and agential generative mechanisms including: income inequality, class structure, social exclusion, social isolation, access to social services, neighbourhood physical environments, gender roles, cultural expectations

431


of motherhood, ethnic segregation and acculturation. A Thesis and Theoretical Framework was presented to support these arguments. I subsequently made the following three critical realist (MCO) propositions. Proposition 7: Social level stressor mechanisms including class, position, racism, inequality, social isolation, segregation, crime and neighbourhood decay tend to cause an increase in psychological stress in the context of social level buffers including social networks (eg. Bonding, bridging, linking networks), trust and community safety, social services, information and emotional and practical support Proposition 8: Social level stressor mechanisms including class, position, racism, inequality, segregation, crime and neighbourhood decay tend to cause a decrease in social level buffers in the context of global-economic mechanisms such as migration, Mega-Malls, Sports Franchise, and urban development. Proposition 9: Migrant related social level mechanisms including acculturation, cultural pregnancy, childbirth and parenting practices, and ethnic geographical integration tend to cause a decrease in social level stress in the context of immigration policy, media, global mechanisms, civil society response, forms of social capital, and access to services. Proposition 10: Mothers expectation of motherhood is a cultural level mechanism that tends to increase maternal psychological stress in the context of media portraying of motherhood, financial stress, and difficult infant temperament, lack of practical and emotional support.

432


10.3 Limitations and Strengths 10.3.1

Samples

10.3.1.1

Qualitative Sample

Mothers were interviewed at a mothers group. Consequently those mothers had experience of support from a mothers group. They placed strong importance on the importance of the groups for establishing and maintaining support networks. One mother noted that support networks may include family, friends and neighbours. The mothers who attended the groups were not from socially disadvantaged backgrounds. There were also few mothers from non-English speaking backgrounds. These limitations impacted on deeper exploration of cultural and social isolation or marginalisation. Of significance is the absent voice from women who might be socially isolated or oppressed. Saturation of some of the emerging concepts was not achieved and resource and time constraints limited the undertaking of further focus groups and interviews. Despite these limitations the qualitative data was rich and contributed significantly to both theory generation and construction. It was notable that the findings from this qualitative study were very similar to those found by Beck in both her original phenolological (Beck 1992) and grounded theory (Beck 1993) studies and her later meta-synthesis (Beck 2002). 10.3.1.2

Quantitative Sample

The size (21,991) of this cross-sectional study of the EDS administered to women postnatally is unique. There have been few previous reports of perinatal depression studies on population samples of this size. Ferguson et al (1996) approached 14,893 women and interviewed 9,316. Other large studies have been of samples less than 3,000 women (Campbell and Cohn 1991; Astbury, Brown et al. 1994; McGill, Benzie-Burrows et al. 1995; Hall, Kotch et al. 1996; Warner, Appleby et al. 1996; Wickberg and Hwang 1997). The prevalence of EDS score greater than 10 (12.2%) and greater than 12 (6.8%) is similar to that found in other studies (O'Hara and Swain 1996) and the prevalence is similar as the 13.1% found by Johnstone and others (2001) in a recent New South Wales study of 490 women. The study sample contained 30 percent missing data which is problematic and would have been of significant concern if this were a confirmatory study. As the study was exploratory I elected to proceed. Two data sets were prepared, one with imputed missing data and the other with deleted missing data. Both data sets were analysed and with similar results which was not surprising given the large sample size. The higher levels of missing data in

433


the Macarthur nursing sector may provide an explanation for the unexplained spatial residual in the southern suburbs (see Figures 89, 94, 104, 107, 112, and 115).

10.3.2

Analytical Approach

10.3.2.1

Cross-section design

The cross sectional design of this study has limitation in relation to drawing inferences from the regularities observed. The direction of the relationships cannot be established in cross sectional quantitative studies. Criteria for inferring a causal relationship in such situations include those proposed by Hill (1965). As noted in Section 2.2 “differencemaking evidence coming from statistics cannot prove causation in any possible means” and evidence about mechanisms is required. Where causal relationships have been established by other evidence the possibility of reverse causality, or bi-directional causality, cannot be excluded in a cross-sectional study such as here. In the case of research regarding disparities in mental health the direction and pathways of causality remain unclear. Higher mental illness in an area could result in depressed socio-economic and social capital indices rather than the reverse. Cross sectional quantitative studies are also unable to identify the course of disease or symptoms over time. It could be that the disease was caused by some prior event not captured in the cross-sectional design. Maternal depression may have existed prior to the pregnancy and be unrelated to the regularities observed. Epidemiologists argue that longitudinal studies are required to overcome these limitations of cross-sectional design but as previously argued the “difference-making” evidence remains insufficient for establishing causation in the absence of evidence concerning mechanisms. Qualitative research not only provides evidence regarding mechanisms but may also be able to establish the direction of the relationships between variables and historical or temporal course. Thus the limitation imposed by the cross sectional design was partly offset in this study by the integrated mixed method design. The strong relationships observed between independent and dependent variables were thus explained using mechanisms identified from qualitative evidence and abductive reasoning (inference to the best explanation). 10.3.2.2

Qualitative Methods

The qualitative (intensive) methods used in this study included interviews and focus groups with qualitative analysis. The strength of these approaches is their ability to

434


provide explanatory power to the analysis (Sayer 2000) and thus the identification of causal mechanisms. Sayer notes, however, that the “concrete patterns and contingent relations [identified] are unlikely to be ‘representative’, ‘average’’ or generalisable. Assessment of the quality of such intensive or qualitative studies is assessed using criteria that differ from those used for extensive quantitative studies (Greenhalgh and Taylor 1997; Cohen and Crabtree 2008; Kitto, Chesters et al. 2008; Miyata and Kai 2009). Table 65 summarises how the criteria proposed by Kitto and colleagues has been addressed in the current study. The use of qualitative methods for emergent theory building was suited to the aim and objectives of this study which was to explain the mechanisms by which multilevel factors influence developmental and life course outcomes with a focus on perinatal depression. Emergent theory building methods such as those used here are also consistent with accepted critical realist methodologies (i.e. grounded theory). I was not able to identify where situation analysis had been previously used for critical realist theory building but, as methodology developed within the interpretist paradigm, its use within a critical realist epistemology is appropriate. Similarly the use of a mixed method study design is also consistent with critical realist methodological pluralism (Sayer 2000, p 19-20; Danermark, Ekstrom et al. 2002, p150-176). The procedural rigour was made explicit through clear articulation of the ontological and epistemology position informing the study which informed the emergent and explanatory study design, methods of data collection and analysis. The methods used were explicitly described in both the empirical and analytical chapters including accessing and interaction with subjects, data collection, coding and analysis. The sampling techniques used were purposeful and sought to include subjects from different communities, and ethnic backgrounds. Similarly the practitioners were purposefully selected. Limitations of representativeness were made clear.

435


Quality Criteria

How criteria has been addressed in the current study

Clarification and Justification Fit of the research questions, aims and choice of methods

 Emergent (grounded) theory methods are suited to developing theory within a critical realist paradigm

Methodological Rigour Procedural rigour

 The epistemological and ontological positions were explicitly stated

Transparency in the way the data was collected and analysed

 Detailed description of data collection and analysis including subject selection, data collection, coding and phases of analysis

Sampling techniques support conceptual generalisability

 Focus groups and practitioners were purposefully sampled. There were limitations of representativeness that limit generalisability

Interpretive Rigour Demonstration of the data/evidence to support interpretation

 Integrative and synergistic use of both quantitative and qualitative data  Use of constant comparative method, situational analysis and testing of theoretical ideas against both quantitative and qualitative data  Reviewed literature following the emergence of conceptual model to enhance theoretical sensitivity  Presentation of the results included extensive use of participants quotes to support interpretation. The study settings were not clearly described  Theoretical sufficiency was used as a criteria to limit the number of interviews  Abductive analysis utilised triangulation and extant theory  The theoretical models were assessed using criteria for

Inference to the Best Explanation

Reflexivity and Relationality Acknowledging and addressing the influence of investigator-participant interaction on the research process

 The prior position of the researcher was described  Theoretical assumption and prior knowledge were made explicit and described  The researcher did not conduct the focus groups thus maintaining some distance from the data collection.

Evaluative Rigour Ensuring political and ethical aspects of the research are addressed

 Ethics approval was obtained from relevant committees  Informed consent was obtained from all participants

Transferability Provision of information about the research context to enable transferability findings to other contexts to be assessed

 The research field was described  Conditions under which the propositions may apply were described.

Table 65: Quality of the Intensive Research Process The interpretive rigor was enhanced by integrative and synergistic use of both quantitative and qualitative data. There was integration of research objectives and data analysis. I used constant comparative method, situational analysis and testing of theoretical ideas against both quantitative and qualitative data. The comparative approach was applied across both the intensive and extensive arms of the study. The theoretical literature was

436


reviewed following the emergence of conceptual model to enhance theoretical sensitivity and the presentation of the results included extensive use of participants’ quotes to support interpretation. I used the emerging theoretical concepts as criteria to limit the number of interviews and focus groups. The abductive theory development used triangulation, extant theory and Inference to the Best Explanation (Hill 1965; Thagard 1978; Ward 2009) to assess the emerging theory. Reflectivity was maintained by early acknowledging my theoretical perspective and prior knowledge. An initial conceptual framework was described to make explicit this prior position and Chapter 3 summarised my prior knowledge of related empirical and theoretical research. Distance from the subjects was achieved by intentionally engaging a research assistant to conduct the focus groups. The ethical and local political aspects of the research were described including the ethics approval. It is uncertain to what extent the findings of the intensive study will be transferrable. The intention of this aspect is not generalisability but rather explanation. The explanations provided were consistent with those found in other studies suggesting that some transferability might be possible. 10.3.2.3

Quantitative Data sources

Individual level variables The study used secondary data sources. The independent variables available for study were limited to those included in the IBIS self-report survey. Consequently I was unable to report on important variables that might have been included if the survey had been specifically designed for the study of perinatal and postpartum depression. Significantly there was limited information on personality trait, psycho-pathological factors, life events and lifestyle behaviours. The IBIS data set had been specifically designed to be linked to the local obstetric data-set. I was only able to undertake this linkage for the Local Government area of Fairfield, Campbelltown, Camden and Wollondilly. Thus key demographic variable such as: age of mother, English language ability, parity, and country of birth were not available for the complete sample of women. A sub-analysis using cross tabulation and logistic regression did not find any association of EDS>10 or EDS>12 with age of mother or parity. Selection bias may have occurred from refusal and non-response in the study population. Importantly not all households of births in the study period were surveyed. The households not questioned with the IBIS questionnaire include mothers who moved to “out of area” locations or mothers who may have refused the visit offered. This may introduce

437


biases as the population who refused an early childhood nurse visit may represent a particular socio-demographic sample. Observational (information) biases may have been present in the survey data. This could have arisen from recall bias, or interviewer / responder bias. A particular problematic feature of self reporting surveys is the mental status of the subject. Depressed women are more likely to have a negative view of their circumstances. This must be taken into account when considering the association found in this study of high EDS with subjective variables such as Rating of own health, “No” reluctance to leave the suburb”, and Difficult financial situation. The EDS is administered in English or via an interpreter where the mother is non-English speaking (NESP). I was unable to report in this study on the percentage of NESP mothers in the full sample but the percentage of NESP mothers in the Local Government Areas of Fairfield, Campbelltown, Camden and Wollondilly (linked data) was 11.8 percent. This may be an important source of information bias in this study. Specific non-English EDS are not currently used in SWS but could be considered for future use. The EDS has, nevertheless, been validated for a number of other languages and ethnic groups, including: Spain (Garcia-Esteve, Ascaso et al. 2003), Turkey (Inandi, Elci et al. 2002), Nepal (Regmi, Sligl et al. 2002), Norway (Eberhard-Gran, Eskild et al. 2001), Japan (Yoshida, Yamashita et al. 2001), India (Banerjee, Banerjee et al. 2000), French speaking Quebec (Des Rivieres-Pigeon, Seguin et al. 2000), China (Zhang, Chen et al. 1999), Germany (Bergant, Nguyen et al. 1998), Sweden (Bagedahl-Strindlund and Monsen Borjesson 1998), Brazil (Da-Silva, Moraes-Santos et al. 1998), Arabic (Ghubash, AbouSaleh et al. 1997), Malaysia (Kit, Janet et al. 1997), and Portugese (Areias, Kumar et al. 1996). South Western Sydney studies have found that Vietnamese and Arabic translations of EDS were acceptable to the women and appear to be suitable screening instruments for postnatal distress and depression in these populations (Matthey, Barnett et al. 1997; Barnett, Matthey et al. 1999). Ecological variables The validity of ecological measures of constructs such as social capital, social cohesion, social isolation, and social exclusion are yet to be established. During the emergent exploratory phase the variables selected for study were informed by the prior knowledge and the concurrent findings of the qualitative studies.

438


Ecological variables were not available for a number of theoretical domains emerging from the quantitative and qualitative studies. In particular I was unable to access measures of “social capital” included in the NSW Health Survey as the enumeration was small. There were limited secondary data sources for access to health services and no ecological measures of the physical nature of the suburbs. The lack of measures of physical environment is a significant weakness of this study. Personal examination of suburbs with low SEIFA scores identified limited evidence of neighbourhood physical decay as reported in similar US and UK studies. Interestingly there was limited concern regarding physical decay arising from the qualitative study, although there was concern regarding temperature control of housing and access to transport. An ecological measure of access to public transport was not developed for this study. Such a measure may be feasible for the South West Sydney region. Crime statistics were only available at the post-code level. Suburbs were assigned a crime score based on their postcode score. The validity of this approach was not tested. I used two aggregated variables (percent social networks and percent “no regret leaving suburb”). I acknowledged the possibility of same-source bias as identified by Radenbush and others(Duncan and Raudenbush 1999; Radenbush and Sampson 1999). As observed by O’Campo (2003) those whose physical or mental health status is poor may rate their neighbourhoods less favourably compared with their healthier neighbours and consequently poor health and poor neighbourhood quality will be highly correlated. Another limitation of the aggregated social network variable was the failure to demonstrate relationships with individual level depressive symptoms when controlling for individual level social network. The obvious implication is that the effect of neighbourhood social networks may be mediated through the individual level social network. In this situation the options are to either not control for individual level social networks or to utilise multi-level structural modelling techniques. This limitation is discussed further below.

439


Areal Units of Measure The suburb of residence was chosen as the closest group-level administrative unit to naturally occurring local neighbourhood environments. The limitations of using administratively defined areal units are well described (Diez Roux 2001; O'Campo 2003; Diez Roux 2004). The administrative boundaries may not align with those of the naturally occurring neighbourhoods. Diez Roux (2004) argues that ideally the collection of new data should be based on theoretically defined areas. In practice this approach is difficult to achieve for large studies. Non-spatial regression techniques treat each area unit in isolation so that there is no recognition in the model of spatial relationships that may exist between a set of related suburbs. A key feature of the Bayesian spatial ecological and multilevel models used in this study is that they incorporated these spatial effects into the modelling. 10.3.2.6

Quantitative Methods

The extensive quantitative methods used in this study included a large cross-sectional population study and multi-level spatial analysis. The analytical methods used included various forms of factor analysis and both frequentist and Bayesian regression studies. The strength of these approaches is their ability to provide descriptive “representative” generalisations (Sayer 2000). Sayer (Sayer 2000, p 21) notes, however, that “although representative of the whole population [extensive studies] are unlikely to be generalisable to other populations at different times and places [and] have limited explanatory power”. Assessment of the quality of such extensive or quantitative epidemiological studies is assessed using criteria as described by Rothman and Greenland (1998) which includes consideration of prevision (lack of random error), validity (lack of systematic error), internal validity and generalisability. Precision The large sample size noted above contributed the precision of this study. External Validity The mean age that the EDS was administered was 2.16 weeks post partum. This is earlier than most other studies which administer assessments in the first 6 – 12 weeks after delivery. The results could potentially be influenced by “Baby/Maternity Blues” experienced in the first few weeks with an overestimate of the prevalence. As with other large prospective cohort studies the postpartum population prevalence of EDS scores is an indicator of possible maternal depression. It was not possible in this

440


study to determine the incidence of new cases over the entire perinatal period. Ideally a longitudinal study is required to accurately assess the incidence of new cases in the perinatal (antenatal plus postpartum) period. The EDS screening tool used is widely accepted for use by maternal child health nurses in primary care. Validation studies of the EDS have demonstrated 68 percent - 86 percent sensitivity and 78 percent - 96 percent specificity and in an Australian sample, 100 percent sensitivity and 89 percent specificity. The positive predictive value has been reported between 70 percent-90 percent. This study reports on both EDS > 10 and EDS >12. Buist, and others (2002) note that using EDS>12 is a sound option for reducing false positives. Using EDS>10 is also useful as many women experience considerable dysfunction and want assistance (Brown, Davey et al. 2001; Buist, Barnett et al. 2002). Internal Validity Internal validity refers to validity of inference for the source population of study subjects and in studies of causation it implies accurate measurement of effects apart from random errors (Rothman and Greenland 1998, p 118). Various types of biases affect internal validity and can be considered of three types: selection bias, information bias and confounding. In this study mothers were visited within the first few weeks of childbirth. It is known that some mothers were not visited. Of the known 25,610 births in the study area 21,991 (86%) were interviewed. Based on unpublished studies on this same population (not presented here) it is highly likely that those not included were disadvantaged mothers. This would be a systematic error which I was unable to correct for. The bias would result in an underestimation of the impact of variables associated with social disadvantage and social exclusion both of which were found to be significant in this study. Information bias may have also been present in the responses to the IBIS survey form. This may have been due to the nature of the mothers self-report or to the nurses application of the questions and recording of maternal responses. Non-differential misclassification tends to underestimate any effect. It is of concern but as Rothman and Greenland observe the significance of non-differential misclassification depends heavily on whether the results are perceived as “positive” or “negative”. In this study the EFA and regression study results were consistent with previous studies. Of greater concern is differential misclassification which occurs when the classification error is dependent upon the value of other variables. Throughout this study I have assumed that mothers with depressive symptoms would have responded differently to other mothers. The result of

441


this bias is to either exaggerate or underestimate an effect. It was not possible to eliminate this bias either statistically, or by use of validating measures at the time of survey. Throughout the interpretation and analysis of the extensive quantitative study I have assumed that depressed mothers may have exaggerated, for example, difficult infant temperament, lack of support, poor self-reported health and financial difficulties. Confounding is an important issue in non-experimental research as here and is well described by Rothman and Greenland (1998, p 120-125). All variables that were statistically associated with increased depressive symptoms were possible confounding factors but “a confounding factor must not be affected by the exposure or the disease”. In an exploratory study where I have taken the stance that there is no a priori theory it is difficult to accurately control for such confounders. The seriousness of including “downstream” variables was such, however, that I elected to use prior knowledge and directed acyclic graphs (DAGs) to ensure that downstream variables were not entered into the regression studies. A confounding factor cannot be an intermediate step in the causal path between the exposure and the outcome. In this exploratory study where theory is emerging from the data it would have been inappropriate to postulate theoretical relationships, thus I deliberately delimited mediation analysis. This is an important caveat of the exploratory approach taken here. I have not described it as a weakness because

theory driven data analysis is appropriately undertaken during the theory confirmation studies described later in this Chapter. The role of these “extraneous risk factors” (Rothman and Greenland 1998, p 122), or distal causes is particularly problematic in multi-level regression studies where ecological level distal causes are mediated through individual level proximal causes. Of particular relevance to this study are multivariate analyses that find that distal factors, such as income, are not statistically significant when more proximal factors, such as social support, are taken into consideration (Aneshensel 2009). Aneshensel (2009) cites Turner and Lloyd (1999) who using mediational analysis did “indeed account fully for SES differences in depressive symptoms by incorporating a broad array of stressors and social psychological mediators, and demonstrating the efficacy of the stress process model for understanding mental health disparities”. In this study the regression study findings were consistent with previous studies and excluded variables were plausible confounders of variables in the final model. This was not the situation with the multi-level studies and I considered it probable that proximal mediation was occurring.

442


In this study I deliberately elected to delimit multi-level mediation analysis or multi-level structural equation modelling as being beyond that scope of this study. The possibility that such mediation might be occurring was, however, taken into account during the theory construction phase (Chapter 8).

10.3.3

Reflection

I initially entered the research field intending to undertake a quantitative hypotheticaldeductive study using frequentist multi-level modelling approaches. I was naïve of the importance of philosophy of science and the contribution that it could make. Early reading in the field of social epidemiology identified concern regarding the general lack of theoretical models to inform such hypothetical-deductive studies. In seeking to understand the process of theory building I found a paucity of literature on how to build a testable theory. Most theory building techniques seemed to be qualitative in nature and critical of quantitative methods. This was despite the claim that both quantitative and qualitative data could be used for grounded theory. Similarly most theory testing approaches were quantitative and critical of qualitative methods. Prominent social epidemiology commentators Muntaner (1999), O’Campo (2003), Krieger (2001) and Raphael (2006) drew attention to the importance of transdisciplinary approaches and possible importance of adopting a realist philosophy. I thus embarked on a critical realist theory building exercise using mixed methodology. 10.3.3.1

Critical Realism

Epidemiology has been firmly based in an empirical philosophical paradigm. As observed earlier logical empiricism (aka positivism) philosophy, together with inductivism, refutationism, and Bayesianism, has had a significant influence on scientific thought and the development of epidemiology methodology. In a critique of causal explanations in social epidemiology Muntaner (1999) argued that the problem lay in the scientific method. Muntaner speculated that “a plausible reason for the lack of explanations in social epidemiology [was] the attachment to an empiricist philosophy [and that] generation of causal explanations in social epidemiology would require abandoning the Humean notion of causality and adopting a realist philosophy that favours generating social theory in addition to observation“. I contended that critical realism might be an appropriate meta-theory for the generation of causal explanations in social epidemiology and may provide answers to the criticisms put

443


forward by Muntaner (1999), O’Campo (2003) and Raphael (2006). The development of realist methodologies in epidemiology and population health is relatively new although advances have been made in policy and program evaluation (Pawson 2006) and evidence based reviews (O'Campo, Kirst et al. 2009). Critical realist theory building has been undertaken in the related domains of disability research (Danermark and Gellerstedt 2004). Bhashar (1975) drew the distinction between intransitive and transitive knowledge. Things that are studied such as physical and social phenomena form the intransitive dimension of science while theories and discourse form part of a transitive dimension. The transitive is the realm of ideas, concepts, discourses and theories. It is possible to have fallible knowledge (transitive) knowledge of reality “which is a social product much like any other, which is no more independent of its production and the me who produce it than motor cars, armchairs, or books” (Bhaskar 1978, p 28). Thus when theories change (transitive dimension) it does not mean that what they are about also changes (Sayer 2000, p 11). By perceiving reality and knowledge about reality as two dimensions it avoids what Bhaskar terms “the epistemic fallacy” (Bhaskar 1978). This fallacy relates to confusing “what is” with “our concepts of it” thus conflating the intransitive into the transitive. But then isn’t the meta-theory of critical realism as proposed by Bhaskar also in the transitive dimension? The possibility that meta-theory might be a constraint to the development of knowledge is explored by Bhaskar and Danermark (2006) in their critique of Gustavsson’s (2004) assertion that “theoretical perspectives also risk introducing a new kind of strait-jacket”. Evident in the analysis in Chapter 9 was the strait-jacket imposed by the critical realist stratified view of reality. Critical realists have not yet determined what ontological levels might exist but argue that the levels are defined by the existence of structures with generative powers. That this meta-theory might be fallible is a possibility that must be considered. The current social epidemiological preoccupation with statistical multilevel methods seems to support the early layered theoretical models proposed by Dalgren and Whitehead (1991) and others (Kaplan 2004). While multi-level methodologists seek to model multilevel mediation, moderation, and cross-level interaction could it be that multilevel statistical model is itself fallible (Oakes 2004). The statistical techniques currently in

444


vogue are seeking to address matters of atomistic and ecological fallacy within clustered populations and questions of composition or context. But that paradigm may be irrelevant if causal mechanisms reside in policy, social and cultural realms that have no, or limited, spatial structure. Here Inference to the Best Explanation may be appropriate, as it was for Darwin, Huygens, and Newton (Thagard 1978), until computation advances are made in areas such as probabilistic networks (Thagard 2000), dynamic-agent models (Auchincloss and Diez Roux 2008) and multilevel spatial structural modelling (Hossain and Laditka 2009). In adopting a critical realist approach I endeavoured to follow the critical realist metatheory as espoused by Bhaskar (1975), Sayer (2000), (Archer 1995) and others. This has limitations only some of which I have discussed above. Similarly the extant theoretical approaches of epidemiology, social epidemiology and multilevel modelling may have themselves imposed theoretical limitations on the theory building described here. 10.3.3.2

Integrated Mixed Methods Design

The approach that I elected to take involved the use of both qualitative and quantitative data sources that would enable me to identify evidence about mechanisms and differences (Russo, 2009). It occurred to me that a concurrent study design would enable me to better integrate the qualitative and quantitative findings in an iterative manner. I later learnt that the quality of such integration was also used to assess the quality of the mixed method design. I have therefore undertaken a self-assessment of the quality of my mixed method approach using frameworks proposed by Teddlie and Tashakkori (2009, p300) and Onwuegbuzie and Johnson (2006). That self-assessment is at Appendix L. I have assessed that the design I have used was appropriate for exploring and describing both individual and group level variables and concepts. The approach was also appropriate for emergent, or grounded theory building. While I did not strictly adhere to the rigours of grounded theory, the study nevertheless is an example of the utility of integrating quantitative and visual data into an emergent study designs. The process did not necessarily go smoothly. The most significant “upset” was the late identification of “expectation” as an important concept resulting in the necessity to rework all the quantitative models. The magnitude of the impact was such that I elected not to rerun all the multi-level Bayesian models. If I had used a sequential design this mishap may not have occurred, but such a design would have impacted on the emergent approach and integration.

445


The design did enable good integration of purpose, objectives, data collection, analysis and triangulation. This was particularly evident in the Theory Construction Phase where the critical realist approach provided considerable rigour. Consequently I believe there was a high degree of interpretive rigour. This was particularly evident when paradoxical findings were identified in relation to country of birth and ecological social support networks. A clear strength of the critical realist approach taken was the extent to which this paradigm was able to support the epistemological, ontological, axiological, methodological and rhetorical positions of quantitative and qualitative research in field of social epidemiology.

446


10.4 Implications for Further Research 10.4.1

Methodological approaches

Critical Realism and Social Epidemiology I have used here the meta-theory of critical realism for the generation of causal explanations in social epidemiology as a response to the criticisms put forward by Muntaner (1999), O’Campo (2003) and Raphael (2006). As observed above the development of realist methodologies in epidemiology and population health is relatively new although advances have been made in policy and program evaluation (Pawson 2006) and evidence based reviews (O'Campo, Kirst et al. 2009). As observed by Muntaner (1999) the “generation of causal explanations in social epidemiology [will] require abandoning the Humean notion of causality and adopting a realist philosophy. This study demonstrates that critical realism can provide the necessary meta-theoretical philosophy for the generation of social epidemiology theory. By stratifying reality critical realism demands that the researcher examines and explains the unobserved generative forces (i.e. social exclusion) that shape the experiences of their human subjects. The fallibility of observations (and thus knowledge) is partly explained by the ontological separation of actual and observed realms together and the influence of context on the generative mechanism(s). As a meta-theory critical realism seems to be ideally suited for social epidemiology theory building and testing. Qualitative methods for confirmatory studies are well supported by critical realism (Sayer 2000; Danermark, Ekstrom et al. 2002) and realist approaches are gaining credibility in relation to evidence-based policy and programme evaluation (McGuire 2005; Pawson 2006; O'Campo, Kirst et al. 2009). From a critical realist perspective quantitative modelling can be very useful to test out possible explanations (Mingers 2006) but findings are “not assumed as closure on reality” (Olsen and Morgan 2005). The “demi-regularities” identified, and hypotheses refuted, do not necessarily reveal “laws”. Scepticism is required.

447


Explanatory Theory Building Method As noted earlier critical commentary of multi-level epidemiology studies, and social epidemiology in general, identified a lack of theoretical models informing methodology and led to calls for population health and social epidemiology theory building that identified and explained underlying causal mechanisms (Muntaner 1999; O'Campo 2003; Carpiano and Daley 2006; Dunn 2006; Raphael 2006). Two dominant approaches to theory building were identified namely: emergent theory building and confirmatory theory testing. Explanatory theory building had been described by Haig (2005) and Danermark and colleagues (2002).

For this study I have

incorporated the emergent and confirmatory theory building approaches within an overarching critical realist explanatory theory building framework. The resulting pluralistic and transdisciplinary Explanatory Theory Building Method has the potential to make a significant contribution to population health and social epidemiology theory building. The Emergent Phase draws on strong qualitative emergent and grounded theory and quantitative exploratory data analysis traditions with their strength for theory generation. The Construction Phase makes explicit the abstract analytical process of abduction and

Inference to Best Explanation (IBE) from where hypothetico-deductive theory testing typically starts and emergent theory building finishes (Lynham 2002). The Confirmatory

Phase embraces case study, probabilistic and hypothetico-deductive methods but within a realist philosophy where propositions are examined in concrete situations, “demiregularities” identified, hypotheses refuted or confirmed, but always with scepticism of establishing “laws” that will be later found fallible. This study has demonstrated that the emergent and construction phases of Explanatory

Theory Building Method can be applied to the field of social epidemiology and population health theory building as they have in disability (Danermark and Gellerstedt 2004) and development research (Olsen 2001). Confirmatory approaches within a realist philosophy have been successfully demonstrated within social epidemiology (O'Campo, Kirst et al. 2009) and health policy and programme evaluation (Greenhalgh, Humphrey et al. 2009).

448


Mixed Method Study Design As argued and demonstrated in this study both qualitative and quantitative methods are able to contribute to the emergent, explanatory and confirmatory phases of theory building. Such a pluralist approach is embraced by critical realist methodologists through what Sayer (2000) calls intensive and extensive study designs respectively. From a

philosophy of epidemiology perspective, Russo (2009) argues that evidence about both “difference-making” (c.f. regularities) and mechanisms is required for the development of causal inference. Focus groups and other qualitative methods provide insight to possible mechanisms that would not be available to the researcher through examination of quantitative regularities alone. As regularities emerge from quantitative data, concurrent qualitative methods enable questions of how and why to be asked within the same sample frame. Responding to the call for more social epidemiology theory building the critical realist mixed method design will enable epidemiologists to complement statistical studies, with which they are familiar, with qualitative methods that examine social structures, cultural and agency properties, and related triggering or conditional mechanisms. It is not uncommon for epidemiologists to use sequential mixed methodologies where qualitative studies are used to inform the subsequent quantitative design. In such situations the qualitative and quantitative studies may not be well integrated and the qualitative methods are not utilised to generate explanation. The concurrent triangulation design used here provided for strong integration with constant comparison across the concurrent studies and triangulation of findings. The study was able to demonstrate the benefits for epidemiology research of using triangulation to formulate explanation and propositions based on the empirical observations. The implication here is that mixed method research has an important role to play in future social epidemiology research.

449


Situational Analysis There are several approaches to analysing the conditions, context or situation that participants experience and that influence causative mechanisms or participants actions. Those approaches include the Conditional/Consequential Matrix, or its earlier variants (Corbin and Strauss 2008 , p 94), Context Charts (Miles and Huberman 1994 , p 102-105), the Research Map (Layder 1993) or Situational Analysis as recently described by Clarke (Clarke 2005; Clarke and Friese 2007; Clarke 2009). I elected to undertake a third cycle of qualitative analysis using the Situational Maps and Social Worlds/Arena Maps recently described by Clarke (2005). Clarke argues that situational analysis is a method suited to the interpretive philosophical tradition. Not surprising the approach worked well within this critical realist study. There has been little published work using this approach. Here the method worked well as a theory generation approach and may be found to be helpful in future studies. The typology used for the situational maps was taken from Clarke (2005) but could be modified in future studies to better reflect the critical realist ontological levels understudy. Latent Variable Theory In this study various forms of factor analysis were used both for exploratory data analysis (Tukey 1980) and abductive theory building (Haig 2005; Haig 2005). Those methods included principal factor analysis, categorical principal components analysis and specification search. The use of the generated latent variables is contentious among epidemiologists, who generally use Variable Theory with its focus on the observable.

Latent variable theory (Borsboom, Mellenbergh et al. 2003; Borsboom 2008) is more commonly used in the psychological sciences and provides a meta-theoretical approach to the use of latent variables. I contend that Latent Variable Theory can also provide the theoretical basis for its use within epidemiology and medical sciences more broadly. As observed by Muntaner (1999, p 124) “social epidemiology [is attached to an] empiricist

philosophy that searches for empirical generalizations (e.g., using observations to build models) while avoiding conjectures about the underlying social mechanisms that would help us understand how social systems work”. Consequently, based on this epistemic position, epidemiology has been reluctant to embrace the use of methods that use latent

variables. At the same time social epidemiology in particular, seeks to examine, and explain, the social determinants of health using abstract concepts such as social exclusion, poverty,

450


social capital and social cohesion. Taxonomies of observable variables are developed and individually examined with a reluctance to generate, and use, latent variables. Composite indices such as the SIEFA index are often seen as less desirable than individual observed variables such as occupation. But what is the underlying mechanism whereby “occupation” is causing the effect. Is it “loss of control”, “social exclusion”, “power”, or some other unobserved “latent” variable or construct? If social epidemiology adopts a more realist approach as proposed by Muntaner, and utilised here, then the way will be open to make greater use of latent variable related methods such as structural equation modelling and psychometric analysis. The technical methods currently available include latent variable regression, multi-level structural equation modelling, and spatial structural equation modelling. What is required is for social epidemiology to go beyond the “observed” by abandoning a Humean notion of causality, adopting a realist philosophy (Muntaner 1999) and using

Latent Variable Theory to measure the “unobserved” (Borsboom 2008). This will entail a re-examination of the current approach to variable measurement where the development and selection of variables is based on emergent qualitative studies as recently described by O’Campo and colleagues (2009) and EDA factor analysis as proposed by Haig (2005) and utilised here. Using such a transdisciplinary approach explanatory measurement models for abstract latent concepts such as social capital, social isolation and social exclusion can be developed. Such measurement models would be analogous to the measurement models used in psychometric studies of the mind and will facilitate future social epidemiological use of structural modelling methods.

451


Analytical models As discussed above (Section 10.3.4) the role played by “extraneous risk factors” or distal causes is problematic in regression studies and in particular in multi-level regression studies where ecological level causes may be mediated through individual level proximal causes. Greater examination of the role played by moderators and mediators will require the use of path analysis methods including structural modelling techniques. Recent advances in multilevel structural modelling are promising and may enable the role of proximal mediators to be elucidated. This may also assist with an understanding of the relative compositional and contextual effects of related variables. Such multi-level models will require theory such as that derived from this study. The critical realist approach is well suited to providing the theoretical basis for such models with its focus on describing both necessary and conditional mechanisms. Critical realism can also provide meta-theoretical support to statistical modelling provided the researcher maintains sceptism as to the findings. Spatial models Exploratory spatial data analysis (ESDA) is a well established methodology including the importance of visualisation as an exploratory method. The approach taken here was consistent with the methods describe by Haining (2003) and Lawson and colleagues (2003). The map decomposition strategy developed by Law and Haining (2004) had not been widely published and was used here to visually examine the co-variant spatial distribution. The approach proved useful and in particular raised questions regarding the spatial distribution of unexplained components. Bayesian methods are increasingly being used for multi-level studies but their use for spatial multilevel studies has been limited. In this study the use of spatial terms (CAR) improved the model fit. As with the spatial ecological studies, multilevel spatial studies can be displayed in GIS software thus allowing visualisation of study findings. The Bayesian multilevel spatial methods used here have not been widely published and are a new contribution to social epidemiology methodology.

452


10.4.2

Future Studies

I will briefly describe here the approach that might be followed to confirm the propositions identified in this study. As with the current study both intensive and extensive study designs will be required. Ideally the studies would be integrated as a mixed method design but this approach would not be essential. In keeping with the critical realist approach Inference to the Best Explanation would continue to be used together with the methods described below. Intensive studies For the intensive studies the propositions in this study provide a basis for future qualitative research questions. Here the research would involve examining how the different structures and mechanisms are manifest in different situations. The focus will be on examining how the proposed mechanisms interact across ontological levels and in specific context. In addition to the methods used here (i.e. emergent/grounded theory and situational analysis) other approaches might include: case studies, network analysis, concept mapping, and interactive qualitative analysis.

Case Studies Missing from this study has been an intensive study of the experiences of depressed mothers. The qualitative findings are consistent with previous qualitative studies but further confirmation and understanding of the phenomenon can be gained from conducting case studies with mothers who have, or are experiencing postnatal depression. Case studies are consistent with critical realist evaluation and confirmation approaches (Koenig 2009). Such case studies could be used to examine the propositions arising from this study. In particular case studies should include migrant women, young unsupported mothers, mothers in depressed neighbourhoods and those living in suburbs with strong aggregated social networks.

Social Network studies Social networks were identified in this study as playing an important buffering role against stress and depression. Questions arose in the study regarding the role of bridging, bonding and linking networks particularly among migrant mothers. Social network measures have been conceptualised and measured using both egocentric and sociomatric network approaches (Lakon, Godette et al. 2008). Ego-centric approaches are taken from the vantage point of the individual while sociometric networks are based on information from all respondents in a social system. Social network analysis is one research method

453


that can assist with understanding these networks. Typically, social network analysis relies on questionnaires and interviews to gather information about the relationships within a defined group. The responses gathered are then mapped. A full description of the methods available is not possible here. The purpose of a social network analysis would be to explore in an intensive manner the nature of the social ties among depressed mothers with a focus on differences across ethnic and migrant groups.

Concept mapping O’Campo and colleagues (2009) recently described a qualitative study that identified pathways by which neighbourhoods affect mental well-being. That study design is well suited for confirmatory and follow-up studies. For that study the authors used concept mapping (Kane and Trochim 2007). Concept mapping is a “structured process, focused on a topic or construct of interest, entailing input from one or more participants that produces an interpretable pictorial view (concept map) of their ideas and how these are interrelated”. The data gathering activities are completed by each individual participant and include brainstorming, sorting, rating and diagramming to represent each individual viewpoint which is incorporated into a group consensus. Both qualitative and quantitative methods are used to create a visual display of how the participants and the group conceptualise a given topic (O'Campo, Salmon et al. 2009).

Interactive qualitative analysis The interactive qualitative analysis approach (Northcutt and McCoy 2004) uses a systems method. Using this method, mothers with postnatal depression would be invited to participate. The participants are invited to produce and analyse data themselves while minimising the effect of the researcher on the content. The role of the researcher is to teach and facilitate the process. The participants are in focus groups and develop diagrams describing their shared experience. The process uses both open and axial coding leading to interview protocols, which in turn guide individual interviews. The interviews and identified ‘affinities’ then lead to mind maps, or systems diagrams that describe the phenomenon. Extensive studies For the extensive studies the testable hypotheses would be derived from the propositions and operationalised through the identification of appropriate study variables both latent and observed. The analytical studies undertaken here could be repeated on a second data set such as that pertaining to the years 2004-2006. In addition structural modelling

454


could be used to examine further the theoretical propositions and the role of moderation and mediation.

A larger data set The individual level data available for this study was from 2002-2006. This study was undertaken only on data from 2002-2003 leaving the possibility of confirming the findings on the larger data-set and/or data from 2004-2006. Group level data would be available from the 2006 Census making it possible to repeat the extensive studies conducted here. During the period 2002 to 2006 substantial migration continued to occur with changing demographic structure in some suburbs. Careful analysis could confirm or refute the findings from the current study. The longer time period would also enable spatio-temporal studies to be conducted at the ecological level using a Bayesian spatial method recently described by Norton and Niu (2009).

Structural models As discussed above the regression models in this study did not adequately explore the effects of moderation and mediation. This was particularly problematic for the multi-level models where ecological effects may have been mediated through the individual level variables. As noted in Chapter 1, I deliberately delimited path analysis and structural modelling as being beyond the scope of the current study. Structural modelling is well established within the psychological (Tolan, Gorman-Smith et al. 2003) and social sciences and can be used for both multi-level and temporal studies (Mehta and Neale 2005). It has not been widely used within epidemiology (Tu 2009). Within a critical realist philosophy the use of latent variables and path analysis is appropriate and should be used to further conduct individual, ecological and multi-level structural studies. In addition a Bayesian spatial structural modelling approach has been described (Liu, Wall et al. 2005).

455


10.5 Implications for Policy and Practice This study has identified significant spatial disparities with depressed mothers living in suburbs with low social capital, low ethnic diversity, low average adult education, low family incomes and high density. Multilevel studies found that these were predominantly compositional effects mirroring the individual level findings. Thus depressed mothers with financial difficulties, poor social networks and a non-Australian background are living in poor neighbourhoods with low social capital. While physical attributes of neighbourhoods may be important, this study identified mechanisms that are generated by social, cultural and global-economic level structures and generative mechanisms. The implications for policy and practice are broad as the stressor and buffering mechanisms might be triggered or constrained at a number of levels. For example, the Global Financial Crisis has had an impact on the Government’s ability to implement maternity leave provision, and several large employers have recently moved their manufacturing to South East Asian economies. At the same time the corporate business entities will build large shopping malls new South West Sydney communities with possible impacts on local suburbs.

10.5.1

Health services

The implications for health services of this study are many and are to found mainly in the qualitative and theory construction chapters. The implications discussed here focus on: opportunities for health early intervention by maternal and child health services, multicultural health, disparities in maternal mental health, and the observed synergy between social exclusion and low social capital. Maternal and Child Health Services The long term consequences of perinatal depression indicates that preventive interventions are warranted. The findings from this study suggest that midwife and nurse home visiting and phone contact may be helpful. A recent comprehensive review, which included a number of sustained nurse home visiting programmes, found that the most promising intervention was the provision of intensive professional postpartum support (Dennis 2005). The efficacy of home visiting for postnatal depression has recently been confirmed (Morrell, Warner et al. 2009). Social support networks were protective in this study suggesting that antenatal interventions that promote friendship groups may be beneficial. The role of antenatal groups in preventing postnatal depression has not yet

456


been confirmed (Austin 2003). But a recent study found that proactive telephone based peer support was protective (Dennis, Hodnett et al. 2009). The findings from this study and recent intervention studies indicate that there is merit in maternal and child health services continuing to develop and evaluate interventions that provide early support for mothers who are “at risk” of developing depression. The studies findings related to expectations have implications for antenatal education and counselling interventions. It may be beneficial to provide more information on the rewards and challenges of early parenthood (Harwood, McLean et al. 2007) Refugee and multicultural health The findings in this study of a strong association of postnatal depressive symptoms with maternal self-report migrant status have been found in other studies and have implications for multicultural health services. The multilevel findings suggested that for migrant mothers, social relations (and possibly bridging networks) were stronger in working class communities and protected against maternal depression. Their risk of depression was higher in communities with strong social networks, high education, and low rates of working class. These paradoxical findings suggest that migrant mothers are socially isolated in these communities. The South West Sydney region has a number of multicultural home visiting services. The implication of these findings is that multicultural home visiting support is also required for migrant mothers living in more affluent communities where they may have poor social networks. Inequalities and access The spatial analysis identified a disparity between the level of nurse home visiting, and low social status, low social capital and higher rates of depression. This inequity of access to support services can be addressed by policy decisions regarding the distribution of nursing resources. I did not undertake a quantitative analysis of the distribution of other health services, but practitioners interviewed spoke of similar disparity in relation to medical services. The implication of these findings is that greater attention should be made by health planners to the equitable distribution of primary health care services that can support mothers during the perinatal period. Mental Health Disparities This study found significant variation between suburbs in relation to maternal depression. Although much of this was compositional some of the difference was due to differences in neighbourhood socioeconomic status and racial or ethnic segregation. Thus social

457


stratification was associated with mental health disparities. This finding has implications for the distribution of mental health services including the distribution of perinatal mental health services. Community Development Services Significant statistical disparities existed across South West Sydney in relation to measures of social capital, social cohesion, and social exclusion. The qualitative findings supported the idea of “depressed communities” and provided a visual description of how mothers and families experience those communities. The regional health services currently provide a number of community development initiatives that predominantly focus on communities with public housing estates. The spatial findings of this study provide support for these initiatives but also indicate other communities with low social capital and increased measures of social exclusion. The findings from this study suggest that social networks may play an important buffering role in those communities. The implication of this finding is that community health and population health services should consider ways of increasing social ties between residents and in particular mothers and their families.

10.5.2

Interagency services

South West Sydney was one of the first regions in New South Wales to implement the Families First (now Families NSW) interagency program aimed at supporting families to raise their children. That program was a “joined up Government” approach that currently focuses on four service domains: Play groups, Family support, This study has identified the important role that social services might play in buffering maternal stress through the strengthening of social support and social inclusion. Examples include: 

Assessment of access to social support within neighbourhoods and the strengthening of structures that enable reciprocity networks for vulnerable pregnant women and mothers (for example isolated first-time mothers)



Provision of practical support for new mothers through either funded or volunteer “home help”. This may also include the development and strengthening of support roles for men

458




Advocacy for the maintenance of community facilities and upkeep such as parks and playgrounds.



Advocacy for amenities for mothers in shopping malls.

Joined up interagency approaches to improving child and family health, such as UK Sure Start initiative have had mixed results (Rutter 2006; Melhuish, Belsky et al. 2008). The study by Mulhuish and colleagues found promising outcomes not identified in earlier evaluations. While these outcomes were among children, the interventions were family focused and many commenced during pregnancy. They are therefore relevant to the study here. The local Families NSW interagency initiative has many similar elements to Sure Start and could be strengthened to support mothers during and after pregnancy using the resources of Community Services, Health, Housing, Local Government, large non-Government Organisations and community based organisations.

10.5.3

Urban planning

South West Sydney is rapidly growing with significant new urban development and redevelopment of older areas. Planning should take into account the isolation and loneliness that mothers can experience as a result of lack of transport (private or public) and poor access to places where other mothers might meet. Based on the findings of this study urban planning could make a contribution to maternal social inclusion and support networks. This could be achieved through provision of: 

Childcare centres and primary schools in a central location within neighbourhood preferably adjacent to each with provision for multi-purpose use



Community centres within neighbourhoods, that enable a variety of community activities (e.g. playgroups, craft groups, after school activities for children), to be conducted by organisations and community groups



Health, welfare and support services, such as early childhood nursing, and family support services



Halls or similar facilities that can be used for civic and cultural activities



Access to shops and convenience stores, post office, banks, ATM, places to socialise and gather after work, open spaces for recreation and libraries



Access to regular affordable public transport.

459


10.5.4

Implications for Government Policy

National and State Policies This study has implications for a broad range of Government policies. The most significant is in relation to current perinatal mental health policies and programmes. The Australian Government has recognised the importance of postnatal depression and has made a commitment to fund routine assessment of women for depression during pregnancy and in their first year of life. There are a number of initiatives but universal implementation of a comprehensive policy that also includes workforce development and robust pathways to care, is still some way off. This study gives further support to the importance of an investment in this area. In addition to the obvious importance of assessment, support and treatment of mothers with depression, this study has identified other areas where governments might intervene including economic support for families, equitable delivery of health and social services, support for migrant families and promotion of socially responsible business and media. The emphasis here is on providing a supportive environment for mothers and their partners before, during and after pregnancy. This takes the current policy emphasis of early childhood intervention to the peri-pregnancy period. Global Policies The finding that global-economic structures may impact on community and family contexts in consistent with the views of Labonte and Torgerson (2005). In their global framework for research policy and political action Labonte and Torgerson argue that Government participation in global policy agreements such as trade, labour, financial and refugee agreements can have a significant impact on domestic and community contexts. Those policies subsequently impact on mothers as outlined earlier. The implication of this study is that national and state governments should assess, or audit, the impact of such global policies and ensure that their impact on vulnerable families is mitigated.

460


10.5.6

Implications for Corporate Business

This study has identified a number of areas where corporate business may be having a negative impact on mothers, families and their neighbourhood context. That includes: loss of financial income during pregnancy and infancy, absent father during the early months, travel time, sense of loss on leaving work and impacts on communities as a result of location of business, retail and sporting franchise. Awareness of these impacts on workers and communities has led some businesses to provide supportive work practices for pregnant women, mothers and fathers and to participate in Corporate Social Responsibility (CSR) initiatives. Potential interventions by Corporate Business related to this study include: 1. Corporate and human resource management practices that demonstrate to women that pregnancy and motherhood is valued and that their status, experience and worth will not be diminished as a result including regular contact during maternity leave and supportive policies toward flexible work practice, breastfeeding and childcare 2. Corporate support for new fathers that recognises the necessity for them to emotionally and practically support their partners during pregnancy, childbirth and early infancy 3. Corporate participation in CSR initiatives that support families and promote social inclusion, social networks and strong vibrant local communities. These initiatives might include: transport to shopping malls, safe facilities for breastfeeding and infants in retail and business areas, philanthropy, staff volunteering programs, and mentoring for community groups 4. Development and utilisation of Community and Family impact assessments tools to guide corporate business and investment decisions 5. Establishment of Corporate led CSR community initiatives similar to the UK’s Industry-led ‘Business in the Community’

461


10.6 Final Comment Drawing on the call for theory building approaches to social epidemiology I have utilised critical realism and a mixed method design to construct an explanatory theory of maternal depression and context. The study will contribute to the increasing role that both realism and mixed methods are playing in explaining the social distribution and social determinants of health. The study found accumulating evidence that maternal stress, during and after pregnancy, is a cause of maternal depression and altered developmental trajectory of her infant. The focus of this study has been on the social context that initiates or conditions maternal stress. Global, economic, social and cultural mechanisms were identified that explain maternal stress and depression within family and neighbourhood contexts. There is a complex intertwining of historical, spatial, cultural, material and relational elements that contribute to the experiences of loss and nurturing. Emerging is the centrality of isolation and expectations lost as possible triggers of stress and depression not only for mothers but also others who have their dreams shattered during life’s transitions. In these situations social and cultural context can either nurture and support or marginalise and isolate. The challenge for policy and practice is to support mothers and their partners during the transition to parenthood within a challenging global-economic context.

462

Bibliography

Bibliography Abbott, M. W. and M. M. Williams (2006). "Postnatal depressive symptoms among Pacific mothers in Auckland: Prevalence and risk factors." Australian and New Zealand Journal of Psychiatry 40(3): 230-238. Abidin, R. (1995). Parenting Stress Index professional manual (3rd ed). Odessa, FL, Psychological Assessment Resources. ABS (2002). 2001 Census of Population and Housing. Canberra, Australian Bureau of Statistics. Acevedo-Garcia, D. and K. Lochner (2003). Residential Segregation and Health. Neighborhoods and Health. I. Kawachi and L. Berkman. Oxford, UK, Oxford University Press. Adamopoulos, J. (2008). On the Entanglement of Culture and Individual Behaviour Multilevel Analysis of Individuals and Cultures. F. Van de Vijver, D. Van Hemert and W. Poortinga. New York, Lawrence Erlbaum Associates. Affonso, D., S. Lovett, et al. (1990). "A standardized interview that differentiates pregnancy and postpartum symptoms from perinatal clinical depression." Birth 17: 121-130. Airey, L. (2003). ""Nae as nice a scheme as it used to be": Lay accounts of neighbourhood incivilities and well-being." Health & Place 9: 129-137. Alder, J., N. Fink, et al. (2007). "Depression and anxiety during pregnancy: A risk factor for obstetrical, fetal and neonatal outcome? A critical review of the literature." The Journal of Maternal-Fetal and Neonatal Medicine 20(3): 189-209. Almedom, A. (2005). "Social Capital and Mental Health: An Interdisciplinary Review of Primary Evidence." Social Science & Medicine 61: 943-964. Almedom, A. and D. Glandon (2008). Social Capital and Mental Health. Social Capital and Health. I. Kawachi, S. Subramanian and D. Kim. New York, Springer. Anderson, J., B. Johnstone, et al. (1999). "Breast-feeding and cognitive development: a meta-analysis." American Journal of Clinical Nutrition 70(4): 525-535. Anderson, R., P. Sorlie, et al. (1996). "Mortality effects of community socioeconomic status." Epidemiology 8: 42-47. Andrienko, N. and G. Andrienko (2006). Exploratory analysis of spatial and temporal data: a systematic approach. Berlin, Springer Verlag. Aneshensel, C. S. (1992). "Social stress: theory and research." Annual Reviews of Sociology 18: 15-38. Aneshensel, C. S. (2009). "Toward Explaining Mental Health Disparities." Journal of Health and Social Behaviour 50: 377-394.

463

Bibliography

Archer, M. (1995). Realist social theory: The morphogentic approach. Cambridge, Cambridge University Press. Areias, M. E., R. Kumar, et al. (1996). "Comparative incidence of depression in women and men, during pregnancy and after childbirth. Validation of the Edinburgh Postnatal Depression Scale in Portuguese mothers." British Journal of Psychiatry 169(1): 30-35. Arends-Toth, J., F. Van de Vijver, et al. (2006). "The influence of method factors on the relation between attitudes and self-reported behaviors in the assessment of acculturation." European Journal of Psychological Assessment 22. Asher, H., H. Weisberg, et al. (1984). Theory-Building and Data Analysis in the Social Sciences. Knoxville, Tennessee, University of Tennessee Press. Astbury, J., S. Brown, et al. (1994). "Birth events, birth experiences and social differences in postnatal depression." Australian Journal of Public Health 18: 179-184. Atkinson, A. (1998). Social exclusion, poverty and unemployment. Exclusion, Employment and Opportunity. CASE Paper number 4. A. Atkinson and J. Hills. London, Centre for Analysis of Social Exclusion. Atkinson, J., A. Paglia, et al. (2002). "Attachment security: A meta-analysis of maternal mental health correlates." Clinical Psychology Review 20: 1019-1040. Attree, P. (2004). "Growing up in disadvantage: a systematic review of the qualitative evidence." Child: Care, Health & Development 30(6): 679-689. Auchincloss, A. and A. Diez Roux (2008). "A New Tool for Epidemiology: The Usefulness of Dynamic-Agent Models in Understanding Place Effects on Health." American Journal of Epidemiology 168(1): 1-8. Austin, M. (2003). "Targeted group antenatal prevention of postnatal depression: a review." Acta Psychiatrica Scandinavica 107(4): 244-250. Austin, P. and J. Tu (2004). "Automated variable selection methods for logistic regression produced unstable models for predicting acute myocardial infarction mortality." Journal of Clinical Epidemiology 57: 1138-1146. Avishai-Eliner, S., K. Brunston, et al. (2002). "Stressed-out, or in (utero)?" Trends in Neurosciences 25(10): 518-524. Avison, W., C. S. Aneshensel, et al., Eds. (2010). Advances in the Conceptualisation of the Stress Process: Essays in Honor of Leonard I Pearlin. New York, Springer. Ayers, J., R. Hofstetter, et al. (2009). "Sorting out t competing effects of acculturation, immigrant stress, and social support in depression: A report on Korean Women in California." The Journal of Nervous and Mental Disease 197(10): 742-747. Bagedahl-Strindlund, M. and K. Monsen Borjesson (1998). "Postnatal depression: a hidden illness." Acta Psychiatrica Scandinavica 98(4): 272-275.

464

Bibliography

Banerjee, N., A. Banerjee, et al. (2000). "Evaluation of postpartum depression using the Edinburgh postnatal depression scale in evaluation of postpartum depression in a rural community in India." International Journal of Social Psychiatry. 46(1): 74-75. Banerjee, S., B. Carlin, et al. (2004). Hierarchical modeling and analysis for spatial data. Boca Raton, Florida, Chapman & Hall. Barker, D. (1992). Fetal and Infant Origins of Adult Disease. London, British Medical Journal. Barker, D., C. Hales, et al. (1993). "Type 2 (non-insulin-dependent) diabetes mellitus, hypertension and hyperlipidaemia (syndrome X): relation to reduced fetal growth. ." Diabetologia 36: 62-67. Barker, D. J. P. (1998). Mothers, babies, and health in later life. Edinburgh, Churchill Livingston. Barnett, B. E., S. Matthey, et al. (1999). "Screening for postnatal depression in women of non-English background." Archives of Women's Mental Health 2(2): 67-74. Baron, R. and D. Kenny (1986). "The moderator-mediator variable distinction in social psychological research: Conceptual, strategic and statistical considerations." Journal of Personality and Social Psychology 51: 1173-1182. Barrera, M. (1986). "Distinctions between social support concepts, measures and models." American Journal of Community Psychology 14: 413-445. Bartley, M. (2004). Health Inequality: An Introduction to Theories, Concepts and Methods. Cambridge, UK, Polity. Baum, S. (2008). Suburban scars: Australian cities and socio-economic deprivations. Urban Research Program. Research Paper 15. Brisbane, Griffith University. Beach, A., A. Henry, et al. (2005). Maternal Depression: An Adverse Early Environment. Perinatal Stress, Mood and Anxiety Disorders. From Bench to Bedside. A. Riecher-Rossler and M. Steiner. Basel, Krager. 173: 70-84. Beck, C. (1995). "Screening methods for postpartum depression." Journal of Obstetric, Gynecologic and Neonatal Nursing 24: 308-312. Beck, C., R. Froman, et al. (2005). "Acculturation level and postpartum depression in Hispanic mothers." MCN: The American Journal of Maternal/Child Nursing 30(5): 299. Beck, C. T. (1992). "The Lived Experience of Postpartum Depression: A Phenomenological Study." Nursing Research 41(3): 166-170. Beck, C. T. (1993). "Teetering on the edge: A substantive theory of postpartum depression." Nursing Research 42(1): 42-48. Beck, C. T. (1995). "The effects of Postpartum Depression on Maternal-Infant Interaction: A Meta-Analysis." Nursing Research 44(5): 298-304.

465

Bibliography

Beck, C. T. (1996). "A meta-analysis of predictors of postpartum depression." Nursing Research 45: 297-303. Beck, C. T. (2001). "Predictors of postpartum depression: an update." Nursing Research 50(5): 275-285. Beck, C. T. (2002). "Postnatal Depression: A Metasynthesis." Qualitative Health Research 12(4): 453-472. Behrens, J. (1997). "Principles and Procedures of Exploratory Data Analysis." Psychological Methods 2(2): 131-160. Ben-Shlomo, Y. and G. Davey Smith (1991). "Deprivation in infancy or in adult life: which is more important for mortality risk?" Lancet 337: 530-534. Ben-Shlomo, Y. and D. Kuh (2002). "A life course approach to chronic disease epidemiology: Conceptual models, empirical challenges and interdisciplinary perspectives." International Journal of Epidemiology 31: 285-293. Bennett, H. A., A. Einarson, et al. (2004). "Prevalence of Depression During Pregnancy: Systematic Review." Obstet Gynecol 103(4): 698-709. Bergant, A., T. Nguyen, et al. (1998). "[Prevalence of depressive disorders in early puerperium]." Gynakologisch-Geburtshilfliche Rundschau.: 232-237. Berkman, L. (1998). "The changing and heterogeneous nature of aging and longevity: a social and biomedical perspective." Annual Review of Gerontology and Geriatrics 8: 37-68. Berkman, L. and I. Kawachi (2000). A Historical Framework for Social Epidemiology. Social Epidemiology. L. Berkman and I. Kawachi. New York, Oxford University Press. Berkman, L. and I. Kawachi, Eds. (2000). Social Epidemiology. New York, Oxford University Press. Berry, J. (1997). "Immigration, acculturation, and adaptation." Applied Pscyhology: An International Review 46: 5-34. Berry, J., Y. Poortinga, et al. (2002). Cross-cultural psychology: Research and applications (2nd ed.). Cambridge, Cambridge University Press. Besag, J., J. York, et al. (1999). "Bayesian image restoration with two applications in spatial statistics." Annals of the Institute of Statistical Mathematics 43: 1-59. Bhaskar, R. (1975). A Realist Theory of Science. Leeds, Leeds Books. Bhaskar, R. (1978). A Realist Theory of Science. Hassocks, Harvester Press. Bhaskar, R. (1979). The Possibility of Naturalism. Philosophical Critique of the Contemporary Human Sciences. London, Routledge. Bhaskar, R. (1998). The Possibility of Naturalism. London, Routledge.

466

Bibliography

Bhaskar, R. and B. Danermark (2006). "Metatheory, Interdisciplinarity and Disability Research: A Critical Realist Perspective." Scandinavian Journal of Disability Research 8(4): 278-297. Bina, R. (2008). "The Impact of Cultural Factors Upon Postpartum Depression: a Literature Review." Health Care for Women International 29: 568-592. Birkeland, R., J. Thompson, et al. (2005). "Adolescent motherhood and postpartum depression." Journal of Clinical Child and Adolescent Psychology 34(2): 292-300. Bisman, J. (2002). The critical realist paradigm as a approach to research in accounting. 2002 AAANZ Conference. Perth, Australia. Blakely, T. and S. Subramanian (2006). Multilevel Studies. Methods in Social Epidemiology. J. Oakes and J. Kaufman. San Francisco, John Wiley & Sons, Inc. Blakely, T. and A. Woodward (2000). "Ecological effects in multi-level studies." Journal of Epidemiology & Community Health 54(5): 367-374. Bobak, M., M. Murphy, et al. (2003). "Determinants of adult mortality in Russia: estimates from sibling data." Epidemiology 14: 603-611. Bonner, A. (2006). Social Exclusion and The Way Out: An individual and community response to human social dysfunction. Chichester, John Wiley & Sons Ltd. Borsboom, D. (2008). "Latent Variable Theory." Measurement 6: 25-53. Borsboom, D., G. Mellenbergh, et al. (2003). "The Theoretical Status of Latent Variables." Psychological Review 110(2): 203-219. Bourdieu, P. (1977). Outline of a Theory of Practice. Cambridge, Cambridge University Press. Bourdieu, P. (1986). The forms of capital. Handbook of theory and research for the sociology of education. J. Richardson. New York, Greenwood. Bourdieu, P. and L. Wacquant (1992). An invitation to reflexive sociology. Chicago, University of Chicago Press. Boyatzis, R. (1998). Transforming qualitative information: Thematic analysis and code development, Sage Publications, Inc. Boyce, P., J. Stubbs, et al. (1993). "The Edinburgh postnatal depression scale: Validation for an Australian sample." Australian and New Zealand Journal of Psychiatry 27: 472-476. Boyce, P. M., G. Parker, et al. (1991). "Personality as a vulnerability factor to depression." British Journal of Psychiatry 159: 106-114. Bronfenbrenner, U. (1979). The ecology of human development: Experiments by nature and design. Cambridge, MA, Harvard University Press. Brown, S., M. Davey, et al. (2001). Victorian survey of recent mothers 2000. Melbourne, Centre for Mothers and Babies, Latrobe University.

467

Bibliography

Brown, S. and J. Lumley (2000). "Physical health problems after childbirth and maternal depression at six to seven months postpartum." International Journal of Obstetrics & Gynaecology 107(10): 1194-1201. Brown, S., J. Lumley, et al. (1994). Missing voices: The experience of motherhood. New York, Oxford University Press. Browne, W. and D. Draper (2006). "A comparison of Bayesian and likelihood-based methods for fitting multilevel models." Bayesian Analysis 3: 473-514. Brunner, E. J. (2000). Toward a New Social Biology. Social Epidemiology. L. Berkman and I. Kawachi. New York, Oxford: 306-311. Buist, A., B. E. Barnett, et al. (2002). "To screen or not to screen - that is the question in perinatal depression." The Medical Journal of Australia 177(Supplement): S101S105. Buka, S., R. Brennan, et al. (2003). "Neighbourhood Support and the Birth Weight of Urban Infants." American Journal of Epidemiology 157(1): 1-8. Byrne, B. (2010). Structural Equation Modeling with AMOS. Basic Concepts, Applications, and Programming. Second Edition. New York, Routledge. Cacioppo, J. and L. Hawkley (2003). "Social Isolation and Health with an Emphasis on Underlying mechanisms." Perspectives in Biology and Medicine 46: S39-S52. Campbell, S. B. and J. F. Cohn (1991). "Course and correlates of postpartum depression in first time mothers." Development and Psychopathology 4: 29-47. Carpiano, R. (2006). "Toward a neighborhood resource-based theory of social capital for health: Can Bourdieu and sociology help." Social Science & Medicine 62(1): 165175. Carpiano, R. (2008). Actual and Potential Neighbourhood Resources for Health: What can Bourdieu offer for understanding. Social Capital and Health. I. Kawachi, S. Subramanian and D. Kim. New York, Springer. Carpiano, R. and D. Daley (2006). "A guide and glossary on postpositivist theory building for population health." Journal of Epidemiology & Community Health 60: 564-570. Casciano, R. and D. Massey (2008). "Neighborhoods, employment, and welfare use: Assessing the influence of neighborhood socioeconomic composition." Social Science Research 37: 544-558. Caughy, M., P. O'Campo, et al. (2003). "When being alone might be better: neighbourhood poverty, social capital, and child mental health." Social Science & Medicine 57(2): 227-237. Charlton, B. (1995). "A critique of Geoffrey Rose's 'population strategy' for preventive medicine." Journal of the Royal Society of Medicine 88: 607-610. Charmaz, K. (2006). Constructing Grounded Theory. London, Sage Publications.

468

Bibliography

Cinciripini, P., J. Blalock, et al. (2010). "Effects of an intensive depression-focused intervention for smoking cessation in pregnancy." Journal of Consulting and Clinical Psychology 78(1): 44-54. Clarke, A. (2005). Situational Analysis: Grounded Theory After the Postmodern Turn. Thousand Oakes, CA, Sage. Clarke, A. (2009). From Grounded Theory to Situational Analysis: What's New? Why? How? Developing Grounded Theory: The Second Generation. J. Morse, P. Stern, J. Corbinet al. Walnut Creek, CA, Left Coast Press. Clarke, A. and C. Friese (2007). Grounded Theorizing Using Situational Analysis. The Sage Handbook of Grounded Teory. A. Bryant and K. Charmaz. Los Angeles, Sage Publications. Cogill, S., H. Caplan, et al. (1986). "Impact of maternal post-natal depression on cognitive development of young children." British Medical Journal 292: 1165-1167. Cohen, D. and B. Crabtree (2008). "Evaluative criteria for qualitative research in health care: Controversies and recommendations." The Annals of Family Medicine 6(4): 331. Congdon, P. (2006). Bayesian Statistical Modelling. Chichester, UK, John Wiley & Sons, Ltd. Corbin, J. and A. Strauss (2008). Basics of Qualitative Research. Los Angeles, Sage Publications. Cox, J., Y. Connor, et al. (1982). "Prospective study of the psychiatric disorders of pregnancy." British Journal of Psychiatry 140: 111-117. Cox, L. J., D. Murray, et al. (1993). "A controlled study of the onset, duration and prevalence of postnatal depression." British Journal of Psychiatry 163: 27-31. Creswell, J. (1998). Qualitative inquiry and research design: Choosing among five traditions. Thousand Oakes, CA, Sage. Creswell, J., M. Fetters, et al. (2004). "Designing a mixed methods study in primary care. ." Annals of Family Medicine 2(1): 7-12. Creswell, J. and A. Tashakkori (2007). "Editorial: Differing perspectives on mixed methods research." Journal of Mixed Methods Research 1(4): 303-308. Crow, G. (2004). Social Networks and Social Exclusion: An Overview of the Debate. Social Networks and Social Exclusion. C. Phillopson, G. Allan and D. Morgan. Hants, UK, Ashgate Publishing. Cubbin, C., F. LeClere, et al. (2000). "Socioeconomic status and injury mortality: individual and neighborhood determinants." Journal of Epidemiology & Community Health 54: 517-552. Culhane, J. and I. Elo (2005). "Neighbourhood context and reproductive health." American Journal of Obstetrics & Gynecology 192(5 Suppl): s22-29.

469

Bibliography

Cummings, E. and P. Davies (1994). "Maternal depression and child development." Journal of Child Psychology & Psychiatry 35: 73-112. Da-Silva, V. A., A. R. Moraes-Santos, et al. (1998). "Prenatal and postnatal depression among low income Brazilian women." Brazilian Journal of Medical & Biological Research 31(6): 799-804. Dahlgren, G. and M. Whitehead (1991). Policies and Strategies to Promote Social Equity in Health. Stockholm, Institute for Future Studies. Danermark, B. (2002). "Interdisciplinary research and critical realism. The example of disability research." Journal of Critical Realism 5: 56-64. Danermark, B., M. Ekstrom, et al. (2002). Explaining Society: Critical realism in the social sciences. London, Routledge. Danermark, B. and L. Gellerstedt (2004). "Social justice: redistribution and recognition - a non-reductionist perspective on disability." Disability and Society 19(4): 339-353. Davey Smith, G., C. Hart, et al. (1997). "Lifetime socioeconomic position and mortality: prospective observational study." British Medical Journal 314: 547-552. Davis, E., L. Glynn, et al. (2007). "Prenatal exposure to maternal depression and cortisol influences infant temperament." Journal of the American Academy of Child & Adolescent Psychiatry 46(6): 737-746. De Silva, M., K. McKenzie, et al. (2005). "Social capital and mental illness: a systematic review." Journal of Epidemiology & Community Health 59: 619-627. Dennis, C.-L. (2005). "Psychosocial and psychological interventions for prevention of postnatal depression: systematic review." British Medical Journal 331(7507): 15. Dennis, C., E. Hodnett, et al. (2009). "Effect of peer support on prevention of postnatal depression among high risk women: multisite randomised controlled trial." British Medical Journal 338(jan15 2): a3064. Dennis, C. and K. McQueen (2009). "The relationship between infant-feeding outcomes and postpartum depression: a qualitative systematic review." Pediatrics 123(4): e736-e751. Denzin, N. and Y. Lincoln (2000). Handbook of Qualitative Research (2nd ed). Thousand Oakes, CA, Sage Publications. Des Rivieres-Pigeon, C., L. Seguin, et al. (2000). "[The Edinburgh Postnatal Depression Scale: the validity of its Quebec version for a population low socioeconomic status mothers]." Canadian Journal of Community Mental Health. 19(1): 201-214. Diaz, M., H. Le, et al. (2007). "Interpersonal factors and perinatal depressive symptomatology in a low-income Latina sample." Cultural Diversity and Ethnic Minority Psychology 13(4): 328. DiClemente, R., R. Crosby, et al., Eds. (2002). Emerging Theories in Health Promotion Practice and Research. San Francisco, CA, Jossey-Bass.

470

Bibliography

Diez Roux, A. (2001). "Investigating Neighbourhood and Area Effects on Health." American Journal of Public Health 91(11): 1783-1789. Diez Roux, A. (2003). The Examination of Neighborhood Effects on Health: Conceptual and Methodological Issues Related to the Presence of Multiple Levels of Organization. Neighborhoods and Health. I. Kawachi and L. Berkman. Oxford, UK, Oxford University Press. Diez Roux, A. (2004). "The Study of Group-level Factors in Epidemiology: Rethinking Variable, Study Designs, and Analytical Approaches." Epidemiologic Reviews 26: 104-111. Diez Roux, A. (2008). "Next steps in understanding the multilevel determinants of health." Journal of Epidemiology & Community Health 62(11): 957. Diez Roux, A., F. Nieto, et al. (1997). "Neighbourhood environments and coronary heart disease: a multilevel analysis." American Journal of Epidemiology 146: 48-63. Dixon-Woods, M., S. Agarwal, et al. (2005). "Synthesising qualitative and quantitative evidence: a review of possible methods." Journal of Health Services Research and Policy 10(1): 45-53. DOH (2003). "NSW Mothers and Babies 2002." NSW Public Health Bulletin 14(S-3). Downey, G. and J. Coyne (1990). "Children of depressed parents: An integrative review." Psychological Bulletin 108: 50-76. Downward, P., J. Finch, et al. (2002). "Critical realism, empirical methods and inference." Cambridge Journal of Economics 26(4): 481-500. Downward, P. and A. Mearman (2007). "Retroduction as mixed-methods triangulation in economic research: reorienting economics into social science." Cambridge Journal of Economics 31: 77-99. Draper, D. (2008). Bayesian Multilevel Analysis and MCMC. Handbook of Multilevel Analysis. J. de Leeuw and E. Meijer. New York, Springer. Dubin, R. (1969). Theory Building (Revised Edition). New York, Free Press. Duncan, C., K. Jones, et al. (1995). "Psychiatric morbidity: a multilevel approach to regional variations in the UK." Journal of Epidemiology & Community Health 49(3): 290-295. Duncan, G. and S. Raudenbush (1999). "Assessing the effects of context in studies of child and youth development." Educational Psychology 34: 29-41. Dunn, J. (2006). "Speaking theoretically about population health." Journal of Epidemiology and Community Health 60(7): 572. Earls, F. and M. Carlson (2001). "The Social Ecology of Child Health and Well-Being." Annual Review of Public Health 22: 143-166.

471

Bibliography

Eastwood, J. (2005). Postpartum Depression: A Population Based Study. School of Public Health and Community Medicine. Sydney, University of New South Wales. Masters of Public Health: 68. Eberhard-Gran, M., A. Eskild, et al. (2001). "The Edinburgh Postnatal Depression Scale: validation in a Norwegian community sample." Nordic Journal of Psychiatry. 55(2): 113-117. Ebrahim, S. and E. Lau (2001). "Commentary: Sick populations and sick individuals." International Journal of Epidemiology 30: 433-434. Eco, U. (1984). Semiotics and the Philosophy of Language. London, Macmillan. Ekbom, A., C. Hsieh, et al. (1996). "Perinatal characteristics in relation to incidence of and mortality from prostate cancer." British Medical Journal 313: 337-341. Ellen, I., T. Mijanovich, et al. (2001). "Neighbourhood effects on health: exploring the links and assessing the evidence." Journal of Urban Affairs 23(3-4): 391-408. Elo, L. and S. Preston (1992). "Effects of early-life conditions on adult mortality: A review." Population Index 58: 186-212. Environmental Systems Research Institute (2008). ArcView 9.3. Redlands, California. Evans, R. and G. Stoddart (1990). "Producing health, consuming health care." Social Science & Medicine 31: 1347. Fergusson, D., L. Horwood, et al. (1996). "Changes in depression during and following pregnancy. ALSPAC Study Team. Study of pregnancy and children." Paediatric and Perinatal Epidemiology 10: 279-293. Field, T., M. Diego, et al. (2006). "Prenatal depression effects on the fetus and newborn: a review." Infant Behavior and Development 29(3): 445-455. Fleetwood, S. (2010, 14 February 2010). "Causal Laws and Tendencies." Retrieved 14 February, 2010, from http://carecon.org.uk/QM/Conference%202008/Papers/Fleetwod.pdf. Florey, L., S. Galea, et al. (2007). Macrosocial Determinants of Population Health in the Context of Globalization. Macrodeterminants of Population Health. S. Gale. Ann Arbor, MI, Springer. Fonagy, p. (1996). Patterns of attachment, interpersonal relationships and health. Social organization and health: toward a health policy for the 21st century. D. Blane, E. J. Brunner and R. Wilkinson. London, Routledge: 125-151. Fone, D. and F. Dunstan (2006). "Mental health, places and people: A multilevel analysis of economic inactivity and social deprivation." Health & Place 12: 332-344. Fone, D., F. Dunstan, et al. (2007). "Does social cohesion modify the association between area income deprivation and health? A multilevel analysis." International Journal of Epidemiology 36: 338-345.

472

Bibliography

Fone, D., K. Lloyd, et al. (2007). "Measuring the neighbourhood using UK benefits data: a multilevel analysis of mental health status." BMC Public Health 7: 69. Forman, D. N., P. Videbech, et al. (2000). "Postpartum depression: identification of women at risk." British Journal of Obstetrics & Gynaecology 107: 1210-1217. Galea, S., Ed. (2007). Macrosocial Determinants of Population Health. Ann Arbor, Springer. Galobardes, B., J. W. Lynch, et al. (2004). "Childhood Socioeconomic Circumstances and Cause-specific Mortality in Adulthood: Systematic Review and Interpretation." Epidemiologic Reviews 26: 7-21. Galster, G. (2004). "Peer Review of Segregation Measures in Iceland and Weinberg (2002) and corresponding Census website." Retrieved November 2008, from http://www.census.gov/hhes/www/housing/housing_patterns/pdf/galster.pdf. Garcia-Esteve, L., C. Ascaso, et al. (2003). "Validation of the Edinburgh Postnatal Depression Scale (EPDS) in Spanish mothers." Journal of Affective Disorders 75(1): 71-76. Gelford, D. and D. M. Teti (1990). "The effects of maternal depression on children." Clinical Psychology Review 10: 329-353. Ghubash, R., M. T. Abou-Saleh, et al. (1997). "The validity of the Arabic Edinburgh Postnatal Depression Scale." Social Psychiatry & Psychiatric Epidemiology 32(8): 474-476. Gifi, A. (1990). Nonlinear multivariate analysis, Chichester : John Wiley & Sons. Gilman, S. E., I. Kawachi, et al. (2002). "Socioeconomic status in childhood and the lifetime risk of major depression." International Journal of Epidemiology 31: 359367. Glanz, K., B. Rimer, et al. (2002). Theory, research, and practice in health behaviour and health education. Health behaviour and health education: Theory, research, and practice. 3rd Edition. K. Glanz, B. Rimer and F. Lewis. San Francisco, JosseyBass. Glaser, B. (1978). Theoretical sensitivity: Advances in the methodology of grounded theory. Mill Valley, Sociological Press. Glaser, B. and A. Strauss (1967). The discovery of grounded theory: Strategies For qualitative research. Chicago, Aldine. Glover, V. and T. O'Connor (2002). "Effects of antenatal stress and anxiety: Implications for development and psychiatry." British Journal of Psychiatry 180: 389-391. Gluckman, P. and M. Hanson, Eds. (2006). The developmental Origins of Health and Disease. Cambridge, Cambridge University Press. Gluckman, P., M. Hanson, et al. (2007). "Early Life Events and Their Consequences for Later Disease: A Life History and Evolutionary Perspective." American Journal of Human Biology 19: 1-19.

473

Bibliography

Gotlib, I. H., V. E. Whiffen, et al. (1989). "Prevalence rates and demographic characteristics associated with depression in pregnancy and the postpartum." Journal of Consulting and Clinical Psychology 57(2): 269-274. Gotlib, I. H., V. E. Whiffen, et al. (1991). "Prospective investigation of postpartum depression: Factors involved in onset and recovery." Journal of Abnormal Psychology 100(2): 122-132. Grbich, C. (2007). Qualitative data analysis: An Introduction. London, Sage Publications. Green, L. and M. Kreuter (1999). Health Promotion Planning. An Educational and Ecological Approach. Third Edition. Mountain View, CA, Mayfield Publishing Company. Green, L. and J. Ottoson (1999). Community and population health, 8th Edition. Boston, McGraw-Hill. Greenhalgh, T., C. Humphrey, et al. (2009). "How Do You Modernize a Health Service? A Realist Evaluation of Whole-scale Transformation in London." The Milbank Quarterly 87(2): 391-416. Greenhalgh, T. and R. Taylor (1997). "How to read a paper: Papers that go beyond numbers (qualitative research)." British Medical Journal 315(7110): 740-743. Greenland, S. (1989). "Modelling and Variable Selection in Epidemiologic Analysis." American Journal of Public Health 79(3): 340-349. Greenland, S. (1992). "Divergent biases in ecological and individual-level studies." Statistics in Medicine 11: 1209-1223. Greenland, S. (1998). "Probability Logic and Probabilistic Induction." Epidemiology 9(3): 322-332. Greenland, S., M. Gago-Dominguez, et al. (2004). "The Value of Risk-Factor ("Black-Box") Epidemiology." Epidemiology 15(5): 529-535. Greenland, S., J. Pearl, et al. (1999). "Causal diagrams for epidemiological research." Epidemiology 10(1): 37-47. Groenewald, T. (2004). "A phenomenological research design illustrated." International Journal of Qualitative Methods 3(1). Guba, E. and Y. Lincoln (1989). Fourth generation evaluation. Newbury Park, CA, Sage. Guba, E. and Y. Lincoln (1994). Competing paradigms in qualitative research. Thousand Oakes, Sage Publication. Guba, E. and Y. Lincoln (2005). Paradigmatic controversies, contradictions, and emerging confluences. Handbook of qualitative research. N. Denzin and Y. Lincoln. Thousand Oakes, CA, Sage. Gustavsson, A. (2004). "The role of theory in disability research - springboard or straitjacket?" Scandinavian Journal of Disability Research 6(1): 55-70.

474

Bibliography

Haack, S. (2004). "An Epistemologist Among the Epidemiologists." Epidemiology 15(5): 521-523. Haig, B. (1995). "Grounded Theory as Scientific Method." Retrieved 14 July, 2006, from http://www.ed.uiuc.edu/EPS/PES-Yearbook/95_docs/haig.html. Haig, B. (2005). "An Abductive Theory of Scientific Method." Psychological Methods 10(4): 371-388. Haig, B. (2005). "Exploratory Factor Analysis, Theory Generation and Scientific Method." Multivariant Behavioural Research 40(3): 303-329. Haining, R. (2003). Spatial Data Analysis: Theory and Practice. Cambridge, Cambridge University Press. Hair, J., R. Anderson, et al. (1998). Multivariate Data Analysis, Fifth Edition. New Jersey, Prentice Hall. Halbreich, U. and S. Karkun (2006). "Cross-cultural and social diversity of prevalence of postpartum depression and depressive symptoms." Journal of Affective Disorders 91(2-3): 97-111. Halfon, N. and M. Hochstein (2002). "Life Course Health Development: An Integrated Framework for Developing Health, Policy and Research." The Milbank Quarterly 80(3): 433-479. Hall, L. A., J. B. Kotch, et al. (1996). "Self-esteem as a mediator of the effects of stressors and social resources on depressive symptoms in postpartum mothers." Nursing Research 45: 231-238. Harwood, K., N. McLean, et al. (2007). "First-Time Mothers' Expectations of Parenthood: What happens when optimistic expectations are not matched by later experiences." Developmental Psychology 43(1): 1-12. Hatton, D., J. Harrison-Hohner, et al. (2007). "Missed antenatal depression among high risk women: a secondary analysis." Archives of Women's Mental Health 10: 121123. Health, N. D. o. (2006). New South Wales Population Health Survey 2003-2004. Sydney, NSW Department of Health. Hedegaard, M., T. Henriksen, et al. (1996). "The relationship between psychological distress during pregnancy and birth weight for gestational age." Acta Obstetricia et Gynecologica Scandinavica 75. Heller, K., R. Price, et al. (1984). Psychology and community change: Challenges of the future. Homewood, IL, Dorsey. Hellevik, O. (1984). Introduction to Causal Analysis. Exploring Survey Data by Crosstabulation. London, George Allen & Unwin. Henderson, J., S. Evans, et al. (2003). "Impact of Postnatal Depression on Breastfeeding Duraction." Birth 30(3): 175-180.

475

Bibliography

Henrichs, J., J. Schenk, et al. (2009). "Maternal Pre-and Postnatal Anxiety and Infant Temperament. The Generation R Study." Infant and Child Development 18(3). Hernandez-Diaz, S., E. Schisterman, et al. (2006). "The Birth Weight "Paradox" Uncovered?" American Journal of Epidemiology 164(11): 1115-1120. Herring, S., J. Rich-Edwards, et al. (2008). "Association of postpartum depression with weight retention 1 year after childbirth." Obesity 16(6): 1296-1301. Hertzman, C., C. Power, et al. (2001). "Using an interactive framework of society and lifecourse to explain self-rated health in early adulthood." Social Science & Medicine 53(12): 1575-1585. Hill, A. (1965). "The environment and disease: association or causation?" Proceedings of the Royal Society of Medicine 58: 295-300. Hinkle, E. (1987). "Stress and Disease: The concept after 50 years." Social Science & Medicine 25(6): 561-566. Hobel, C. and J. Culhane (2003). "Role of Psychosocial and Nutritional Stress on Poor Pregnancy Outcome." Journal of Nutrition 133: 1709S-1717S. Hoffman, S. and M. Hatch (2000). "Depressive symptomatology during pregnancy: Evidence for an association with decreased fetal growth in pregnancies of lower social class women." Health Psychology 19: 535-543. Hogue, C., S. Hoffman, et al. (2001). "Stress and preterm delivery: a conceptual framework." Paediatric and Perinatal Epidemiology 15(s2): 30-40. Hortulanus, R. and A. Machielse (2006). The issue of social isolation. Social Isolation in Modern Society. R. Hortulanus, A. Machielse and L. Meeuwesen. London, Routledge. Hortulanus, R., A. Machielse, et al. (2006). Social Isolation in Modern Society. London, Routledge. Hosmer, D. and S. Lemeshow (2000). Applied logistic regression (2nd Edition). New York, John Wiley. Hossain, M. and J. Laditka (2009). "Using hospitalization for ambulatory care sensitive conditions to measure access to primary care: an application of spatial structural equation modelling." International Journal of Health Geographics: 1-14. House, J., K. Landis, et al. (1988). "Social relationships and health." Science 241: 540545. Huizink, A. (2008). "Prenatal stress exposure and temperament: A review." European Journal of Developmental Science, 2 1(2): 77-99. Huizink, A., P. de Medina, et al. (2002). "Psychological measures of prenatal stress as predictors of infant temperament." Journal of the American Academy of Child & Adolescent Psychiatry 41: 1078-1085.

476

Bibliography

Hulse, K. and W. Stone (2007). "Social Cohesion, Social Capital and Social Exclusion: A cross cultural comparison." Policy Studies 28(2): 109-128. Hutchison, P., D. Abrams, et al. (2007). The Social Psychology of Exclusion. Multidisciplinary Handbook of Social Exclusion Research. D. Abrams, J. Christian and D. Gordon. Chichester, John Wiley & Sons Ltd. Inandi, T., O. C. Elci, et al. (2002). "Risk factors for depression in postnatal first year, in eastern Turkey." International Journal of Epidemiology 31(6): 1201-1207. Jaccard, J. and J. Jacoby (2010). Theory Construction and Model-Building Skills: A Practical Guide for Social Scientists. New York, The Guilford Press. Jeppesen, S. (2005). "Critical Realism as an Approach to Unfolding Empirical Findings." the Journal of Transdisciplinary Environmental Studies 4(1): 1-9. Johnson, G. (2004). "Small area mapping of prostate cancer incidence in New York State (USA) using fully Bayesian hierarchical modelling." International Journal of Health Geographics 2: 29. Johnstone, S. J., P. M. Boyce, et al. (2001). "Obstetric risk factors for postnatal depression in urban and rural community samples." Australian and New Zealand Journal of Psychiatry 35(1): 69-74. Kajantie, E. (2006). "Fetal origins of stress-related adult disease." Annals of the New York Academy of Sciences 1083(1 Stress, Obesity, and Metabolic Syndrome): 11-27. Kane, M. and W. Trochim (2007). Concept Mapping for Planning and Evaluation. Thousand Oakes, CA, Sage Publications. Kaplan, A. (1964). The conduct of inquiry: Methodology for behavioural science. San Francisco, CA, Chandler. Kaplan, G. (2004). "What's Wrong with Social Epidemiology, and How Can We Make It Better." Epidemiologic Reviews 26: 124-135. Kapoor, A., E. Dunn, et al. (2006). "Fetal programming of hypothalamo-pituitary-adrenal function: prenatal stress and glucocorticoids." 572(1): 31-44. Kawachi, I. and L. Berkman, Eds. (2003). Neighborhoods and Health. Oxford, Oxford University Press. Kawachi, I., S. Subramanian, et al. (2008). Social Capital and Health. Social Capital and Health. I. Kawachi, S. Subramanian and D. Kim. New York, Springer. Keating, D. and C. Hertzman (1999). Developmental health and the wealth of nations. New York, Guilford Press. Kelle, U. (2001). "Sociological Explanations Between Micro and Macro and Integration of Qualitative and Quantitative Methods. FQS (Forum: Qualitative Social Research) 2(1),." Retrieved February 2009, from Http://www.qualitative-research.net/fqs/fqseng.htm

477

Bibliography

Kelsall, J. and J. Wakefield (2002). "Modeling spatial variation in disease risk." Journal of the American Statistical Association 97(459): 692-701. Kinderman, P. (2005). "A psychological model of mental disorder." Harvard Review of Psychiatry 13(4): 206-217. Kinney, D., K. Munir, et al. (2008). "Prenatal stress and risk for autism." Neuroscience and biobehavioral reviews 32(8): 1519-1532. Kit, L. K., G. Janet, et al. (1997). "Incidence of postnatal depression in Malaysian women." Journal of Obstetrics & Gynaecology Research 23(1): 85-89. Kitto, S., J. Chesters, et al. (2008). "Quality in qualitative research. Criteria for authors and assessors in the submission and assessment of qualitative research articles for the Medical Journal of Australia." Medical Journal of Australia 188(4): 243-246. Koenig, G. (2009). "Realistic Evaluation and Case Studies." Evaluation 15(1): 9-30. Kramer, M., L. Goulet, et al. (2001). "Socio-economic disparities in preterm birth: causal pathways and mechanisms." Paediatric and Perinatal Epidemiology 15 (Suppl 2): 104-123. Kramer, M. and C. Hogue (2009). "Is Segregation Bad for Your Health." Epidemiologic Reviews 31: 178-194. Krieger, N. (1994). "Epidemiology and the web of causation: has anyone seen the spider?" Social Science & Medicine 39(7): 887-903. Krieger, N. (2000). Discrimination and Health. Social Epidemiology. L. Berkman and I. Kawachi. New York, Oxford University Press. Krieger, N. (2001). "Theories for social epidemiology in the 21st century: an ecosocial perspective." International Journal of Epidemiology 30: 668-677. Kruger, D., T. Reischl, et al. (2007). "Neighborhood Social Conditions Mediate the Association Between Physical Deterioration and Mental Health." American Journal of Community Psychology 40: 261-271. Kuh, D. and Y. Ben-Shlomo (1997). A life course approach to chronic disease epidemiology. Oxford, Oxford University Press. Kuhn, T. (1962). The structure of scientific revolutions. Chicago, University of Chicago Press. Kuhn, T. (1996). The structure of scientific revolutions. 3rd Edition. Chicago, University of Chicago Press. Kulldorff, M. (1997). "A spatial scan statistic." Communications in Statistics - Theory and Methods 26(6): 1481-1496. Kulldorff, M., E. Feuer, et al. (1997). "Breast cancer clusters in the northeast United States: A geographic analysis." American Journal of Epidemiology 146(2): 161170.

478

Bibliography

Kulldorff, M., K. Rand, et al. (1998). SaTScan v2.1: Software for the spatial and spacetime scan statistics. Bethesda, MD, National Cancer Institute. Labonte, R. (1999). "Social Capital and community development: practitioner emptor." Australian and New Zealand Journal of Public Health 23: 430-433. Labonte, R. and R. Torgerson (2005). "Interrogating globalization, health and development: Toward a comprehensive framework for research, policy and political action." Critical Public Health 15(2): 157-179. Lakon, C., D. Godette, et al., Eds. (2008). Network-Based Approaches for Measuring Social Capital. New york, Springer. Lancaster, C., K. Gold, et al. (2010). "Risk factors for depressive symptoms during pregnancy: a systematic review." American Journal of Obstetrics and Gynecology 202(1): 5-14. Lane, A., R. Keville, et al. (1997). "Postnatal depression and relation among mothers and their partners: Prevalence and predictors." British Journal of Psychiatry 171: 550555. Lansakara, N., S. Brown, et al. (2009). "Birth Outcomes, Postpartum Health and Primary Care Contacts of Immigrant Mothers in an Australian Nulliparous Pregnancy Cohort Study." Maternal and Child Health Journal Retrieved 24 March, 2010, from http://www.springerlink.com.ezproxy1.library.usyd.edu.au/content/a925q325hnr11 257/fulltext.pdf. Law, J. and R. Haining (2004). "A Bayesian Approach to Modeling Binary Data: The Case of High-Intensity Crime Areas." Geographical Analysis 36(3): 197-216. Lawson, A., W. Browne, et al. (2003). Disease Mapping with WinBUGS and MLwiN. Chichester, England, John Wiley and Sons Ltd. Layder, D. (1993). New Strategies in Social Research: An Introduction and Guide. Cambridge, UK, Polity Press. Leeder, S. and A. Dominello (1999). "Social capital and its relevance to health and family policy." Australian and New Zealand Journal of Public Health 23: 424-429. Leigh, B. and J. Milgrom (2008). "Risk factors for antenatal depression, postnatal depression and parenting stress." BMC Psychiatry 8: 24. Leon, D., I. Kupilova, et al. (1996). "Failure to realise growth potential in utero and adult obesity in relation to blood pressure in 50 year old Swedish men." British Medical Journal 312: 401-406. Lipton, P. (2004). Inference to the Best Explanation. 2nd Edition. London, Routledge. Lithell, H., P. McKeigue, et al. (1996). "Relation of size at birth to non-insulin dependent diabetes and insulin concentrations in men aged 50-60 years." British Medical Journal 312: 406-410. Liu, X., M. Wall, et al. (2005). "Generalised spatial structural equational models." Biostatistics 6(4): 539-557.

479

Bibliography

Logsdon, C., J. C. Birkimer, et al. (2000). "The link of social support and postpartum depressive symptoms in African-American women with low incomes." The American Journal of Maternal/Child Nursing 25(5): 262-266. Lovejoy, C., P. Graczyk, et al. (2000). "Maternal depression and parenting behaviour: A meta-analytic review." Clinical Psychology Review 20(5): 561-592. Lundberg, O. (1993). "The impact of childhood living conditions on illness and mortality in adulthood." Social Science & Medicine 36: 1047-1052. Lupien, S., B. McEwen, et al. (2009). "Effects of stress throughout the lifespan on the brain, behaviour and cognition." Nature Reviews Neuroscience 10(6): 434-445. Lynch, J. (2000). "Social epidemiology: some observations on the past, present and future." Australasian Epidemiologist 7: 7-15. Lynch, J., G. Davey Smith, et al. (2000). "Income inequality and mortality: importance to health of individual income, psychosocial environment, or material conditions." British Medical Journal 320: 1200-1204. Lynch, J. W., G. A. Kaplan, et al. (1997). "Why do poor people behave poorly? Variation in adult health behaviours and psychological characteristics by stages of the socioeconomic life course." Social Science & Medicine 44: 809-834. Lynham, S. (2002). "The General Method of Theory-Building Research in Applied Disciplines." Advances in Developing Human Resources 4(3): 221-241. Lyons-Ruth, K., R. Wolfe, et al. (2000). "Depression and the parenting of young children: Making the case for early preventive mental health services." Harvard Review of Psychiatry 8: 148-153. Macintyre, S. and A. Ellaway (2003). Neighborhoods and Health: An Overview. Neighborhoods and Health. I. Kawachi and L. Berkman. Oxford, Oxford University Press. Macintyre, S., A. Ellaway, et al. (2002). "Place effects on health: how can we conceptualise, operationalise, and measure them?" Social Science & Medicine 55: 125-139. Macintyre, S., S. Maciver, et al. (1993). "Area, class, and health: should we be focusing on places or people?" Journal of Social Policy 22: 213-234. MacNab, Y. (2004). "Bayesian spatial and ecological models for small-area accident and injury analysis." Accident Analysis and Prevention 36: 1019-1028. MacNab, Y., Z. Qiu, et al. (2004). "Hierarchical Bayes Analysis of Multilevel Health Services Data: A Canadian Neonatal Mortality Study." Health Services and Outcomes Research Methodology 5: 5-26. Maly, M. (2000). "The Neighborhood Diversity Index: A Complementary Measure of Racial Residential Settlement." Journal of Urban Affairs 22(1): 37-47.

480

Bibliography

Martins, C. and E. Gaffan (2000). "Effects of early maternal depression on patterns of infant-mother attachment: a meta-analytic investigation." Journal of Child Psychology & Psychiatry & Allied Disciplines 41(6): 737-746. Massey, D. (2001). The prodigal paradigm returns: Ecology comes back to sociology. Does It Take a Village? Community Effects on Children, Adolescents, and Families. A. Booth and A. Crouter. Mahwah, NJ, Lawrence Erlbaum Associates Publishers: 41-48. Matthews, S. and M. Meaney (2005). Maternal Adversity, Vulnerability and Disease. Perinatal Stress, Mood and Anxiety Disorders. From Bench to Bedside. A. Riecher-Rossler and M. Steiner. Basel, Karger. Matthey, S., B. E. Barnett, et al. (1997). "Vietnamese and Arabic women's responses to the Diagnostic Interview Schedule (depression) and self-report questionnaires: cause for concern." Australian and New Zealand Journal of Psychiatry 31: 360369. Maxwell, J. (2005). Qualitative Research Design: An Interactive Approach. 2nd Ed. Thousand Oakes, Sage Publications. Mayo, D. and A. Spanos (2004). "When Can Risk-Factor Epidemiology Provide Reliable Tests." Epidemiology 15(5): 523-524. McCormick, J. (2001). "Commentary: Reflections on sick individuals and sick populations." International Journal of Epidemiology 30: 434-435. McCulloch, A. (2007). "The changing structure of ethnic diversity and segregation in England, 1991-2001." Environment and Planning A 39: 909-927. McGill, H., V. L. Benzie-Burrows, et al. (1995). "Postnatal depression: a Christchurch Study." New Zealand Medical Journal: 162-165. McGrath, J., K. Records, et al. (2008). "Maternal depression and infant characteristics." Infant Behavior & Development 31(1): 71-80. McGuire, W. (2005). "Beyond EBM: new directions for evidence-based public health." Perspectives in Biology and Medicine 48(4): 557-569. McGuire, W. (2006). Beyond EBM: Critical Realism as the Foundation for Evidence-Based Public Health. Comparative Program on Health and Society Lupina Foundation Working Papers Series 2004-2005. J. Cohen and J. Keelan. Toronto, Munk Centre for International Studies, University of Toronto: 115-129. McMichael, A. (1999). "Prisoners of the Proximate: Loosening the Constraints on Epidemiology in an age of change." American Journal of Epidemiology 149: 887897. McQuail, D. (1994). McQuails Mass Communication Theory (4th Ed.). London, Sage Publications. Meaney, M. (2010). "Epigenetics and the Biological Definition of Gene x Environment Interactions." Child Development 81(1): 41-79.

481

Bibliography

Meeuwesen, L. (2006). Health and Isolation. Social Isolation in Modern Society. R. Hortulanus, A. Machielse and L. Meeuwesen. London, Routledge. Mehta, P. and M. Neale (2005). "People Are Variables Too: Multilevel Structural Equations Modelling." Psychological Methods 10(3): 259-284. Melhuish, E., J. Belsky, et al. (2008). "Effects of fully-established Sure Start Local Programmes on 3-year-old children and their families living in England: a quasiexperimental observational study." The Lancet 372(9650): 1641-1647. Menaghann, E. (1990). "Social stress and individual distress." Research in Community Mental Health 6: 107-141. Merton, R. (1968). Social Theory and Social Structure. New York, Free Press. Meulman, J., A. Van der Kooij, et al. (2004). Principal components analysis with nonlinear optimal scaling transformations for ordinal and nominal data. Handbook of Quantitative Methodology for the Social Sciences. D. Kaplan. Thousand Oaks, CA, Sage Publications: 49-70. Miles, M. and A. Huberman (1994). Qualitative Data Analysis: An Expanded Sourcebook. Thousand Oakes, CA, Sage Publications. Millar, J. (2007). Social Exclusion and Social Policy Research: Defining Exclusion. Multidisciplinary Handbook of Social Exclusion Research. D. Abrams, J. Christian and D. Gordon. Chichester, UK, John Wiley & Sons. Miller, S. (2003). Impact on Inference Quality. Handbook of Mixed Methods in Social and Behavioral Research. A. Tashakkori and C. Teddlie. Thousand Oaks, Sage Publications. Mingers, J. (2006). "A critique of statistical modelling in management science from a critical realist perspective: its role within multimethodology." Journal of the Operations Research Society 57: 202-219. Misra, D., B. Guyer, et al. (2003). "Integrated perinatal health framework. A multiple determinants model with a life span approach." American Journal of Preventive Medicine 25(1): 65-75. Miyata, H. and I. Kai (2009). "Reconsidering Evaluation Criteria for Scientific Adequacy in Health Care Research: An Integrative Framework of Quantitative and Qualitative Criteria." International Journal of Qualitative Methods 8(1): 64. Modarres, A. (2004). "Neighbourhood Integration: Temporality and Social Fracture." Journal of Urban Affairs 26(3): 351-377. Modell, S. (2009). "In defence of triangulation: A critical realist approach to mixed methods research in management accounting." Management Accounting Research 20: 208221. Moore, S., V. Haines, et al. (2006). "Lost in translation: a genealogy of the "social capital" concept in public health." Journal of Epidemiology & Community Health 60: 729734.

482

Bibliography

Morenoff, J. (2003). "Neighbourhood mechanisms and the spatial dynamics of birth weight." AJS 108(5): 976-1017. Morgensten (1982). "Uses of ecological analysis in epidemiological research." American Journal of Public Health 72: 1336-1344. Morgenstern, H. (1998). Ecological Studies. Modern Epidemiology. 2nd Edition. K. Rothman and S. Greenland. Philadephia, Lippincott-Raven: 459-490. Morgenstern, H. (2000). Ecological Study. Encyclopedia of Epidemiologic Methods. M. Gail and J. Benichou. Chichester, UK, John Wiley & Sons Ltd: 315-336. Morrell, C., R. Warner, et al. (2009). "Psychological interventions for postnatal depression: cluster randomised trial and economic evaluation. The PoNDER trial." Health Technology Assessment 13: 30. Munro, B. (2001). Statistical Methods for Health Care Research. Philadelphia, Lippincott. Muntaner, C. (1999). "Invited Commentary: Social Mechanisms, Race, and Social Epidemiology." American Journal of Epidemiology 150(2): 121-126. Muntaner, C. and M. Gomez (2003). "Qualitative and quantitative research in social epidemiology: is complementarity the only issue." Gaceta Sanitaria 17(Supl 3): 5357. Murray, L. and P. J. Cooper (1996). "The impact of postpartum depression on child development." International Review of Psychiatry 8: 55-63. Murray, L., A. Fiori-Cowley, et al. (1996). "The impact of postnatal depression and associated adversity on early mother-infant interaction and later infant outcome." Child Development 67: 2512-2526. Murray, L., A. Hipwell, et al. (1996). "The cognitive development of 5-year-old children of postnatally depressed mothers." Journal of Child Psychology & Psychiatry & Allied Disciplines 37(8): 927-935. Murray, L., C. Stanley, et al. (1996). "The role of infant factors in posnatal depression and mother-infant interactions." Developmental Medicine and Child Neurology 38(2): 109-119. Nager, A., L. M. Johansson, et al. (2006). "Neighbourhood socioeconomic environment and risk of postpartum psychosis." Archives of Women's Mental Health 9(2): 81-86. Nahas, V. and N. Amasheh (1999). "Culture care meanings and experiences of postpartum depression among Jordanian Australian women: A Transcultural study." Journal of Transcultural Nursing 10: 37-45. National Research Council and Institute of Medicine (2000). From Neurons to Neighbourhoods: The Science of Early Childhood Development. Committee on Integrating the Science of Early Childhood Development. Washington, DC, National Academy Press. Nauck, B. (2008). Acculturation. Multilevel Analysis of Individuals and Cultures. F. Van de Vijver, D. Van Hemert and Y. Poortinga. New York, Lawrence Erlbaum Associates.

483

Bibliography

Ng, S. (1991). "Does Epidemiology Need a New Philosophy?" American Journal of Epidemiology 133(11): 1073-1077. NHMRC (2000). Postnatal depression. A systematic review of published scientific literature to 1999. Canberra, National Health and Medical Research Council. Northcutt, N. and D. McCoy (2004). Interactive Qualitative Analysis. A Systems Method for Qualitative Research. Thousand Oaks, CA, Sage Publications. Norton, J. and X. Niu (2009). "Intrinsically Autoregressive Spatiotemporal Models With Application to Aggregated Birth Outcomes." Journal of the American Statistical Association 104(486): 638-649. Notkola, V., S. Punsar, et al. (1985). "Socio-economic conditions in childhood and mortality and morbidity caused by coronary heart disease in adulthood in rural Finland." Social Science & Medicine 21: 517-523. Nutbeam, D. and E. Harris, Eds. (1999). Theory in a nutshell: A guide to health promotion theory. Sydney, Australia, McGraw-Hill. O'Campo, P. (2003). "Invited Commentary: Advancing Theory and Methods for Multilevel Models of Residential Neighbourhoods and Health." American Journal of Epidemiology 157: 9-13. O'Campo, P., M. Kirst, et al. (2009). "Community-Based Services for Homeless Adults Experiencing Concurrent Mental Health and Substance Use Disorders: A Realist Approach to Synthesizing Evidence." Journal of Urban Health: Bulletin of the New York Academy of Medicine 86(6): 965-989. O'Campo, P., C. Salmon, et al. (2009). "Neighbourhoods and mental well-being: What are the pathways." Health & Place 15: 56-68. O'Connor, T., J. Heron, et al. (2002). "Antenatal anxiety predicts child behavioural/emotional problems independently of postnatal depression." Journal of the American Academy of Child & Adolescent Psychiatry 41: 1470-1477. O'Donnell, K., T. O'Connor, et al. (2009). "Prenatal Stress and Neurodevelopment of the Child: Focus on the HPA Axis and Role of the Placenta." Developmental Neuroscience 31(4): 285-292. O'Hara, M. and A. Swain (1996). "Rates and risk of postnatal depression - a metaanalysis." International Review of Psychiatry 8: 37-54. O'Hara, M. W. (1995). Postpartum Depression. New York, Springer-Verlag. O'Hara, M. W., D. J. Neunaber, et al. (1984). "Prospective study of postpartum depression: Prevalence, course, and predictive factors." Journal of Abnormal Psychology 93(2): 158-171. Oakes, J. (2004). "The (mis)estimation of neighborhood effects: causal inference for a practicable social epidemiology." Social Science & Medicine 58: 1929-1952. Oakes, J. and J. Kaufman, Eds. (2006). Methods in Social Epidemiology. San Francisco, John Wiley & Sons, Inc.

484

Bibliography

Olsen, W. (2001). "Stereotypical and traditional views about the gender division of labour in Indian labour markets." Journal of Critical Realism 4(1): 4-12. Olsen, W. and J. Morgan (2005). "A critical epistemology of analytical statistics: addressing the sceptical realist." Journal for the Theory of Social Behaviour 35(3): 255-284. Onozawa, K., R. Kumar, et al. (2003). "High EPDS scores in women from ethnic minorities living in London." Archives of Women's Mental Health 6(Suppl2): s51-s55. Onwuegbuzie, A. and R. Johnson (2006). "The Validity Issue in Mixed Research." Research in Schools 13(1): 48-63. Onwuegbuzie, A. and N. Leech (2006). "Linking research questions to mixed methods data analysis procedures." The Qualitative Report 11(3): 474-498. Orr, S. (2004). "Social support and pregnancy outcome: a review of the literature." Clinical Obstetrics and Gynecology 47(4): 842-855. Orr, S. and C. Miller (1995). "Maternal depressive symptoms and the risk of poor pregnancy outcome." Epidemiologic Reviews 17: 165-171. Osborne, J., A. Costello, et al. (2008). Best Practices in Exploratory Factor Analysis. Best Practices in Quantitative Methods. J. Osborne. Los Angeles, Sage Publications. Pawson, R. (2006). Evidence-Based Policy: A Realist Perspective. London, Sage Publications. Pawson, R. and N. Tiley (1997). Realistic Evaluation. London, Sage. Pearce, N. (1996). "Traditional epidemiology, modern epidemiology, and public health." American Journal of Public Health 86: 678-683. Pearce, N. and D. Crawford-Brown (1989). "Critical Discussion in Epidemiology: Problems with the Popperian approach." Journal of Clinical Epidemiology 42(3): 177-184. Pearl, J. (1995). "Causal diagrams for empirical research." Biometrika 155: 176-184. Pearlin, L. (1989). "The Sociological Study of Stress." Journal of Health and Social Behaviour 30(3): 241-256. Pearlin, L., E. Menaghan, et al. (1981). "The Stress Process." Journal of Health and Social Behaviour 22: 337-356. Peirce, C. (1960). Collected Papers of Charles Sanders Peirce. C. Hartshorne and P. Weiss. Cambridge, MAS, Belknap Ress of Harvard University Press. Perkin, M., J. Bland, et al. (1993). "The effect of anxiety and depression during pregnancy on obstetric complications." British Journal of Obstetrics & Gynaecology 100: 629634. Piantadosi, S., D. Byar, et al. (1988). "The ecological fallacy." American Journal of Epidemiology 127: 893-903.

485

Bibliography

Pickett, K. and M. Pearl (2001). "Multilevel analysis of neighbourhood socioeconomic context and health outcomes: a critical review." Journal of Epidemiology & Community Health 55: 111-122. Poole, C. and K. Rothman (1998). "Our conscientious objection to the epidemiology wars." Journal of Epidemiology & Community Health 52: 613-614. Popper, K. (1959). The logic of scientific discovery (in German). New York, Basic Books. Power, C. and C. Hertzman (1997). "Social and biological pathways linking early life and adult disease." British Medical Bulletin 53: 210-221. Preston, S., M. Hill, et al. (1998). "Childhood conditions that predict survival to advanced ages among African-Americans." Social Science & Medicine 47: 1231-1246. Pretty, G., H. Chipuer, et al. (2003). "Sense of place amongst adolescents and adults in two rural Australian towns: The discriminating features of place attachement, sense of community and place dependence in relation to place identity." Journal of Environmental Psychology 23(3): 273-287. Putnam, R. (1993). "The prosperous community: Social capital and public life." American Prospect 13: 35-42. Putnam, R. (1995). "Bowling alone: America's declining social capital." Journal of Democracy 6: 65-78. Putnam, R. (2000). Bowling alone: The collapse and revival of American community. New York, Simon and Schuster. Radenbush, S. and R. Sampson (1999). "Econometrics: toward a science of assessing ecological settings, with application to the systematic social observation of neighbourhoods." Sociolological Methodology 29: 1-41. Rajaratnam, J., J. Burke, et al. (2006). "Maternal and child health and neighborhood context: The selection and construction of area-level variables." Health & Place 12: 547-556. Raphael, D. (2006). "Social determinants of health: present status, unanswered questions and future directions." International Journal of Health Services 36(4): 651-677. Reardon, S. (2006). A conceptual framework for measuring segregation and its association with population outcomes. Methods in Social Epidemiology. J. Oakes and J. Kaufman. San Francisco, CA, Jossey-Bass. Regmi, S., W. Sligl, et al. (2002). "A controlled study of postpartum depression among Nepalese women: validation of the Edinburgh Postpartum Depression Scale in Kathmandu." Tropical Medicine & International Health 7(4): 378-382. Rice, F., G. Harold, et al. (2009). "Disentangling prenatal and inherited influences in humans with an experimental design." Proceedings of the National Academy of Sciences 106(7): 2464. Richman, J., V. Raskin, et al. (1991). "Gender roles, social support and postpartum depressive symptomatology." Journal of Nervous & Mental Disease 179: 139-147.

486

Bibliography

Ritter, C., S. E. Hobfoll, et al. (2000). "Stress, psychosocial resources, and depressive symptomatology during pregnancy in low-income, inner-city women." Health Psychology 19(6): 576-585. Ritzer, G. (2008). Modern Sociological Theory (7th Ed.). New York, McGraw-Hill. Robert, S. (1999). "Socioeconomic position and health: the independent contribution of community socioeconomic context." Annual Review of Sociology 25: 489-516. Roman, L., J. Gardiner, et al. (2009). "Alleviating perinatal depressive symptoms and stress: a nurse-community health worker randomized trial." Archives of Women's Mental Health 12(6): 379-391. Room, G. (1995). Beyond the Threshold: the Measurement and Analysis of Social Exclusion. Bristol, The Policy Press. Rose, G. (1985). "Sick Individuals and Sick Populations." International Journal of Epidemiology 14(1): 32-38. Rose, G. (1992). The strategy of preventive medicine. Oxford, England, Oxford University. Rosenberg, A. (2000). Philosophy of science: A contemporary introduction. London, Routledge. Ross, G. and C. Wu (1995). "The links between education and health." American Sociological Review 60: 719-745. Rothman, K. and S. Greenland (1998). Causation and Causal Inference. Modern Epidemiology. K. Rothman and S. Greenland. Philadelphia, Lippincott Williams & Wilkins. Rothman, K. and S. Greenland, Eds. (1998). Modern Epidemiology: Second Edition. Philadelphia, Lippincott Williams & Wilkins. Rothman, K. and S. Greenland (1998). Precision and Validity in Epidemiologic Studies. Modern Epidemiology, 2nd Ed. K. Rothman and S. Greenland. Philadelphia, PA, Lippincott Williams and Wilkins. Rubio, D., K. Kraemer, et al. (2008). "Factors associated with alcohol use, depression, and their co-occurrence during pregnancy." Alcoholism, clinical and experimental research 32(9): 1543-1551. Russo, F. (2009, February 2010). "Causal webs in epidemiology." Retrieved February 2010, from http://philsciarchive.pitt.edu/archive/00004970/01/Causal_webs_in_epidemiology.pdf. Rutter, M. (2006). "Is Sure Start an Effective Preventive Intervention?" Child and Adolescent Psychiatry and Mental Health 11(3): 135-141. Saldana, J. (2009). The Coding Manual for Qualitative Researchers. Los Angeles, Sage Publications.

487

Bibliography

Sandler, I., S. Braver, et al. (2000). Stress: Theory, research and action. Handbook of Community Psychology. J. Rappaport and E. Seidman. New York, Kluwer Academic/Plenum Publishers. Sarantakos, S. (1998). Varieties of social research. Social Research, 2nd Edition. S. Sarantakos. South Yarra, Australia, Macmillian. Saunders, P. (2003). Can Social Exclusion Provide a New framework for Measuring Poverty. SPRC Discussion Paper No. 127. Sydney, Social Policy Research Centre, University of New South Wales. Saunders, P. and L. Adelman (2005). Income Poverty, Deprivation and Exclusion: A Comparative Study of Australia and Britian, SPRC Discussion Paper No. 141. Sydney, Social Policy Research Centre, University of New South Wales. Sayer, A. (1992). Method in Social Science. London, Routledge. Sayer, A. (2000). Realism and Social Science. London, Sage Publications. Schumacker, R. (2006). "Conducting Specification Searches with AMOS." Structural Equation Modeling 13(1): 118-129. Schwartz, S. and A. Diez Roux (2001). "Commentary: Causes of incidence and causes of cases - a Durkheimian perspecitive on Rose." International Journal of Epidemiology 30: 435-439. Seckl, J. (2004). "Prenatal glucocorticoids and long-term programming." European Journal of Endocrinology 151: U49-U62. Seguin, L., L. Potvin, et al. (1999). "Socio-environmental factors and postnatal depressive symptomatology: A longitudinal study." Women & Health 29(1): 57-72. Seguin, L., L. Potvin, et al. (1999). "Depressive symptoms in the late postpartum among low socioeconomic status women." Birth 26: 157-163. Shoemaker, P., J. Tankard, et al. (2004). How to build social science theories. Thousand Oaks, Sage Publications. Shonkoff, J. P. and S. J. Meisels (2000). Handbook of Early Childhood Intervention. Cambridge, Cambridge University Press. Simpson, E. (1949). "Measurement of diversity." Nature 163: 688. Skapinakis, P., G. Lewis, et al. (2005). "Mental health inequalities in Wales, UK: multi-level investigation of the effects of area deprivation." British Journal of Psychiatry 186: 417-442. Small, R., J. Lumley, et al. (2003). "Cross-cultural Experiences of Maternal Depression: Associations and Contributing Factors for Vietnamese, Turkish and Filipino Immigrant Women in Victoria, Australia." Ethnicity & Health 8(3): 189-206. Smith, M. (2008). "Testable theory development for small-N studies: critical realism and middle-range theory." Retrieved 15 February, 2010, from http://aisel.aisnet.org/confirm2008/51.

488

Bibliography

Sohr-Preston, S. and L. Scaramella (2006). "Implications of Timing of Maternal Depressive Symptons for Early Cognitive and Language Development." Clinical Child and Family Psychology Review 9(1): 65-83. Spiegelhalter, D., N. Best, et al. (2002). "Bayesian Measures of Model Complexity and Fit (with Discussion)." Journal of the Royal Statistic Society B 64: 583-640. Spiegelhalter, D., A. Thomas, et al. (2003). WinBUGS Version 1.4 User Manual. Cambridge, UK, Medical Research Council Biostatistics Unit. Stack, C. (1974). All our kin: Strategies for survival in a black community. NewYork, Harper & Row. Stafford, M., M. De Silva, et al. (2008). "Neighbourhood social capital and common mental disorder: testing the link in a general population sample." Health & Place 14: 394405. Stamp, G. E. and C. A. Crowther (1994). "Postnatal depression: A South Australian prospective survey." Australian and New Zealand Journal of Obstetrics and Gynaecology 34: 1564-1167. Steer, R., T. Scholl, et al. (1992). "Self-reported depression and negative pregnancy outcomes." Journal of Clinical Epidemiology 45: 1093-1099. Stein, A., P. J. Cooper, et al. (1989). "Social adversity and perinatal complications: their relation to postnatal depression." British Medical Journal 298(6680): 1073-1074. Stewart, D., A. Gagnon, et al. (2008). "Postpartum depression symptoms in newcomers." Canadian Journal of Psychiatry 53(2): 121-124. Stone, E., Y. Lin, et al. (2008). "A final common pathway for depression? Progress toward a general conceptual framework." Neuroscience and biobehavioral reviews 32(3): 508-524. Stott, D. (1973). "Follow-up study from birth of the effects of prenatal stress." Developmental Medicine and Child Neurology 15: 770-787. Strauss, A. (1987). Qualitative analysis for social scientists. Cambridge, UK, Cambridge University Press. Strauss, C. and J. Corbin (1990). Basics of qualitative resaerch: Grounded theory procedures and techniques. Newbury Park, CA, Sage Publications. Stuchbery, M., S. Matthey, et al. (1998). "Postnatal depression and social supports in Vietnamese, Arabic and Anglo-Celtic mothers." Social Psychiatry and Psychiatric Epidemiology 33: 483-490. Subramanian, S. (2004). "The relevance of multilevel statistical methods for identifying causal neighbourhood effects." Social Science & Medicine 58: 1961-1967. Subramanian, S., K. Lochner, et al. (2003). "Neighbourhood differences in social capital: a compositional artifact or a contextual construct." Health & Place 9: 33-44.

489

Bibliography

Surkan, P., I. Kawachi, et al. (2008). "Childhood overweight and maternal depressive symptoms." Journal of Epidemiology and Community Health 62(5): e11. Surkan, P., K. Peterson, et al. (2006). "The role of social networks and support in postpartum women's depression: a multiethnic urban sample." Maternal & Child Health Journal 10(4): 375-383. Susser, E. (2004). "Eco-Epidemiology: Thinking Outside the Black Box." Epidemiology 15(5): 519-520. Susser, M. (1973). Causal thinking in the health sciences: concepts and strategies in epidemiology. New York, Oxford Press. Susser, M. (1991). "What is a Cause and How do we know one? A Grammar for Pragmatic Epidemiology." American Journal of Epidemiology 133(7): 635-648. Susser, M. (1994a). "The logic in ecological: I. The logic of analysis." American Journal of Public Health 84: 825-829. Susser, M. (1994b). "The logic in ecological. II: The logic of design." American Journal of Public Health 84: 830-835. Susser, M. (1998). "Does risk factor epidemiology put epidemiology at risk? Peering into the near future." Journal of Epidemiology & Community Health 52(10): 608-611. Susser, M. and E. Susser (1996). "Choosing a future for epidemiology: II. From black box to Chinese boxes and eco-epidemiology." American Journal of Public Health 86: 674-677. Swain, J., J. Lorberbaum, et al. (2007). "Brain basis of early parent-infant interactions: psychology, physiology, and in vivo functional neuroimaging studies." Journal of Child Psychology & Psychiatry & Allied Disciplines 48: 262-287. Sword, W., S. Watt, et al. (2006). "Postpartum health, service needs, and access to care experiences of immigrant and Canadian-born women." Journal of Obstetric, Gynecologic and Neonatal Nursing 35: 717-727. SWSAHS (2004). 2002-2003 Annual Report. Liverpool, South Western Sydney Area Health Service. Syme, S. and L. Berkman (1976). "Social class, susceptibility and sickness." American Journal of Epidemiology 104: 1-8. Tabachnick, B. and L. Fidell (2001). Using Multivariate Statistics. Boston, Allyn and Bacon. Talge, N., C. Neal, et al. (2007). "Antenatal maternal stress and long-term effects on child neurodevelopment: how and why?" Journal of Child Psychology & Psychiatry 48(3): 245-261. Tashakkori, A. and C. Teddlie, Eds. (2003). Handbook of Mixed Methods in Social and Behavioural Research. Thousand Oaks, Sage Publications.

490

Bibliography

Tashakkori, A. and C. Teddlie (2003). Major issues and controversies in the use of mixed methods in the social and behavioural sciences. Handbook of mixed methods in social and behavioural research. A. Tashakkori and Teddlie. Thousand Oakes, CA, Sage. Taylor, S. and R. Repetti (1997). "Health psychology: what is an unhealthy environment and how does it get under the skin?" Annual Review of Psychology 48: 411-447. Teddlie, C. and A. Tashakkori (2009). Foundations of Mixed Methods Research. Los Angeles, Sage. Thagard, P. (1978). "The Best Explanation: Criteria for Theory Choice." Journal of Philosophy 75(2): 76-92. Thagard, P. (1988). Computational philosophy of science. Cambridge, MA, MIT Press. Thagard, P. (1992). Conceptual revolutions. Princeton, NJ, Princeton University Press. Thagard, P. (2000). Coherence in thought and action. Cambridge, MA, MIT Press. Thagard, P. (2000). "Probabilistic networks and explanatory coherence." Cognitive Science Quarterly 1(1): 91–114. Thomas, A. and S. Chess (1977). Temperament and development. New York, Brunner Mazel. Thygesen, L., G. Andersen, et al. (2005). "A philosophical analysis of the Hill criteria." British Medical Journal 59(6): 512. Tobin, G. and C. Begley (2004). "Methodological rigour within a qualitative framework." Journal of Advanced Nursing 48(4): 388-396. Tolan, P., D. Gorman-Smith, et al. (2003). "The Developmental Ecology of Urban Males' Youth Violence." Developmental Psychology 39(2): 274-291. Torraco, R. (2002). "Research Methods for Theory Building in Applied Disciplines: A Comparative Analysis." Advances in Developing Human Resources 4(3): 355-376. Tsakloglou, P. and F. Papadopoulos (2002). "Aggregate level and determining factors of social exclusion in twelve European countries." Journal of European Social Policy 12(3): 211-226. Tu, Y. (2009). "Commentary: Is structural equation modelling a step forward for epidemiologists." International Journal of Epidemiology 38: 549-551. Tukey, J. (1980). "We Need Both Exploratory and Confirmatory." The American Statistician 34(1): 23-25. Turner, R. (2010). Understanding Health Disparities: The Promise of the Stress Process Model. Advances in the Conceptualization of the Stress Process: Essays in Honor of Leonard I Pearlin. W. Avison, C. S. Aneshensel, S. Schieman and B. Wheaton. New York, Springer.

491

Bibliography

Turner, R. and D. Lloyd (1999). "The Stress Process and the Social Distribution of Depression." Journal of Health and Social Behaviour 40: 374-404. Uchino, B., J. Cacioppo, et al. (1996). "The relationship between social support and physiological processes: A review with emphasis on underlying mechanisms and implications for health." Psychological Bulletin 119: 488-531. Van de Vijver, F., D. Van Hemert, et al. (2008). Conceptual Issues in Multilevel Models. Multilevel Analysis of Individuals and Cultures. F. Van de Vijver, D. Van Hemert and Y. Poortinga. New York, Lawrence Erlbaum Associates. Wandersman, A. and P. Florin (2000). Citizen Participation and Community Organizations. Handbook of Community Psychology. J. Rappaport and E. Seidman. New York, Kluwer Academic/Plenum Publishers. Wandersman, A. and M. Nation (1998). "Urban Neighbourhoods and mental health: Psychological contributions to understanding toxicity, resilience, and interventions." American Psychologist 53(6): 647-656. Ward, A. (2009). "The role of causal criteria in causal inferences: Bradford Hill's" aspects of association"." Epidemiologic Perspectives & Innovations 6(1): 2. Warner, R., L. Appleby, et al. (1996). "Demographic and obstetric risk factors for postnatal psychiatric morbidity." British Journal of Psychiatry 168: 607-611. Watt, R. (2002). "Emerging theories into the social determinants of health: implications for oral health promotion." Community Dentistry and Oral Epidemiology 30: 241-247. Weich, S., M. Blanchard, et al. (2002). "Mental health and the built environment: crosssectional survey of individual and contextual risk factors for depression." British Journal of Psychiatry 180: 428-433. Weich, S., L. Twigg, et al. (2003). "Contextual risk factors for the common mental disorders in Britain: a multilevel investigation of the effects of place." Journal of Epidemiology & Community Health 57: 616-621. Weinstock, M. (2008). "The long-term behavioural consequences of prenatal stress." Neuroscience and biobehavioral reviews 32(6): 1073-1086. Weissman, M. and M. Olfson (1995). "Depression in women: Implications for health care research." Science 269: 799-801. Welberg, L. and J. Secki (2001). "Prenatal Stress, Glucocorticoids and the Programming of the Brain." Journal of Neuroendocrinology 13: 113-128. Whittlemore, R. (2005). "Combining evidence in nursing research: methods and implications." Nursing Research 54(1): 56-62. Wickberg, B. and C. P. Hwang (1997). "Screening for postnatal depression in a population-based Swedish sample." Acta Psychiatrica Scandinavica 95(1): 62-66. Williams, H. and A. Carmichael (1985). "Depression in mothers in a multi-ethnic urban industrial municipality in Melbourne. Aetiological factors and effects on infants and preschool children." Journal of Child Psychology and Psychiatry 27: 277-288.

492

Bibliography

Wilson, L. M., A. J. Reid, et al. (1996). "Antenatal psychosocial risk factors associated with adverse postnatal family outcomes." Canadian Medical Association Journal 154: 785-799. Wong, I., B. Cowling, et al. (2009). "A multilevel analysis of the effects of neighbourhood income inequality on individual self-rated health in Hong Kong." Social Science & Medicine 68: 124-132. Woolley, C. (2009). "Meeting the Mixed Methods Challenge of Integration in a Sociological Study of Structure and Agency." Journal of Mixed Methods Research 3(1): 7-25. Wright, R. (2007). "Prenatal maternal stress and early caregiving experiences: implications for childhood asthma risk." Paediatric and Perinatal Epidemiology 21(s3 Insights from the Epidemiology of Asthma): 8-14. Yeung, H. (1997). "Critical realism and realist research in human geography: a method or a philosophy in search of a method." Progress in Human Geography 21(1): 51-74. Yin, R. (2006). "Mixed Methods Research: Are the Methods Genuinely Integrated or Merely Parallel?" Research in the Schools 13(1): 41-47. Yoshida, K., H. Yamashita, et al. (2001). "Postnatal depression in Japanese mothers and the reconsideration of 'Satogaeri bunben'." Pediatrics International. 43(2): 189-193. Young, J. (2006). "Developmental origins of obesity: a sympathoadrenal perspective." International Journal of Obesity 30: S41-S49. Yu, C. (2005). Misconceived relationships between logical positivism and quantitative research, Revised 18 July 2005. Annual Meeting of the 2001 American Educational Research Association Seattle, WA, Research Method Forum. Zajicek, E. (1981). Psychiatric problems in pregnancy. Pregnancy: A psychological and social map. S. Wolkind and E. Zajicek. London, Academic Press. Zelkowitz, P. and T. H. Milet (1995). "Screening for postpartum depression in a community sample." Canadian Medical Association Journal 40: 80-85. Zhang, R., Q. Chen, et al. (1999). "[Study for the factors related to postpartum depression]." Chung-Hua Fu Chan Ko Tsa Chih [Chinese Journal of Obstetrics & Gynecology]. 34(4): 231-233. Zhu, L., D. Gorman, et al. (2006). "Hierarchical Bayesian spatial models for alcohol availablity, drug "hot spots" and violent crime." International Journal of Health Geographics 5: 54. Ziersch, A. and F. Baum (2004). "Involvement in civil society groups: Is it good for your health?" Journal of Epidemiology & Community Health 58: 493-500. Zimmerman, M. (2000). Empowerment Theory: Psychological, Organisational and Community Levels of Analysis. Handbook of Community Psychology. J. Rappaport and E. Seidman. New York, Kluwer Academic/Plenum Publishers.

493

Bibliography

494

List of Appendices

List of Appendices

A.


B


C


D


E

Individual Level Imputed Data Analysis

F

Group Level Variables – Technical Notes

G

Group Level Correlation and Bivariate Analysis

H

Group Level Likelihood Linear Regression

I

Group Level Bayesian CAR Bivariate Analysis

J

Final Spatial Regression Model EDS > 9

K

Ecological Exploratory Factor Analysis

L

Mixed Method Evaluation

495

List of Appendices

496

Appendix A: Ethics Approval


497


498


499


500

Appendix B: Consent


501

Appendix B: Consent

502

Appendix B: Consent

503

Appendix B: Consent

504

Appendix C: EPDS


505

Appendix C: EPDS

506

Appendix D: IBIS


507

Appendix D: IBIS

508

Appendix E: Imputed Analysis

Individual Level Imputed Data Analysis Univariate and Bivariate Analysis 3,000

Frequency

2,000

Mean =5.36 Std. Dev. =4.04 N =21,758

1,000

0 -10.00

0.00

10.00

20.00

30.00

EDS_Missing

Figure 148: Distribution of the imputed EDS Variable Descriptive Statistics for Imputed EDS Variable

EDS_Missing N

Statistic

21758

Range

Statistic

29.00

Minimum

Statistic

.00

Maximum

Statistic

29.00

Mean

Statistic

5.3646

Std. Error

.02739

Std. Deviation

Statistic

4.03967

Variance

Statistic

16.319

Skewness

Statistic

1.103

Std. Error Kurtosis

Statistic Std. Error

Valid N (listwise) 21758

.017 1.692 .033

Figure 149: Descriptive Statistics for the imputed EDS Variable

509


Table 66: Bivariate Logistic Regression, EDS > 9 Imputed, Significant Associations OR Variable

B

SE

Wald

Df

Sig


0.573

0.042

185.494

1

Sole Parent (D)

0.324

0.065

24.992

Household Size (C)

Categories

1–5

95% CI

Exp(B)

Lower

Upper

0.000

1.773

1.633

1.926

1

0.000

1.383

1.218

1.570

7.689

2

0.021

6 – 10

0.025

0.056

0.206

1

0.650

0.975

0.875

1.087

> 10

0.631

0.232

7.375

1

0.007

1.880

1.192

2.964

0.092

0.059

2.420

1

0.120

1.096

0.976

1.231

Accommodation (D)

0.086

0.080

1.159

1

0.282

1.090

0.932

1.274


0.276

0.077

12.801

1

0.000

1.318

1.133

1.532


0.171

0.013

168.346

1

0.000

1.186

1.156

1.217

143.517

2

0.000

Blended Family (D)

Socioeconomic


Car Access (C)

Good Average

0.462

0.072

41.512

1

0.000

1.587

1.379

1.826

Poor

1.073

0.091

138.109

1

0.000

2.925

2.446

3.499

75.909

2

0.000

Regular Occasional

0.271

0.054

25.316

1

0.000

1.311

1.180

1.458

No

0.458

0.057

64.173

1

0.000

1.581

1.413

1.768

1.097

0.072

230.166

1

0.000

2.996

2.600

3.452

0.547

0.027

402.305

1

0.000

1.728

1.638

1.822

0.393

0.027

206.362

1

0.000

1.482

1.404

1.563

Planned Pregnancy (D)

0.254

0.042

35.691

1

0.000

1.289

1.186

1.401

Previous Stillbirth (D)

0.193

0.074

6.779

1

0.009

1.212

1.049

1.401

0.120

0.023

27.060

1

0.000

1.127

1.077

1.179

0.171

0.043

15.885

1

0.000

1.186

30.440

2

0.000

Health Mother’s Health (D) Mother’s Health (S) Child’s Health (S)

Health Mother(D) Maternal Health(O) Health Child(O)

Pregnancies

Social Network Suburb Duration (S)

IODURAT

Suburb Duration (D) Suburb Duration (C)

Regret Leaving Suburb

Regret Leaving Suburb (S)

510

3+

1.091

1.291

2 yrs

0.217

0.056

15.207

1

0.000

1.243

1.114

1.386

1 yr or less

0.236

0.046

25.964

1

0.000

1.266

1.156

1.386

67.437

3

0.000


0.234

0.049

22.482

1

0.000

1.263

1.147

1.391

No not much

0.426

0.063

45.249

1

0.000

1.531

1.352

1.733

No not at all

0.492

0.077

40.700

1

0.000

1.636

1.406

1.902

0.180

0.022

65.754

1

0.000

1.198

1.146

1.251


OR Variable

B

SE

Wald

Df

Sig

Regret Leaving Suburb (D)

0.321

0.048

45.649

1

Support Network (S)

0.333

0.018

325.399

Support Network (D)

0.733

0.047


Categories

Yes

95% CI

Exp(B)

Lower

Upper

0.000

1.378

1.256

1.513

1

0.000

1.396

1.346

1.447

244.204

1

0.000

2.082

1.899

2.283

137.424

2

0.000

Sometimes

0.472

0.058

67.052

1

0.000

1.604

1.432

1.796

No

0.764

0.082

86.141

1

0.000

2.147

1.827

2.523


0.410

0.035

136.629

1

0.000

1.507

1.407

1.614


0.558

0.050

126.126

1

0.000

1.747

1.585

1.925

211.620

2

0.000

Emotional Support

Yes Sometimes

0.627

0.069

82.888

1

0.000

1.873

1.636

2.144

No

1.087

0.090

146.038

1

0.000

2.964

2.485

3.535

0.567

0.039

211.157

1

0.000

1.762

1.633

1.902

0.782

0.057

190.174

1

0.000

2.185

1.955

2.442

Comforts Baby

0.974

0.429

5.148

1

0.023

2.648

1.142

6.143


1.235

0.383

10.377

1

0.001

3.437

1.622

7.285

Baby Trouble Sleeping

0.322

0.024

175.206

1

0.000

1.380

1.315

1.447

Baby Demanding

0.283

0.022

164.195

1

0.000

1.328

1.271

1.386

Baby Content

0.323

0.031

105.480

1

0.000

1.381

1.299

1.469

Baby Difficult Feeder

0.171

0.027

39.544

1

0.000

1.186

1.125

1.251

Baby Difficult to Comfort

0.270

0.027

98.381

1

0.000

1.310

1.242

1.382

Health of Child

0.393

0.027

206.362

1

0.000

1.482

1.404

1.563

Emotional Support (S) Emotional Support (D)

I

Attachment



511


Table 67: Bivariate Logistic Regression, EDS > 12 Imputed, Significant Associations OR

95% CI

B

SE

Wald

df

Sig

Exp(B)

Lower

Upper


0.544

0.064

71.942

1

0.000

1.722

1.519

1.953

Sole Parent (D)

0.711

0.086

69.091

1

0.000

2.036

1.722

2.408

11.205

2

0.004

Variable

Category

Demographic

Household Size (C)

1–5 6 – 10

0.040

0.083

0.234

1

0.629

1.041

0.885

1.225

> 10

0.968

0.290

11.102

1

0.001

2.632

1.490

4.651

0.307

0.084

13.477

1

0.000

1.359

1.154

1.600

46.022

6

0.000

Blended Family (D)

Socioeconomic Accommodation (C)

Mortgage Owned

0.027

0.130

0.044

1

0.834

1.028

0.796

1.327

Private rent

0.198

0.076

6.751

1

0.009

1.219

1.050

1.416

Public Rent

0.401

0.120

11.093

1

0.001

1.493

1.179

1.890

Parents

0.385

0.095

16.222

1

0.000

1.469

1.218

1.771

Caravan

0.464

0.739

0.394

1

0.530

1.590

0.374

6.764

Refuge

1.841

0.370

24.783

1

0.000

6.305

3.054

13.017

Accommodation (D)

0.366

0.109

11.282

1

0.001

1.441

1.164

1.784


0.271

0.116

5.414

1

0.020

1.311

1.044

1.647


0.245

0.020

151.448

1

0.000

1.277

1.229

1.328

163.712

2

0.000


Car Access (C)

Good Average

0.513

0.118

19.063

1

0.000

1.671

1.327

2.104

Poor

1.487

0.135

120.573

1

0.000

4.425

3.394

5.771

67.792

2

0.000

Regular Occasional

0.357

0.081

19.600

1

0.000

1.428

1.220

1.673

Poor

0.636

0.082

60.970

1

0.000

1.890

1.611

2.217

1.510

0.088

291.633

1

0.000

4.527

3.807

5.384

0.704

0.042

286.121

1

0.000

2.021

1.863

2.193

0.385

0.042

85.619

1

0.000

1.470

1.355

1.595

0.209

0.063

10.893

1

0.001

1.233

1.089

1.396

Planned Pregnancy (D)

0.469

0.063

55.256

1

0.000

1.599

1.413

1.809

Previous Stillbirth (D)

0.228

0.109

4.368

1

0.037

1.256

1.014

1.556

Health Mother’s Health (D) Mother’s Health (S) Child’s Health (S) Breast Feeding

Health Mother(D) Maternal Health(O) Health Child(O) IDBF

Pregnancies

512


OR

95% CI

B

SE

Wald

df

Sig

Exp(B)

Lower

Upper

Suburb Duration (S)

0.197

0.035

32.104

1

0.000

1.218

1.138

1.304

Suburb Duration (D)

0.307

0.064

23.205

1

0.000

1.360

1.200

1.541

32.779

2

0.000

Variable

Category

Social Network

Suburb Duration (C)


3+ 2 yrs

0.280

0.086

10.710

1

0.001

1.324

1.119

1.565

1 yr or more

0.395

0.070

31.699

1

0.000

1.484

1.293

1.702

70.442

3

0.000


0.161

0.077

4.394

1

0.036

1.175

1.011

1.365

No not much

0.419

0.096

19.140

1

0.000

1.521

1.260

1.835

No not at all

0.826

0.105

62.463

1

0.000

2.284

1.861

2.803

0.261

0.033

63.675

1

0.000

1.298

1.218

1.384

0.492

0.069

51.009

1

0.000

1.635

1.429

1.872

Support Network (S)

0.401

0.026

239.715

1

0.000

1.493

1.419

1.571

Support Network (D)

0.919

0.067

188.989

1

0.000

2.507

2.199

2.858

127.666

2

0.000

Regret Leaving Suburb (S) Regret Leaving Suburb (D)


Yes Sometimes

0.526

0.085

38.415

1

0.000

1.693

1.433

2.000

No

1.090

0.106

105.387

1

0.000

2.974

2.415

3.662


0.540

0.048

127.181

1

0.000

1.717

1.563

1.886


0.708

0.071

100.185

1

0.000

2.031

1.768

2.333

209.954

2

0.000

Emotional Support (C)

Yes Sometimes

0.856

0.094

82.553

1

0.000

2.354

1.957

2.831

No

1.388

0.113

151.982

1

0.000

4.007

3.213

4.996

Emotional Support (S)

0.730

0.050

210.877

1

0.000

2.075

1.880

2.290

Emotional Support (D)

1.045

0.076

187.730

1

0.000

2.844

2.449

3.303

Comforts Baby

1.111

0.546

4.135

1

0.042

3.038

1.041

8.866


1.811

0.417

18.895

1

0.000

6.118

2.703

13.844

Baby Trouble Sleeping (S)

0.406

0.035

133.746

1

0.000

1.501

1.402

1.609

Baby Demanding (S)

0.392

0.031

157.531

1

0.000

1.479

1.392

1.572

Baby Content (S)

0.390

0.045

74.686

1

0.000

1.477

1.352

1.614

Baby Difficult Feeder (S) Baby Difficult to Comfort (S)

0.208

0.039

28.207

1

0.000

1.231

1.140

1.330

0.384

0.038

101.614

1

0.000

1.468

1.363

1.582

0.385

0.042

85.619

1

0.000

1.470

1.355

1.595

Attachment


Health of Child (S)


513


Logistic Regression EDS >9 (Missing Values Imputed) Exp(B) B

S.E.

Wald

df

Sig.

Upper

95% C.I for EXP(B) Lower

Upper

Country of Birth(D)(1)

.378

.058

42.037

1

.000

1.460

1.302

1.637

Sole Parent(D)(1)

.116

.125

.866

1

.352

1.123

.879

1.434

5.513

2

.063

Household Size(O) Household Size(O)(1)

-.127

.075

2.835

1

.092

.881

.760

1.021


.530

.342

2.402

1

.121

1.698

.869

3.319

Blended Family(D)(1)

.024

.080

.086

1

.769

1.024

.875

1.198

Accommodation(D)(1)

-.343

.127

7.291

1

.007

.710

.553

.910

Unemployed Father(D)(1)

-.063

.106

.351

1

.553

.939

.763

1.156

.080

.017

20.667

1

.000

1.083

1.046

1.121

6.379

2

.041

Financial Situation(O) Car Access(O) Car Access(O)(1)

-.171

.077

4.845

1

.028

.843

.724

.981

Car Access(O)(2)

.055

.084

.426

1

.514

1.056

.896

1.245

Health Mother(D)(1)

.763

.098

60.743

1

.000

2.144

1.770

2.597

Health Child(O)

.150

.036

17.186

1

.000

1.162

1.082

1.247

Planned Pregnancy(D)(1)

.152

.059

6.715

1

.010

1.164

1.038

1.307

Suburb Duration(D)(1)

.046

.058

.631

1

.427

1.047

.935

1.172

No Regret Leaving(D)(1)

.166

.063

6.805

1

.009

1.180

1.042

1.336

Social Support(D)(1)

.278

.070

15.995

1

.000

1.321

1.153

1.514

Practical Support(D)(1)

.247

.074

11.027

1

.001

1.280

1.106

1.481

Emotional Support(D)(1)

.356

.082

18.590

1

.000

1.427

1.214

1.677

-.425

1.216

.122

1

.726

.654

.060

7.085

-1.862

1.319

1.991

1

.158

.155

.012

2.063

.195

.042

21.124

1

.000

1.215

1.118

1.321

Comforts Child(D)(1) Avoid Contact Child(D)(1) Baby Sleep Trouble(O) Baby Demanding(O)

.093

.039

5.747

1

.017

1.098

1.017

1.185

-.014

.040

.122

1

.727

.986

.912

1.067


.036

.046

.608

1

.436

1.037

.947

1.134

Baby not content(O)

.134

.048

7.749

1

.005

1.143

1.040

1.256

-1.905

1.324

2.071

1

.150

.149


Constant

Table 68: Logistic Regression EDS >9 Imputed Data Table 68: shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >9 entered into the initial logistic regression model. Employing a 0.05 criterion of statistical significance, country of birth, accommodation, financial situation, maternal health, unplanned pregnancy, social support, practical support, emotional support, baby with difficulty sleeping, demanding, a difficult feeder, not being content, and one dummy variable coding access to a car, had significant partial effects. The correlation matrix revealed no signs of multicollinearity.

514


The following predictors were not significant: sole parenthood, household size, blended family, suburb duration, no regret leaving the suburb, The Hosmer and Lemeshow Test was a not significant (Chi-square = 6.944, df 8, p = 0.543) indicating that the data fit the model well. The model was able to correctly classify 99.5 percent of EDS >9 for an overall success rate of 85.0 percent. The area under the ROC curve was 67.1 (95% CI: 65.7 – 68.5) indicating a less than adequate fit.

95.0% C.I for EXP(B) B

S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

Country of Birth(D)

.326

.056

33.419

1

.000

1.386

1.241

1.548

Accommodation(D)

-.363

.123

8.745

1

.003

.695

.546

.885

.066

.017

15.121

1

.000

1.068

1.033

1.104

7.080

2

.029

Financial Situation(O) Car Access(O) Car Access(O)(1)

-.162

.075

4.677

1

.031

.850

.734

.985

Car Access(O)(2)

.079

.081

.961

1

.327

1.082

.924

1.267

Maternal Health(O)

.359

.035

102.576

1

.000

1.432

1.336

1.535

Planned Pregnancy(D)

.119

.056

4.515

1

.034

1.126

1.009

1.257


.142

.062

5.291

1

.021

1.153

1.021

1.301

Social Support(O)

.148

.027

29.640

1

.000

1.160

1.099

1.223


.215

.073

8.642

1

.003

1.239

1.074

1.430


.326

.081

16.143

1

.000

1.385

1.182

1.624

Baby Sleep Trouble(O)

.208

.035

35.239

1

.000

1.231

1.149

1.318

Baby Demanding(D)

.507

.111

21.008

1

.000

1.661

1.337

2.063

Baby not content(O)

.154

.044

11.980

1

.001

1.166

1.069

1.272

-4.218

.147

819.257

1

.000

.015

Constant

Table 69: Forward Stepwise (Conditional) Logistic Regression of EDS >9 (Imputed) The Hosmer and Lemeshow Test was a not significant (Chi-square = 4.531., df 8, p = 0.8.6) indicating that the data fit the model well. The area under the ROC curve was 67.2 (95% CI: 65.9 – 68.4) indicating a less than adequate fit.

515



S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

Country of Birth(D)

.361

.050

51.811

1

.000

1.434

1.300

1.583


.068

.015

21.226

1

.000

1.071

1.040

1.102

Maternal Health(O)

.396

.032

152.152

1

.000

1.486

1.395

1.583

Social Support(O)

.160

.024

46.191

1

.000

1.174

1.121

1.229


.329

.068

23.393

1

.000

1.390

1.216

1.588

Baby Demanding(D)

.469

.103

20.775

1

.000

1.598

1.306

1.954


.192

.032

36.477

1

.000

1.211

1.138

1.289

Baby not content(O)

.120

.040

8.810

1

.003

1.127

1.042

1.220

-4.153

.133

977.663

1

.000

.016

Constant

Table 70: Parsimonious Preliminary Main Effect Model EDS >9 (Imputed Dataset) The Hosmer and Lemeshow Test was a not significant (Chi-square = 4.369., df 8, p = 0.822) indicating that the data fit the model well. The model was able to correctly classify 99.6 percent of EDS >9 for an overall success rate of 84.0 percent. The area under the ROC curve was 67.1 (95% CI: 65.9 – 68.3) indicating a less than adequate fit.


S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

Baby Demanding(D)

.469

.103

20.716

1

.000

1.598

1.306

1.956

Country of Birth(D)

.365

.050

52.641

1

.000

1.440

1.305

1.589


.064

.015

18.080

1

.000

1.066

1.035

1.097

Maternal Health(O)

.392

.032

147.769

1

.000

1.479

1.389

1.576

Social Support(O)

.151

.024

39.275

1

.000

1.163

1.109

1.219


.171

.065

6.935

1

.008

1.187

1.045

1.348


.282

.072

15.429

1

.000

1.326

1.152

1.526


.188

.032

34.896

1

.000

1.207

1.134

1.285

Baby not content(O)

.122

.041

8.991

1

.003

1.129

1.043

1.223

-4.121

.134

952.044

1

.000

.016

Constant

Table 71: Final Model EDS >9 (Imputed Dataset) The Hosmer and Lemeshow Test was a not significant (Chi-square = 4.369., df 8, p = 0.822) indicating that the data fit the model well. The model was able to correctly classify 99.6 percent of EDS >9 for an overall success rate of 84.0 percent. The area under the ROC curve was 67.2 (95% CI: 66.0 – 68.4) indicating a less than adequate fit.

516


chgdev9M

6.00

4.00

2.00

0.00 0.00000

0.20000

0.40000

0.60000

0.80000


Figure 150: Deviance versus Predicted Probability EDS >9 Imputed The curve that extends from the lower left to the upper right corresponds to cases in which the dependent variable has a value of 0. The change in deviance plot illustrates that there are no significantly outlying cases.


0.01500

0.01000

0.00500

0.00000 0.00000

0.20000

0.40000

0.60000

0.80000


Figure 151: Cooks Distances versus Predicted Probabilities EDS >9 Imputed The Cook’s distances plot generally follows that of the change in deviance. There are cases in both the upper left and upper right that are outliers. The EDS > 9 Imputed model does not fit as well as the EDS > 9.

517


Logistic Regression EDS > 12 (Missing Values Imputed) 95% C.I for EXP(B) B

S.E.

Wald

df

Sig.

Exp(B)

Lower

Upper

Country of Birth(D)

.211

.089

5.565

1

.018

1.234

1.036

1.470

Sole Parent(D)

.295

.166

3.158

1

.076

1.343

.970

1.860

6.346

2

.042

Household Size(O) Household Size(O)(1)

-.073

.112

.425

1

.514

.930

.747

1.157


.955

.400

5.695

1

.017

2.599

1.186

5.696

Blended Family(D)

.083

.117

.506

1

.477

1.087

.864

1.366

Accommodation(D)

-.514

.191

7.217

1

.007

.598

.411

.870


-.146

.153

.908

1

.341

.864

.640

1.167


.115

.026

18.961

1

.000

1.122

1.065

1.181

2.365

2

.307

Car Access(O) Car Access(O)(1)

-.091

.113

.648

1

.421

.913

.732

1.139

Car Access(O)(2)

.123

.118

1.075

1

.300

1.131

.896

1.426

Maternal Health(O)

.506

.059

73.274

1

.000

1.659

1.477

1.862

-.115

.059

3.811

1

.051

.891

.793

1.000

Planned Pregnancy(D)

.143

.088

2.682

1

.101

1.154

.972

1.370

Suburb Duration(D)

.130

.086

2.321

1

.128

1.139

.963

1.347


.221

.092

5.754

1

.016

1.248

1.041

1.495

Social Support(O)

.163

.039

17.048

1

.000

1.177

1.089

1.272


.277

.105

6.945

1

.008

1.319

1.074

1.621


.491

.112

19.129

1

.000

1.634

1.311

2.036


.240

.058

17.265

1

.000

1.271

1.135

1.423

Baby Demanding(D)

.646

.148

19.138

1

.000

1.908

1.428

2.548


.005

.057

.008

1

.927

1.005

.898

1.125


.035

.065

.292

1

.589

1.036

.912

1.176

Baby not content(O)

.149

.068

4.771

1

.029

1.161

1.015

1.327

-5.996

.237

638.644

1

.000

.002

Health Child(O)

Constant

Table 72: Logistic Regression EDS > 12 (Missing Values Imputed) Table 72: shows the coefficient, Wald test and odds ratio for each of the predictors of EDS >12 entered into the initial logistic regression model. Employing a 0.05 criterion of statistical significance, country of birth, accommodation, financial situation, maternal health, unplanned pregnancy, social support, practical support, emotional support, baby with difficulty sleeping, demanding, a difficult feeder, not being content, and one dummy variable coding access to a car, had significant partial effects. The correlation matrix revealed no signs of multicollinearity.

518


The following predictors were not significant: sole parenthood, household size, blended family, suburb duration, no regret leaving the suburb, The Hosmer and Lemeshow Test was a not significant (Chi-square = 8.358, df 8, p = 0.399) indicating that the data poorly fit the model. The model was able to correctly classify 99.9 percent of EDS >9 for an overall success rate of 94.2 percent. The area under the ROC curve was 71.9 (95% CI: 69.9 – 73.9) indicating an adequate fit.

95% C.I for EXP(B) B Sole Parent(D)

.388

S.E. .160

Household Size(O)

Wald

df

Sig.

5.865

1

.015

6.270

2

.044

Exp(B)

Lower

Upper

1.474

1.077

2.017


-.046

.109

.178

1

.673

.955

.771

1.183


.976

.400

5.958

1

.015

2.653

1.212

5.808

-.485

.188

6.628

1

.010

.616

.425

.891


.120

.026

21.546

1

.000

1.127

1.072

1.185

Maternal Health(O)

.510

.059

74.311

1

.000

1.665

1.483

1.870

Accommodation(D)

Health Child(O)

-.125

.059

4.512

1

.034

.882

.786

.990


.258

.091

8.034

1

.005

1.294

1.083

1.546

Social Support(O)

.170

.039

18.907

1

.000

1.186

1.098

1.280


.280

.105

7.155

1

.007

1.323

1.078

1.624


.510

.111

20.925

1

.000

1.665

1.338

2.071


.253

.052

23.413

1

.000

1.288

1.163

1.427

Baby Demanding(D)

.658

.147

20.093

1

.000

1.930

1.448

2.573

Baby not content(O)

.156

.066

5.521

1

.019

1.169

1.026

1.332

-5.923

.232

650.204

1

.000

.003

Constant

Table 73: Forward Stepwise (Conditional) Logistic Regression of EDS >12 The Hosmer and Lemeshow Test was a not significant (Chi-square = 9.263., df 8, p = 0.321) indicating that the data fit the model poorly. The model was able to correctly classify 99.9 percent of EDS >9 for an overall success rate of 94.2 percent. The area under the ROC curve was 71.9 (95% CI: 69.8 – 73.9) indicating an adequate fit.

519


*

95% C.I for EXP(B) Wald

df

Sig.

Baby Demanding(D)

.655

B

.133

24.140

1

.000

1.925

1.483

2.500

Country of Birth(D)

.178

.076

5.418

1

.020

1.195

1.028

1.388


.129

.022

33.014

1

.000

1.137

1.089

1.189

Maternal Health(O)

.477

.049

94.817

1

.000

1.611

1.464

1.774

Social Support(O)

.208

.033

39.012

1

.000

1.231

1.153

1.314


.514

.092

31.328

1

.000

1.672

1.397

2.003


.249

.042

34.397

1

.000

1.283

1.180

1.394

-5.719

.186

947.055

1

.000

.003

Constant

S.E.

Exp(B)

Lower

Upper

Table 74: Parsimonious Preliminary Main Effect Model EDS >12 (Missing Values Imputed) The Hosmer and Lemeshow Test was a not significant (Chi-square = 4.965., df 8, p = 0.761) indicating that the data fit the model well. The model was able to correctly classify 100 percent of EDS >9 for an overall success rate of 93.8 percent. The area under the ROC curve was 70.7 (95% CI: 68.7 – 72.8) indicating an adequate fit.


Wald

df

Sig.

.137

20.960

1

.000

1.869

1.430

2.443

.197

.077

6.539

1

.011

1.218

1.047

1.416

.122

.023

28.887

1

.000

1.130

1.081

1.181

.468

.049

89.735

1

.000

1.596

1.449

1.759

.189

.034

30.501

1

.000

1.208

1.130

1.292

.223

.092

5.823

1

.016

1.249

1.043

1.497


.449

.098

20.939

1

.000

1.567

1.293

1.899


.205

.047

18.739

1

.000

1.228

1.119

1.347

Baby not content(O)

.106

.060

3.110

1

.078

1.112

.988

1.251

-5.805

.206

794.633

1

.000

.003

Baby Demanding(D)

.626

Country of Birth(D) Financial Situation(O) Maternal Health(O) Social Support(O) Practical Support(D)

Constant

S.E.

Exp(B)

Lower

Upper

Table 75: Final Model EDS >12 (Missing Values Imputed) The Hosmer and Lemeshow Test was a not significant (Chi-square = 3.250., df 8, p = 0.918) indicating that the data fit the model well. The model was able to correctly classify 100 percent of EDS >9 for an overall success rate of 93.8 percent. The area under the ROC curve was 70.6 (95% CI: 68.8 – 72.5) indicating an adequate fit.

520


10.00

chgdev

8.00

6.00

4.00

2.00

0.00 0.00000

0.20000

0.40000

0.60000


Figure 152: Deviance versus Predicted Probability EDS >12 Imputed The curve that extends from the lower left to the upper right corresponds to cases in which the dependent variable has a value of 0. The change in deviance plot illustrates that there are no significantly outlying cases.


0.05000

0.04000

0.03000

0.02000

0.01000

0.00000 0.00000

0.20000

0.40000

0.60000


Figure 153: Cooks Distances versus Predicted Probabilities EDS >12 Imputed The Cook’s distances plot generally follows that of the change in deviance. There are cases in the upper left that may be outliers.

521


522

Appendix F: Ecological Variables

Group Level Variables Technical Notes Segregation, Integration and Diversity Measures Used Three measures of ethnic diversity were selected based on the advice given to the US Bureau of the Census by Galster (2004). Those recommended were: Diversity Index (Maly 2000), Entropy Index (Modarres 2004), and the Simpson’s Index (Simpson 1949). For the following formulas let: = the proportion of individuals in group m (m = 1, 2, … , M) in tract i = the proportion of individuals in group m (m = 1,2, … , M) in the whole metropolitan study area 

Simpson’s D (or Herfindahl-Hirschman) index

Range: [0, 1]; 0 completely homogeneous; 1 completely heterogeneous 

Entropy (information, Shannon-Weaver, or Shannon-Wiener) Index

Range: [0, lnM]; 0 when on of the groups has a probability 1 (completely homogeneous); lnM when 

= 1/M (completely heterogeneous)

Maly (neighbourhood diversity) index

Range: [0, 1]; 0 = completely homogeneous when the entire suburb comprised of a group with a miniscule metropolitan presence; 1 = heterogeneous when all groups in suburb are matching their metropolitan wide share.

523

Appendix F: Ecological Variables

Index of concentration at the extremes (ICE) Brooks-Gunn argued that it is the positive effect of concentrated affluence that matters. But proportions of affluent and poor families are correlated across neighbourhoods with considerable multicollinearity into statistical estimates. To overcome this problem, Massey (2001) proposed conceptualizing the concentration of affluence and poverty as falling along a continuum ranging from a negative extreme where all families are poor to a neutral point where affluent and poor families are equally balanced to a positive extreme where all families are affluent. To measure variation on this continuum, Massey proposed an index of concentration at the extremes. This measure is computed as the number of affluent families in a Census tract minus the number of poor families, divided by the total number of families in the Census tract for whom there is data on income.

where Ai = the number of families or persons classified as affluent in neighbourhood i, Pi is the number of families or persons classified as poor in neighbourhood i, and Ti is the total population of neighbourhood i for whom there is data on income. The index varies from a theoretical minimum of -1.0 (all families are poor) to a theoretical maximum of +1.0 (all families are affluent) and passes through 0 (affluent and poor are equally balanced). ICE circumvents the multicollinearity problem posed by including measures of neighbourhood affluence and poverty in the same statistical models.

524

Appendix G: Ecological Bivariate

Group Level Correlation and Bivariate Analysis

ENTROPY

ENTROPY NORMALIZED

SIMPSON INDEX

MALY INDEX

VOLUNTEER %

NO VOLUNTEER %

HOME VISIT RATE

HOME VISIT %

NURSE VISIT RATE

POOR %

DENSITY ENTROPY ENTROPY NORMALIZED SIMPSON INDEX MALY INDEX VOLUNTEER % NO VOLUNTEER % HOME VISIT RATE HOME VISIT % NURSE VISIT RATE POOR % RICH % INDEX OF EXTREMES SMOKING % UNPLANNED PREG % PUBLIC ACCOMODATION % BREAST FEEDING % NOT SUPPORT % SUPPORT % NO PRACTICAL SUPPORT % PRACTICAL SUPPORT % NO EMOTIONAL SUPPORT % EMOTIONAL SUPPORT % NO REGRET LEAVING % REGRET LEAVING % POOR HEALTH % GOOD HEALTH % HOME OWNED % HOME RENTED % UNEMPLOYED RATE DIFFERENT ADDRESS 5 YEARS SAME ADDRESS 5 YEARS DIFFERENT ADDRESS 1 YEAR HOUSE VACANCY RATE APPARTMENT % SINGLE HOUSE % MULTISTOREY APPARTMENT % CLASS 3 % CLASS 2 % CLASS 1 % CLASS INDEX OF EXTREMES SOLE PARENT % SCHOOL YEAR 8 OR LESS % IRSD IRSD DECILE CRIME RATE

DENSITY

Table 76: Ecological Correlation Matrix

1.00 0.48 0.48 0.53 0.09 -0.58 0.27 -0.12 -0.12 -0.35 0.49 -0.55 -0.53 -0.21 0.33 0.30 0.18 0.54 -0.54 0.38 -0.38 0.35 -0.35 0.39 -0.39 0.38 -0.38 -0.40 0.36 0.60 -0.10 0.06 -0.14 0.09 0.34 -0.20 0.36 0.55 -0.25 -0.52 -0.55 0.37 0.61 -0.60 -0.61 0.10

0.48 1.00 1.00 0.98 0.25 -0.79 0.34 -0.02 -0.02 -0.48 0.41 -0.45 -0.44 -0.37 0.14 0.08 0.06 0.52 -0.52 0.11 -0.11 0.26 -0.26 0.39 -0.39 0.18 -0.18 -0.27 0.20 0.52 0.07 -0.09 -0.08 0.16 0.31 -0.24 0.35 0.53 -0.39 -0.43 -0.49 0.09 0.70 -0.51 -0.56 0.01

0.48 1.00 1.00 0.98 0.25 -0.79 0.34 -0.02 -0.02 -0.48 0.41 -0.45 -0.44 -0.37 0.14 0.08 0.06 0.52 -0.52 0.11 -0.11 0.26 -0.26 0.39 -0.39 0.18 -0.18 -0.27 0.20 0.52 0.07 -0.09 -0.08 0.16 0.31 -0.24 0.35 0.53 -0.39 -0.43 -0.49 0.09 0.70 -0.51 -0.56 0.01

0.53 0.98 0.98 1.00 0.16 -0.82 0.32 -0.01 -0.02 -0.46 0.50 -0.54 -0.53 -0.36 0.21 0.16 0.13 0.60 -0.60 0.16 -0.16 0.30 -0.30 0.46 -0.46 0.22 -0.22 -0.34 0.26 0.60 0.06 -0.08 -0.09 0.20 0.35 -0.26 0.39 0.60 -0.47 -0.50 -0.56 0.19 0.77 -0.60 -0.63 0.06

0.09 0.25 0.25 0.16 1.00 -0.16 0.06 -0.15 -0.15 -0.39 0.08 -0.09 -0.09 0.20 0.08 0.13 -0.02 0.00 0.00 0.03 -0.03 -0.02 0.02 0.14 -0.14 0.02 -0.02 -0.04 0.02 0.05 -0.15 0.11 -0.13 -0.03 0.15 -0.06 0.17 0.09 0.26 -0.14 -0.12 0.01 0.14 -0.09 -0.10 -0.12

-0.58 -0.79 -0.79 -0.82 -0.16 1.00 -0.42 0.05 0.06 0.39 -0.64 0.69 0.67 0.07 -0.42 -0.30 -0.38 -0.54 0.54 -0.22 0.22 -0.34 0.34 -0.55 0.55 -0.38 0.38 0.48 -0.43 -0.76 -0.01 0.08 0.07 -0.20 -0.32 0.33 -0.34 -0.81 0.41 0.77 0.81 -0.42 -0.83 0.79 0.80 -0.13

0.27 0.34 0.34 0.32 0.06 -0.42 1.00 0.13 0.12 -0.09 -0.16 0.11 0.14 -0.28 -0.05 -0.26 0.11 0.06 -0.06 -0.05 0.05 0.07 -0.07 0.01 -0.01 0.08 -0.08 0.27 -0.26 0.00 -0.02 0.09 -0.17 -0.31 -0.28 0.32 -0.27 0.12 0.22 -0.18 -0.15 -0.24 0.23 -0.04 -0.06 -0.24

-0.12 -0.02 -0.02 -0.01 -0.15 0.05 0.13 1.00 1.00 0.67 -0.13 0.14 0.14 -0.08 0.02 0.07 0.01 -0.08 0.08 0.00 0.00 -0.05 0.05 -0.04 0.04 0.05 -0.05 0.08 -0.08 -0.12 0.06 -0.05 0.00 -0.03 0.05 0.04 0.01 -0.08 0.05 0.06 0.07 -0.13 -0.07 0.10 0.08 -0.08

-0.12 -0.02 -0.02 -0.02 -0.15 0.06 0.12 1.00 1.00 0.67 -0.13 0.14 0.14 -0.08 0.02 0.07 0.01 -0.08 0.08 0.00 0.00 -0.05 0.05 -0.04 0.04 0.05 -0.05 0.07 -0.07 -0.12 0.06 -0.04 0.00 -0.03 0.05 0.04 0.01 -0.08 0.06 0.06 0.07 -0.13 -0.07 0.10 0.08 -0.08

-0.35 -0.48 -0.48 -0.46 -0.39 0.39 -0.09 0.67 0.67 1.00 -0.27 0.25 0.26 0.20 0.12 0.10 0.14 -0.21 0.21 -0.06 0.06 0.12 -0.12 -0.11 0.11 0.08 -0.08 0.08 -0.03 -0.20 0.17 -0.17 0.23 -0.13 -0.19 0.05 -0.22 -0.17 0.07 0.13 0.16 0.00 -0.44 0.22 0.24 0.20

0.49 0.41 0.41 0.50 0.08 -0.64 -0.16 -0.13 -0.13 -0.27 1.00 -0.98 -0.99 0.26 0.58 0.67 0.42 0.57 -0.57 0.47 -0.47 0.40 -0.40 0.64 -0.64 0.40 -0.40 -0.80 0.77 0.88 -0.28 0.16 -0.20 0.54 0.67 -0.68 0.69 0.82 -0.52 -0.76 -0.80 0.84 0.82 -0.92 -0.90 0.38

525

526

INDEX OF EXTREMES

SMOKING %

UNPLANNED PREG %


BREAST FEEDING %

NOT SUPPORT %

SUPPORT %


PRACTICAL SUPPORT %

NO EMOTIONAL SUPPORT %


RICH %


-0.55 -0.45 -0.45 -0.54 -0.09 0.69 0.11 0.14 0.14 0.25 -0.98 1.00 1.00 -0.26 -0.62 -0.68 -0.46 -0.61 0.61 -0.47 0.47 -0.46 0.46 -0.66 0.66 -0.43 0.43 0.82 -0.79 -0.93 0.21 -0.08 0.14 -0.49 -0.62 0.64 -0.64 -0.88 0.53 0.82 0.86 -0.84 -0.82 0.95 0.94 -0.42

-0.53 -0.44 -0.44 -0.53 -0.09 0.67 0.14 0.14 0.14 0.26 -0.99 1.00 1.00 -0.25 -0.61 -0.68 -0.44 -0.60 0.60 -0.47 0.47 -0.44 0.44 -0.65 0.65 -0.42 0.42 0.82 -0.78 -0.92 0.24 -0.11 0.16 -0.51 -0.65 0.66 -0.67 -0.86 0.53 0.80 0.84 -0.85 -0.83 0.94 0.93 -0.40

-0.21 -0.37 -0.37 -0.36 0.20 0.07 -0.28 -0.08 -0.08 0.20 0.26 -0.26 -0.25 1.00 0.42 0.47 0.48 -0.02 0.02 0.12 -0.12 0.15 -0.15 0.34 -0.34 0.20 -0.20 -0.29 0.34 0.20 -0.14 0.07 0.07 0.21 0.03 -0.21 0.01 0.22 0.09 -0.30 -0.27 0.40 -0.10 -0.20 -0.18 0.23

0.33 0.14 0.14 0.21 0.08 -0.42 -0.05 0.02 0.02 0.12 0.58 -0.62 -0.61 0.42 1.00 0.65 0.61 0.51 -0.51 0.33 -0.33 0.42 -0.42 0.50 -0.50 0.44 -0.44 -0.58 0.55 0.67 -0.12 0.01 -0.04 0.15 0.27 -0.42 0.27 0.66 -0.35 -0.64 -0.66 0.64 0.43 -0.67 -0.64 0.52

0.30 0.08 0.08 0.16 0.13 -0.30 -0.26 0.07 0.07 0.10 0.67 -0.68 -0.68 0.47 0.65 1.00 0.53 0.50 -0.50 0.47 -0.47 0.49 -0.49 0.59 -0.59 0.45 -0.45 -0.70 0.68 0.68 -0.09 -0.04 0.04 0.32 0.45 -0.54 0.47 0.62 -0.28 -0.62 -0.62 0.72 0.38 -0.67 -0.66 0.34

0.18 0.06 0.06 0.13 -0.02 -0.38 0.11 0.01 0.01 0.14 0.42 -0.46 -0.44 0.48 0.61 0.53 1.00 0.33 -0.33 0.20 -0.20 0.29 -0.29 0.52 -0.52 0.42 -0.42 -0.30 0.29 0.50 -0.08 -0.01 -0.04 0.02 -0.05 -0.14 -0.03 0.60 -0.24 -0.62 -0.61 0.42 0.31 -0.50 -0.47 0.31

0.54 0.52 0.52 0.60 0.00 -0.54 0.06 -0.08 -0.08 -0.21 0.57 -0.61 -0.60 -0.02 0.51 0.50 0.33 1.00 -1.00 0.48 -0.48 0.56 -0.56 0.57 -0.57 0.41 -0.41 -0.48 0.43 0.67 -0.07 -0.01 -0.08 0.17 0.42 -0.40 0.44 0.62 -0.49 -0.55 -0.58 0.39 0.60 -0.65 -0.63 0.29

-0.54 -0.52 -0.52 -0.60 0.00 0.54 -0.06 0.08 0.08 0.21 -0.57 0.61 0.60 0.02 -0.51 -0.50 -0.33 -1.00 1.00 -0.48 0.48 -0.56 0.56 -0.57 0.57 -0.41 0.41 0.48 -0.43 -0.67 0.07 0.01 0.08 -0.17 -0.42 0.40 -0.44 -0.62 0.49 0.55 0.58 -0.39 -0.60 0.65 0.63 -0.29

0.38 0.11 0.11 0.16 0.03 -0.22 -0.05 0.00 0.00 -0.06 0.47 -0.47 -0.47 0.12 0.33 0.47 0.20 0.48 -0.48 1.00 -1.00 0.64 -0.64 0.29 -0.29 0.35 -0.35 -0.34 0.31 0.41 -0.29 0.23 -0.24 0.25 0.36 -0.34 0.34 0.33 -0.07 -0.33 -0.34 0.34 0.36 -0.41 -0.39 0.24

-0.38 -0.11 -0.11 -0.16 -0.03 0.22 0.05 0.00 0.00 0.06 -0.47 0.47 0.47 -0.12 -0.33 -0.47 -0.20 -0.48 0.48 -1.00 1.00 -0.64 0.64 -0.29 0.29 -0.35 0.35 0.34 -0.31 -0.41 0.29 -0.23 0.24 -0.25 -0.36 0.34 -0.34 -0.33 0.07 0.33 0.34 -0.34 -0.36 0.41 0.39 -0.24

0.35 0.26 0.26 0.30 -0.02 -0.34 0.07 -0.05 -0.05 0.12 0.40 -0.46 -0.44 0.15 0.42 0.49 0.29 0.56 -0.56 0.64 -0.64 1.00 -1.00 0.48 -0.48 0.46 -0.46 -0.37 0.35 0.54 -0.05 0.00 0.00 0.08 0.23 -0.33 0.23 0.47 -0.16 -0.47 -0.47 0.39 0.29 -0.48 -0.48 0.36

NO REGRET LEAVING %

REGRET LEAVING %

POOR HEALTH %

GOOD HEALTH %

HOME OWNED %

HOME RENTED %

UNEMPLOYED RATE


SAME ADDRESS 5 YEARS

DIFFERENT ADDRESS 1 YEAR


EMOTIONAL SUPPORT %


-0.35 -0.26 -0.26 -0.30 0.02 0.34 -0.07 0.05 0.05 -0.12 -0.40 0.46 0.44 -0.15 -0.42 -0.49 -0.29 -0.56 0.56 -0.64 0.64 -1.00 1.00 -0.48 0.48 -0.46 0.46 0.37 -0.35 -0.54 0.05 0.00 0.00 -0.08 -0.23 0.33 -0.23 -0.47 0.16 0.47 0.47 -0.39 -0.29 0.48 0.48 -0.36

0.39 0.39 0.39 0.46 0.14 -0.55 0.01 -0.04 -0.04 -0.11 0.64 -0.66 -0.65 0.34 0.50 0.59 0.52 0.57 -0.57 0.29 -0.29 0.48 -0.48 1.00 -1.00 0.47 -0.47 -0.54 0.52 0.69 -0.07 -0.03 -0.02 0.18 0.35 -0.39 0.38 0.70 -0.36 -0.70 -0.70 0.55 0.54 -0.70 -0.68 0.30

-0.39 -0.39 -0.39 -0.46 -0.14 0.55 -0.01 0.04 0.04 0.11 -0.64 0.66 0.65 -0.34 -0.50 -0.59 -0.52 -0.57 0.57 -0.29 0.29 -0.48 0.48 -1.00 1.00 -0.47 0.47 0.54 -0.52 -0.69 0.07 0.03 0.02 -0.18 -0.35 0.39 -0.38 -0.70 0.36 0.70 0.70 -0.55 -0.54 0.70 0.68 -0.30

0.38 0.18 0.18 0.22 0.02 -0.38 0.08 0.05 0.05 0.08 0.40 -0.43 -0.42 0.20 0.44 0.45 0.42 0.41 -0.41 0.35 -0.35 0.46 -0.46 0.47 -0.47 1.00 -1.00 -0.37 0.34 0.53 -0.03 -0.04 0.10 0.04 0.19 -0.25 0.22 0.50 -0.21 -0.51 -0.50 0.38 0.35 -0.50 -0.48 0.26

-0.38 -0.18 -0.18 -0.22 -0.02 0.38 -0.08 -0.05 -0.05 -0.08 -0.40 0.43 0.42 -0.20 -0.44 -0.45 -0.42 -0.41 0.41 -0.35 0.35 -0.46 0.46 -0.47 0.47 -1.00 1.00 0.37 -0.34 -0.53 0.03 0.04 -0.10 -0.04 -0.19 0.25 -0.22 -0.50 0.21 0.51 0.50 -0.38 -0.35 0.50 0.48 -0.26

-0.40 -0.27 -0.27 -0.34 -0.04 0.48 0.27 0.08 0.07 0.08 -0.80 0.82 0.82 -0.29 -0.58 -0.70 -0.30 -0.48 0.48 -0.34 0.34 -0.37 0.37 -0.54 0.54 -0.37 0.37 1.00 -0.98 -0.79 -0.12 0.25 -0.21 -0.53 -0.61 0.76 -0.65 -0.69 0.46 0.65 0.68 -0.84 -0.55 0.79 0.78 -0.44

0.36 0.20 0.20 0.26 0.02 -0.43 -0.26 -0.08 -0.07 -0.03 0.77 -0.79 -0.78 0.34 0.55 0.68 0.29 0.43 -0.43 0.31 -0.31 0.35 -0.35 0.52 -0.52 0.34 -0.34 -0.98 1.00 0.75 0.11 -0.23 0.22 0.53 0.60 -0.75 0.64 0.65 -0.40 -0.63 -0.65 0.85 0.50 -0.75 -0.74 0.42

0.60 0.52 0.52 0.60 0.05 -0.76 0.00 -0.12 -0.12 -0.20 0.88 -0.93 -0.92 0.20 0.67 0.68 0.50 0.67 -0.67 0.41 -0.41 0.54 -0.54 0.69 -0.69 0.53 -0.53 -0.79 0.75 1.00 -0.10 -0.03 -0.05 0.33 0.52 -0.57 0.55 0.91 -0.57 -0.86 -0.90 0.80 0.81 -0.97 -0.96 0.43

-0.10 0.07 0.07 0.06 -0.15 -0.01 -0.02 0.06 0.06 0.17 -0.28 0.21 0.24 -0.14 -0.12 -0.09 -0.08 -0.07 0.07 -0.29 0.29 -0.05 0.05 -0.07 0.07 -0.03 0.03 -0.12 0.11 -0.10 1.00 -0.94 0.91 0.08 -0.14 -0.07 -0.11 -0.06 -0.15 0.08 0.07 -0.12 -0.28 0.14 0.13 0.02

0.06 -0.09 -0.09 -0.08 0.11 0.08 0.09 -0.05 -0.04 -0.17 0.16 -0.08 -0.11 0.07 0.01 -0.04 -0.01 -0.01 0.01 0.23 -0.23 0.00 0.00 -0.03 0.03 -0.04 0.04 0.25 -0.23 -0.03 -0.94 1.00 -0.90 -0.12 0.10 0.17 0.05 -0.08 0.24 0.06 0.07 0.01 0.22 -0.01 -0.01 -0.09

-0.14 -0.08 -0.08 -0.09 -0.13 0.07 -0.17 0.00 0.00 0.23 -0.20 0.14 0.16 0.07 -0.04 0.04 -0.04 -0.08 0.08 -0.24 0.24 0.00 0.00 -0.02 0.02 0.10 -0.10 -0.21 0.22 -0.05 0.91 -0.90 1.00 0.19 -0.08 -0.15 -0.05 -0.02 -0.12 0.02 0.02 0.03 -0.35 0.09 0.09 0.13

527

528

APPARTMENT %

SINGLE HOUSE %

MULTISTOREY APPARTMENT %

CLASS 3 %

CLASS 2 %

CLASS 1 %

CLASS INDEX OF EXTREMES

SOLE PARENT %


IRSD


HOUSE VACANCY RATE


0.09 0.16 0.16 0.20 -0.03 -0.20 -0.31 -0.03 -0.03 -0.13 0.54 -0.49 -0.51 0.21 0.15 0.32 0.02 0.17 -0.17 0.25 -0.25 0.08 -0.08 0.18 -0.18 0.04 -0.04 -0.53 0.53 0.33 0.08 -0.12 0.19 1.00 0.59 -0.61 0.59 0.28 -0.28 -0.20 -0.25 0.44 0.32 -0.35 -0.36 0.21

0.34 0.31 0.31 0.35 0.15 -0.32 -0.28 0.05 0.05 -0.19 0.67 -0.62 -0.65 0.03 0.27 0.45 -0.05 0.42 -0.42 0.36 -0.36 0.23 -0.23 0.35 -0.35 0.19 -0.19 -0.61 0.60 0.52 -0.14 0.10 -0.08 0.59 1.00 -0.69 0.98 0.40 -0.25 -0.35 -0.38 0.59 0.54 -0.55 -0.56 0.28

-0.20 -0.24 -0.24 -0.26 -0.06 0.33 0.32 0.04 0.04 0.05 -0.68 0.64 0.66 -0.21 -0.42 -0.54 -0.14 -0.40 0.40 -0.34 0.34 -0.33 0.33 -0.39 0.39 -0.25 0.25 0.76 -0.75 -0.57 -0.07 0.17 -0.15 -0.61 -0.69 1.00 -0.73 -0.47 0.38 0.39 0.44 -0.66 -0.43 0.57 0.56 -0.34

0.36 0.35 0.35 0.39 0.17 -0.34 -0.27 0.01 0.01 -0.22 0.69 -0.64 -0.67 0.01 0.27 0.47 -0.03 0.44 -0.44 0.34 -0.34 0.23 -0.23 0.38 -0.38 0.22 -0.22 -0.65 0.64 0.55 -0.11 0.05 -0.05 0.59 0.98 -0.73 1.00 0.43 -0.28 -0.37 -0.40 0.60 0.56 -0.58 -0.58 0.25

0.55 0.53 0.53 0.60 0.09 -0.81 0.12 -0.08 -0.08 -0.17 0.82 -0.88 -0.86 0.22 0.66 0.62 0.60 0.62 -0.62 0.33 -0.33 0.47 -0.47 0.70 -0.70 0.50 -0.50 -0.69 0.65 0.91 -0.06 -0.08 -0.02 0.28 0.40 -0.47 0.43 1.00 -0.55 -0.96 -0.99 0.73 0.76 -0.94 -0.93 0.37

-0.25 -0.39 -0.39 -0.47 0.26 0.41 0.22 0.05 0.06 0.07 -0.52 0.53 0.53 0.09 -0.35 -0.28 -0.24 -0.49 0.49 -0.07 0.07 -0.16 0.16 -0.36 0.36 -0.21 0.21 0.46 -0.40 -0.57 -0.15 0.24 -0.12 -0.28 -0.25 0.38 -0.28 -0.55 1.00 0.34 0.45 -0.35 -0.49 0.52 0.51 -0.24

-0.52 -0.43 -0.43 -0.50 -0.14 0.77 -0.18 0.06 0.06 0.13 -0.76 0.82 0.80 -0.30 -0.64 -0.62 -0.62 -0.55 0.55 -0.33 0.33 -0.47 0.47 -0.70 0.70 -0.51 0.51 0.65 -0.63 -0.86 0.08 0.06 0.02 -0.20 -0.35 0.39 -0.37 -0.96 0.34 1.00 0.99 -0.73 -0.68 0.90 0.88 -0.37

-0.55 -0.49 -0.49 -0.56 -0.12 0.81 -0.15 0.07 0.07 0.16 -0.80 0.86 0.84 -0.27 -0.66 -0.62 -0.61 -0.58 0.58 -0.34 0.34 -0.47 0.47 -0.70 0.70 -0.50 0.50 0.68 -0.65 -0.90 0.07 0.07 0.02 -0.25 -0.38 0.44 -0.40 -0.99 0.45 0.99 1.00 -0.74 -0.73 0.93 0.92 -0.37

0.37 0.09 0.09 0.19 0.01 -0.42 -0.24 -0.13 -0.13 0.00 0.84 -0.84 -0.85 0.40 0.64 0.72 0.42 0.39 -0.39 0.34 -0.34 0.39 -0.39 0.55 -0.55 0.38 -0.38 -0.84 0.85 0.80 -0.12 0.01 0.03 0.44 0.59 -0.66 0.60 0.73 -0.35 -0.73 -0.74 1.00 0.54 -0.83 -0.80 0.54

0.61 0.70 0.70 0.77 0.14 -0.83 0.23 -0.07 -0.07 -0.44 0.82 -0.82 -0.83 -0.10 0.43 0.38 0.31 0.60 -0.60 0.36 -0.36 0.29 -0.29 0.54 -0.54 0.35 -0.35 -0.55 0.50 0.81 -0.28 0.22 -0.35 0.32 0.54 -0.43 0.56 0.76 -0.49 -0.68 -0.73 0.54 1.00 -0.84 -0.84 0.14

-0.60 -0.51 -0.51 -0.60 -0.09 0.79 -0.04 0.10 0.10 0.22 -0.92 0.95 0.94 -0.20 -0.67 -0.67 -0.50 -0.65 0.65 -0.41 0.41 -0.48 0.48 -0.70 0.70 -0.50 0.50 0.79 -0.75 -0.97 0.14 -0.01 0.09 -0.35 -0.55 0.57 -0.58 -0.94 0.52 0.90 0.93 -0.83 -0.84 1.00 0.98 -0.38

CRIME RATE


IRSD DECILE


-0.61 -0.56 -0.56 -0.63 -0.10 0.80 -0.06 0.08 0.08 0.24 -0.90 0.94 0.93 -0.18 -0.64 -0.66 -0.47 -0.63 0.63 -0.39 0.39 -0.48 0.48 -0.68 0.68 -0.48 0.48 0.78 -0.74 -0.96 0.13 -0.01 0.09 -0.36 -0.56 0.56 -0.58 -0.93 0.51 0.88 0.92 -0.80 -0.84 0.98 1.00 -0.34

0.10 0.01 0.01 0.06 -0.12 -0.13 -0.24 -0.08 -0.08 0.20 0.38 -0.42 -0.40 0.23 0.52 0.34 0.31 0.29 -0.29 0.24 -0.24 0.36 -0.36 0.30 -0.30 0.26 -0.26 -0.44 0.42 0.43 0.02 -0.09 0.13 0.21 0.28 -0.34 0.25 0.37 -0.24 -0.37 -0.37 0.54 0.14 -0.38 -0.34 1.00

529


Bivariate Analysis Un-standardized Coefficients Variable

R

R Square

ZGNSPEMPC ZGYSPEMPC ZGSIMPSON ZGNSPTPC ZGYSPTPC ZGVOLYPC ZGIRSDDEC ZGDENSITY ZGENTROPY ZGENTNORM ZGSCH8LPC ZGPHEALTHPC ZGGHEALTHPC ZGNSPRACPC ZGYSPRACPC ZGOCC3R ZGNRGRETPC ZGYRGRETPC ZGRICHPC ZGIRSD ZGUNEMPLR ZGOCCICE ZGICE ZGPOORPC ZGPLPREGPC ZGOCC1R ZGHOUSEPC ZGOWNEDPC ZGAPPPC ZGAPHIPC ZGRENTPC ZGVISRATE ZGMALY ZGBREASTPC ZGONEPFPC ZGDADD5R ZGVACANPC ZGCRIMR ZGDADD1R ZGSADD5R ZGSMOKPC ZGACCPUBPC ZGUHVRATE

0.563 0.563 0.537 0.525 0.525 0.519 0.498 0.488 0.488 0.488 0.481 0.456 0.456 0.446 0.446 0.439 0.432 0.432 0.431 0.425 0.420 0.419 0.415 0.388 0.382 0.374 0.296 0.291 0.268 0.268 0.250 0.215 0.209 0.209 0.202 0.194 0.179 0.155 0.150 0.144 0.098 0.092 0.004

0.317 0.317 0.288 0.275 0.275 0.270 0.248 0.238 0.238 0.238 0.232 0.208 0.208 0.199 0.199 0.193 0.187 0.187 0.186 0.180 0.177 0.176 0.172 0.150 0.146 0.140 0.088 0.085 0.072 0.072 0.063 0.046 0.043 0.044 0.041 0.038 0.032 0.024 0.023 0.021 0.010 0.008 0.000

Standardized Coefficients

B

Beta

t

Sig

B

Std. Error

0.563 -0.563 0.537 0.525 -0.525 -0.519 -0.498 0.468 0.488 0.488 0.481 0.456 -0.456 0.446 -0.446 0.439 0.432 -0.432 -0.431 -0.425 0.420 -0.419 -0.415 0.388 0.382 -0.374 -0.296 -0.291 0.268 0.268 0.250 -0.215 -0.209 0.209 0.202 -0.194 0.179 0.155 -0.150 0.144 -0.098 0.092 -0.004

0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007

0.563 -0.563 0.537 0.525 -0.525 -0.519 -0.498 0.488 0.488 0.488 0.481 0.456 -0.456 0.446 -0.446 0.439 0.432 -0.432 -0.431 -0.425 0.420 -0.419 -0.415 0.388 0.382 -0.374 -0.296 -0.291 0.268 0.268 0.250 -0.215 -0.209 0.209 0.202 -0.194 0.179 0.155 -0.150 0.144 -0.098 0.092 -0.004

94.851 -94.851 88.625 85.890 -85.890 -84.691 -80.012 77.951 77.944 77.944 76.515 71.505 -71.505 69.421 -69.421 68.127 66.763 -66.763 -66.648 -65.351 64.562 -64.339 -63.561 58.588 57.661 -56.178 -43.257 -42.378 38.767 38.724 36.026 -30.732 -29.714 29.827 28.743 -27.532 25.349 21.806 -21.146 20.290 -13.711 12.826 -0.496

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.620

0.551 -0.574 0.525 0.513 -0.537 -0.531 -0.510 0.456 0.476 0.476 0.469 0.444 -0.469 0.433 -0.458 0.427 0.419 -0.445 -0.444 -0.437 0.408 -0.432 -0.428 0.375 0.369 -0.387 -0.310 -0.304 0.254 0.254 0.237 -0.229 -0.222 0.196 0.188 -0.208 0.165 0.141 -0.164 0.130 -0.112 0.078 -0.018

0.574 -0.551 0.548 0.537 -0.513 -0.507 -0.486 0.479 0.500 0.500 0.494 0.469 -0.444 0.458 -0.433 0.452 0.445 -0.419 -0.419 -0.412 0.433 -0.406 -0.402 0.400 0.395 -0.361 -0.283 -0.277 0.282 0.281 0.264 -0.202 -0.195 0.223 0.216 -0.180 0.193 0.168 -0.136 0.158 -0.084 0.106 0.011

Table 77: Bivariate Linear Regression, EDS>9, descending R

530

95% CI for B

Std. Error


Unstandardized Coefficients Variable ZGSIMPSON ZGENTROPY ZGENTNORM ZGSCH8LPC ZGVOLYPC ZGNSPTPC ZGYSPTPC ZGNRGRETPC ZGYRGRETPC ZGDENSITY ZGIRSDDEC ZGOCC3R ZGPHEALTHPC ZGGHEALTHPC ZGUNEMPLR ZGOCCICE ZGIRSD ZGRICHPC ZGICE ZGOCC1R ZGPOORPC ZGNSPEMPC ZGYSPEMPC ZGAPPPC ZGAPHIPC ZGVISRATE ZGHOUSEPC ZGOWNEDPC ZGPLPREGPC ZGSMOKPC ZGNSPRACPC ZGYSPRACPC ZGRENTPC ZGVACANPC ZGBREASTPC ZGMALY ZGDADD5R ZGDADD1R ZGONEPFPC ZGUHVRATE ZGCRIMR ZGSADD5R ZGACCPUBPC


95% CI for B

R

R Squared

B

Std. Error

Beta

t

Sig

B

Std. Error

0.699 0.654 0.654 0.619 0.586 0.528 0.528 0.527 0.527 0.516 0.515 0.484 0.476 0.476 0.466 0.463 0.455 0.448 0.431 0.415 0.402 0.340 0.340 0.318 0.316 0.297 0.294 0.286 0.276 0.254 0.239 0.239 0.218 0.180 0.169 0.156 0.128 0.127 0.125 0.110 0.096 0.091 0.020

0.488 0.427 0.427 0.384 0.343 0.278 0.278 0.278 0.278 0.266 0.265 0.234 0.226 0.226 0.217 0.215 0.207 0.201 0.185 0.172 0.161 0.116 0.116 0.101 0.100 0.088 0.086 0.082 0.076 0.064 0.057 0.057 0.048 0.032 0.029 0.024 0.016 0.016 0.016 0.012 0.009 0.008 0.000

0.699 0.654 0.654 0.619 -0.586 0.528 -0.528 0.527 -0.527 0.494 -0.515 0.484 0.476 -0.476 0.466 -0.463 -0.455 -0.448 -0.431 -0.415 0.402 0.340 -0.340 0.318 0.316 -0.297 -0.294 -0.286 0.276 -0.254 0.239 -0.239 0.218 0.180 0.169 -0.156 -0.128 -0.127 0.125 0.110 0.096 0.091 -0.020

0.005 0.005 0.005 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007

0.699 0.654 0.654 0.619 -0.586 0.528 -0.528 0.527 -0.527 0.516 -0.515 0.484 0.476 -0.476 0.466 -0.463 -0.455 -0.448 -0.431 -0.415 0.402 0.340 -0.340 0.318 0.316 -0.297 -0.294 -0.286 0.276 -0.254 0.239 -0.239 0.218 0.180 0.169 -0.156 -0.128 -0.127 0.125 0.110 0.096 0.091 -0.020

136.175 120.415 120.415 109.976 -100.69 86.581 -86.581 86.425 -86.425 83.902 -83.642 77.120 75.381 -75.381 73.403 -72.842 -71.286 -69.848 -66.482 -63.536 61.108 50.423 -50.423 46.746 46.488 -43.288 -42.826 -41.561 39.988 -36.527 34.361 -34.361 31.148 25.496 23.883 -22.005 -17.991 -17.889 17.543 15.482 13.408 12.740 -2.762

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.006

0.689 0.643 0.643 0.608 -0.597 0.516 -0.540 0.515 -0.539 0.483 -0.527 0.472 0.463 -0.488 0.454 -0.476 -0.468 -0.461 -0.443 -0.428 0.389 0.327 -0.353 0.305 0.303 -0.310 -0.307 -0.299 0.262 -0.267 0.226 -0.253 0.204 0.166 0.155 -0.170 -0.142 -0.141 0.111 0.096 0.082 0.077 -0.034

0.709 0.664 0.664 0.631 -0.574 0.540 -0.516 0.539 -0.515 0.506 -0.503 0.496 0.488 -0.463 0.478 -0.451 -0.443 -0.435 -0.418 -0.402 0.414 0.353 -0.327 0.331 0.330 -0.283 -0.280 -0.272 0.289 -0.240 0.253 -0.226 0.232 0.194 0.183 -0.142 -0.114 -0.113 0.139 0.124 0.110 0.105 -0.006

Table 78: Bivariate Linear Regression, EDS > 9 Imputed, descending R

531


Unstandardized Coefficients Variable ZGNSPEMPC ZGYSPEMPC ZGNSPTPC ZGYSPTPC ZGNRGRETPC ZGYRGRETPC ZGSIMPSON ZGIRSDDEC ZGRICHPC ZGVOLYPC ZGPHEALTHPC ZGGHEALTHPC ZGICE ZGUNEMPLR ZGENTROPY ZGENTNORM ZGOCC3R ZGIRSD ZGDENSITY ZGPOORPC ZGSCH8LPC ZGOCCICE ZGHOUSEPC ZGNSPRACPC ZGYSPRACPC ZGPLPREGPC ZGAPPPC ZGAPHIPC ZGOWNEDPC ZGOCC1R ZGRENTPC ZGVACANPC ZGMALY ZGONEPFPC ZGCRIMR ZGBREASTPC ZGDADD5R ZGVISRATE ZGACCPUBPC ZGSADD5R ZGDADD1R ZGUHVRATE ZGSMOKPC


R

R Squared

B

Std. Error

Beta

t

Sig

B

Std. Error

0.593 0.593 0.549 0.549 0.524 0.524 0.511 0.497 0.484 0.480 0.478 0.478 0.474 0.474 0.468 0.468 0.462 0.456 0.455 0.455 0.439 0.429 0.425 0.421 0.421 0.415 0.405 0.405 0.381 0.366 0.345 0.310 0.301 0.290 0.272 0.234 0.172 0.158 0.150 0.109 0.080 0.033 0.030

0.352 0.352 0.301 0.301 0.274 0.274 0.261 0.247 0.234 0.231 0.228 0.228 0.225 0.225 0.219 0.219 0.213 0.208 0.207 0.207 0.192 0.184 0.180 0.177 0.177 0.172 0.164 0.164 0.145 0.134 0.119 0.096 0.091 0.084 0.074 0.055 0.030 0.025 0.220 0.012 0.006 0.001 0.001

0.593 -0.593 0.549 -0.549 0.524 -0.524 0.511 -0.497 -0.484 -0.480 0.478 -0.478 -0.474 0.474 0.468 0.468 0.462 -0.456 0.436 0.455 0.439 -0.429 -0.425 0.421 -0.421 0.415 0.405 0.405 -0.381 -0.366 0.345 0.310 -0.301 0.290 0.272 0.234 -0.172 -0.158 0.150 0.109 -0.080 0.033 -0.030

0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007

0.593 -0.593 0.549 -0.549 0.524 -0.524 0.511 -0.497 -0.484 -0.480 0.478 -0.478 -0.474 0.474 0.468 0.468 0.462 -0.456 0.455 0.455 0.439 -0.429 -0.425 0.421 -0.421 0.415 0.405 0.405 -0.381 -0.366 0.345 0.310 -0.301 0.290 0.272 0.234 -0.172 -0.158 0.150 0.109 -0.080 0.033 -0.030

102.655 -102.65 91.456 -91.456 85.665 -85.665 82.825 -79.840 -77.045 -76.355 75.824 -75.824 -75.089 75.104 73.744 73.744 72.588 -71.339 71.148 71.283 68.000 -66.265 -65.389 64.732 -64.732 63.541 61.798 61.748 -57.377 -54.798 51.311 45.471 -43.995 42.245 39.460 33.472 -24.406 -22.302 21.087 15.222 -11.177 4.605 -4.246

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000

0.582 -0.604 0.537 -0.560 0.512 -0.536 0.499 -0.509 -0.496 -0.493 0.466 -0.490 -0.487 0.462 0.455 0.455 0.449 -0.468 0.424 0.443 0.426 -0.442 -0.437 0.408 -0.434 0.402 0.392 0.392 -0.394 -0.379 0.332 0.297 -0.314 0.277 0.259 0.220 -0.186 -0.172 0.136 0.095 -0.094 0.019 -0.045

0.604 -0.582 0.560 -0.537 0.536 -0.512 0.523 -0.485 -0.471 -0.468 0.490 -0.466 -0.462 0.487 0.480 0.480 0.474 -0.443 0.448 0.468 0.451 -0.417 -0.412 0.434 -0.408 0.428 0.418 0.418 -0.368 -0.353 0.359 0.324 -0.288 0.304 0.286 0.247 -0.159 -0.144 0.164 0.123 -0.066 0.047 -0.016

Table 79: Bivariate Linear Regression, EDS > 12, descending R

532

95% Cl for B


Unstandardized Coefficients Variable ZGNRGRETPC ZGYRGRETPC ZGSIMPSON ZGENTROPY ZGENTNORM ZGVOLYPC ZGNSPTPC ZGYSPTPC ZGIRSDDEC ZGNSPEMPC ZGYSPEMPC ZGDENSITY ZGSCH8LPC ZGOCC3R ZGIRSD ZGUNEMPLR ZGRICHPC ZGOCCICE ZGICE ZGVISRATE ZGPLPREGPC ZGOCC1R ZGPOORPC ZGNSPRACPC ZGYSPRACPC ZGHOUSEPC ZGOWNEDPC ZGPHEALTHPC ZGGHEALTHPC ZGAPHIPC ZGAPPPC ZGRENTPC ZGDADD5R ZGBREASTPC ZGDADD1R ZGSADD5R ZGONEPFPC ZGACCPUBPC ZGVACANPC ZGSMOKPC ZGCRIMR ZGMALY ZGUHVRATE


95% CI for B

R

R Squared

B

Std. Error

Beta

t

Sig.

Lower Bound

Upper Bound

0.476 0.476 0.453 0.436 0.436 0.386 0.381 0.381 0.368 0.358 0.358 0.352 0.351 0.289 0.287 0.276 0.271 0.270 0.266 0.250 0.236 0.233 0.225 0.200 0.200 0.196 0.192 0.191 0.191 0.187 0.186 0.156 0.156 0.146 0.142 0.132 0.118 0.076 0.075 0.059 0.032 0.019 0.018

0.226 0.226 0.206 0.191 0.191 0.149 0.145 0.145 0.136 0.128 0.129 0.124 0.123 0.084 0.082 0.076 0.073 0.073 0.071 0.063 0.056 0.054 0.065 0.040 0.040 0.038 0.037 0.037 0.037 0.035 0.034 0.024 0.024 0.021 0.020 0.017 0.014 0.006 0.006 0.003 0.001 0.000 0.000

0.476 -0.476 0.453 0.436 0.436 -0.386 0.381 -0.381 -0.368 0.358 -0.358 0.337 0.351 0.289 -0.287 0.276 -0.271 -0.270 -0.266 -0.250 0.236 -0.233 0.255 0.200 -0.200 -0.196 -0.192 0.191 -0.191 0.187 0.186 0.156 -0.156 0.146 -0.142 0.132 0.118 0.076 0.075 -0.059 0.032 -0.019 -0.018

0.006 0.006 0.006 0.006 0.006 0.007 0.007 0.007 0.007 0.007 0.007 0.006 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007 0.007

0.476 -0.476 0.453 0.436 0.436 -0.386 0.381 -0.381 -0.368 0.358 -0.358 0.352 0.351 0.289 -0.287 0.276 -0.271 -0.270 -0.266 -0.250 0.236 -0.233 0.255 0.200 -0.200 -0.196 -0.192 0.191 -0.191 0.187 0.186 0.156 -0.156 0.146 -0.142 0.132 0.118 0.076 0.075 -0.059 0.032 -0.019 -0.018

75.376 -75.376 70.907 67.614 67.614 -58.230 57.391 -57.391 -55.244 53.522 -53.522 52.423 52.238 42.083 -41.695 39.973 -39.236 -39.139 -38.410 -36.003 33.875 -33.386 36.769 28.440 -28.440 -27.787 -27.254 27.146 -27.146 26.536 26.341 22.010 -22.083 20.625 -20.053 18.537 16.607 10.573 10.544 -8.178 4.513 -2.659 -2.526

0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.000 0.008 0.012

0.463 -0.488 0.441 0.424 0.424 -0.398 0.368 -0.394 -0.382 0.345 -0.372 0.325 0.338 0.276 -0.300 0.262 -0.285 -0.284 -0.279 -0.264 0.223 -0.247 0.241 0.186 -0.214 -0.209 -0.206 0.177 -0.205 0.173 0.172 0.142 -0.170 0.132 -0.156 0.118 0.104 0.062 0.061 -0.073 0.018 -0.033 -0.032

0.488 -0.463 0.466 0.449 0.449 -0.373 0.394 -0.368 -0.355 0.372 -0.345 0.350 0.364 0.303 -0.273 0.289 -0.257 -0.257 -0.252 -0.236 0.250 -0.219 0.269 0.214 -0.186 -0.182 -0.178 0.205 -0.177 0.201 0.200 0.170 -0.143 0.160 -0.128 0.146 0.132 0.090 0.089 -0.045 0.046 -0.005 -0.004

Table 80: Bivariate Linear Regression, EDS > 12 Imputed, descending R

533


534

Appendix H: Ecological Regression

Group Level Likelihood Linear Regression Group-Level Linear Regression EDS >9 In the bivariate analysis all candidate variables were statistically significant with the grouplevel variable EDS >9. The correlation matrix identified variables with a correlation greater than 0.9. A set of variables was selected for analysis and is listed above. Table 81 shows the standardized and unstandardised coefficients, significance, confidence intervals and collinearity statistics for both Forward and Backward Regression studies. The correlation matrix revealed no signs of multicollinearity. Diagnostics are presented below for the Backward Regression.

Forward Regression Unstandardized Coefficients Std. B Error -1.39E.085 015 .382 .104

(Constant) NOT SUPPORT % NURSE VISIT RATE ENTROPY

Standardized Coeff.

95% Confidence Interval for B

Beta

t

Sig

Lower

Upper

Collinearity Statistics Toleranc e VIF

.000

1.000

-.170

.170

.382

3.664

.000

.175

.589

.679

1.473

.320

.099

.320

3.225

.002

.123

.517

.748

1.336

.353

.117

.353

3.028

.003

.122

.585

.542

1.844

-.251 a Dependent Variable: EDS9 %

.100

-.251

-2.501

.014

-.450

-.052

.734

1.361

APPARTMENT %

Backward Regression

(Constant) ENTROPY NURSE VISIT RATE NOT SUPPORT % APPARTMENT %

Unstand ardized Coefficients Std. B Error -1.39E.085 015 .353 .117

Standar dized Coefficie nts Beta


Sig

Lower

Upper

Colinearity Statistics Toleranc e VIF

.000

1.000

-.170

.170

.353

3.028

.003

.122

.585

.542

1.844

.320

.099

.320

3.225

.002

.123

.517

.748

1.336

.382 -.251

.104 .100

.382 -.251

3.664 -2.501

.000 .014

.175 -.450

.589 -.052

.679 .734

1.473 1.361


Table 81: EDS9 Coefficients

535


ANOVA

Model 9

Sum of Squares

df

Mean Square

Regression

29.174

4

7.294

Residual

70.826

96

.738

100.000

100

Total

F

Sig.

9.886

.000(i)

Table 82: ANOVA for Model 9 of the Backward Regression – EDS9% ANOVA tests the acceptability of the model. The regression row displays information about the variation accounted for by the model which in this case is 29 percent and 71 percent of the variation is not accounted for by the model. The significance value of the F statistic is less than 0.05 (0.000) which means that the variation explained by the model is not due to chance.

Change Statistics R Square

Adjusted R Square

Std. Error of the Estimate

R Square Change

F Change

df1

Model 1

R

df1

df2

.578

.334

.243

.86980191

.334

3.682

12

88

.000

2

.578

.334

.252

.86490494

.000

.001

1

88

.979

3

.578

.334

.260

.86030522

.000

.045

1

89

.832

4

.575

.331

.265

.85760868

-.003

.430

1

90

.513

5

.571

.326

.268

.85574338

-.004

.600

1

91

.440

6

.570

.325

.274

.85189457

-.001

.165

1

92

.685

7

.566

.320

.277

.85025680

-.005

.639

1

93

.426

8

.553

.305

.269

.85505269

-.015

2.075

1

94

.153

9

.540

.292

.262

.85893408

-.014

1.874

1

95

.174

DurbinWatson

1.905

Table 83: Model Summary and Durbin Watson for Backward Regression - EDS9% Model Fit The model summary table reports the strength of the relationship between the model and the dependent variable. R for the final backward model is 0.54 indicating a moderate relationship. The R Square is 0.292 indicating that approximately 30 percent of the variation in EDS9 at suburb level is explained by the model.

536


Checking independence The Durbin-Watson estimate is 1.905 which is close to two which indicates that the data points were independent. The independence assumption is satisfied. Residual Statistics Minimum

Maximum

Mean

Std. Deviation

N

Predicted Value

-1.2192985

2.1419802

.0000000

.54013235

101

Std. Predicted Value

-2.257

3.966

.000

1.000

101

Standard Error of Predicted Value

.092

.598

.174

.079

101

Adjusted Predicted Value

-1.1738553

1.5848877

-.0004866

.53001350

101

Residual

-2.81997681

2.98291850

.00000000

.84158009

101

Std. Residual

-3.283

3.473

.000

.980

101

Stud. Residual

-3.388

3.521

.000

1.008

101

Deleted Residual

-3.00266123

3.06689429

.00048660

.89261599

101

Stud. Deleted Residual

-3.592

3.754

.000

1.033

101

Mahal. Distance

.146

47.549

3.960

6.108

101

Cook's Distance

.000

.173

.013

.031

101

Centered Leverage Value

.001

.475

.040

.061

101


Table 84: Residuals Statistics Checking for outliers The case wise analysis was done at three standard deviations. The Standardised Residual minimum and maximums exceed +_ 3 indicating that there are outliers. The Casewise diagnostics have identified those as the suburbs of Pairiewood and Regents Park.

Case Number

SSC_NAME

Std. Residual

EDS9 %

Predicted Value

Residual

77

Prairiewood

3.473

3.31193

.3290161

2.98291844

81

Regents Park

-3.283

-2.59717

.2228085

-2.81997677


Table 85: Casewise Diagnostics for (+-3 SD) Backward Regression – EDS9%

537


Analysis with +- 2 standard deviations identified seven outlying suburbs. They are Chullora, Long Point, Mount Lewis, Old Guildford, Prairiewood, Regents Park and Ruse.

SSC_NAME

Std. Residual

EDS9 %

Predicted Value

Case Number 28

Chullora

-2.343

-2.59717

-.5846683

Residual -2.01250000

59

Long Point

-2.537

-2.59717

-.4177594

-2.17940889

Mount Lewis

-2.040

-2.59717

-.8453517

-1.75181663

Old Guildford

2.751

2.19963

-.1629709

2.36260373

Prairiewood

3.473

3.31193

.3290161

2.98291844

-3.283

-2.59717

.2228085

-2.81997677

2.334

1.45983

-.5449727

2.00480033

68 72 77 81 85

Regents Park Ruse


Table 86: Casewise Diagnostics for (+-2 SD) Backward Regression – EDS9% Normality of the Error Term 25

20

Frequency

Mean =4.6E-17 Std. Dev. =0.98 N =101 15

10

5

0 -4

-2

0

2

4

Regression Standardized Residual

Figure 154: Histogram of EDS9 Regression Residual The histogram of the residuals approximates the curve of a normal distribution and confirms the assumption of normality of the error term.

538


1.0

Expected Cum Prob

0.8

0.6

0.4

0.2

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Observed Cum Prob

Figure 155: Normal P-P Plot of Regression Standardised Residual The P-P plotted residuals follow the 45-degree line and confirms the assumption of normality.

Checks for Independence of the Error Term


4

Prairiewood

2

0

-2

Regents Park

R Sq Linear = -2.22E-16

-4 -2

0

2

Regression Standardized Predicted Value

Figure 156: Scatter Plot of Regression Standardised Residual to Standardised Predicted Value – 95% Confidence Intervals Shown A check for the independence of the error term was undertaken using a plot of residuals by the predicted values. There is a good scatter with no relationship identified. Further

539


checks were undertaken by plotting the standardised residuals against the significant variables. All plots were similar to that of Entropy below.

3.00000

Standardized Residual

2.00000

1.00000

0.00000

-1.00000

R Sq Linear = 0

-2.00000

-3.00000 -2.00000

-1.00000

0.00000

1.00000

2.00000

3.0000

ENTROPY

Figure 157: Scatter Plot of Standardised Residual to Standardised Variable Entropy

540


Identification of Influential Points A plot of the Cook’s Distance against Centered Leverage value was undertaken. The following Scatter Plot identifies two points far to the right of the others. The suburb of Chullora has high leverage and high influence. Its high leverage gives it extra weigh in the computation of the regression line, and the high influence indicates that it did affect the slop of the regression line.

6.00000

Chullora

Cook's Distance

5.00000

4.00000

3.00000

2.00000

1.00000

0.00000

Moorebank Glen Alpine Regents Park Campbelltown Holsworthy Cabramatta Leumeah Cabramatta West Yennora Camden ElderslieAirds IngleburnBirrong Miller Currans HillBonnyrigg Wattle Grove Bankstown Canley Heights ClaymoreCamden South 0.00000

0.20000

0.40000

Blairmount

0.60000


Figure 158: Scatter Plot of Cook’s Distance to Centered Leverage Value

Note: Results of the analysis of EDS > 9 (Imputed missing data) are presented in the summary table below. The full results and diagnostics are not presented.

541


Group-Level Linear Regression EDS >12 In the bivariate analysis all candidate variables were statistically significant with the grouplevel variable EDS >12. The correlation matrix identified variables with a correlation greater than 0.9. A set of variables was selected for analysis and is listed above. Table 87 shows the standardized and unstandardized coefficients, significance, confidence intervals and collinearity statistics for both Forward and Backward Regression studies. The correlation matrix revealed no signs of multicollinearity. Diagnostics are presented below for the Backward Regression.

Forward Regression M o d el

1

(Constant)

Unstand ardized Coefficients Std. B Error -3.24E.094 016

NOT SUPPORT %

.336

Standar dized Coefficie nts

95% Confidence Interval for B

Beta

.095

t

.336

Sig

Lower

Upper

.000

1.000

-.187

.187

3.555

.001

.149

.524


1.000

1.000


Backward Regression

(Constant) ENTROPY NO VOLUNTEER % NURSE VISIT RATE SMOKING % NO REGRET LEAVING % CLASS 3 % SCHOOL YEAR 8 OR LESS % HOME OWNED %

Unstand ardized Coefficients Std. B Error 1.18E.087 015 .318 .137

Beta


Sig

Lower

Upper


.000

1.000

-.172

.172

.318

2.316

.023

.045

.591

.403

2.480

-.260

.112

-.260

-2.332

.022

-.482

-.039

.610

1.639

.324

.102

.324

3.188

.002

.122

.526

.734

1.363

-.387

.151

-.387

-2.562

.012

-.688

-.087

.333

3.007

.328

.160

.328

2.047

.044

.010

.646

.296

3.377

.710

.199

.710

3.563

.001

.314

1.106

.191

5.231

-.357

.160

-.357

-2.230

.028

-.676

-.039

.296

3.382

.437

.167

.437

2.619

.010

.106

.768

.273

3.661


Table 87: EDS > 12 Coefficients

542

Standar dized Coefficie nts


ANOVA

Model 7

Sum of Squares

df

Mean Square

Regression

30.081

8

3.760

Residual

69.919

92

.760

100.000

100

Total

F

Sig.

4.948

.000

Table 88: ANOVA of Model 7 of the Backward Regression – EDS12% ANOVA Tests the acceptability of the model. The regression row displays information about the variation accounted for by the model which in this case is 30 percent and 70 percent of the variation is not accounted for by the model. The significance value of the F statistic is less than 0.05 (0.000) which means that the variation explained by the model is not due to chance. Model Fit

Change Statistics

Model R

R Square

Adjusted R Square

Std. Error of the Estimate

R Square Change

F Change

df1

df2

Sig. F Change

1

.574

.330

.221

.88280695

.330

3.022

14

86

.001

2

.574

.330

.229

.87782466

.000

.021

1

86

.886

3

.573

.328

.237

.87360881

-.001

.157

1

87

.693

4

.571

.326

.243

.87001287

-.002

.269

1

88

.605

5

.569

.323

.248

.86703442

-.003

.385

1

89

.537

6

.565

.320

.252

.86468480

-.004

.507

1

90

.478

7

.548

.301

.240

.87177166

-.019

2.514

1

91

.116

DurbinWatson

2.049

Table 89: Model Summary for Backward Regression The model summary table reports the strength of the relationship between the model and the dependent variable. R for the final backward model is 0.672 indicating a good relationship. The R Square is 0.452 indicating that approximately 45 percent of the variation in EDS9M at suburb level is explained by the model.

543


Checking independence The Durbin-Watson estimate is 2.049 which is close to two which indicates that the data points were independent. The independence assumption is satisfied. Residual Statistics Minimum Predicted Value

Maximum

Mean

Std. Deviation

N

-1.4707472

1.3670754

.0000000

.54846426

101

-2.682

2.493

.000

1.000

101

.122

.738

.241

.099

101

-1.5474583

3.4577341

.0222924

.65299609

101

-2.00226092

2.47272325

.00000000

.83617400

101

Std. Residual

-2.297

2.836

.000

.959

101

Stud. Residual

-3.307

2.892

-.008

1.061

101

-5.41827726

3.50625610

-.02229242

1.10010400

101

-3.504

3.016

-.009

1.080

101

Mahal. Distance

.981

70.700

7.921

9.595

101

Cook's Distance

.000

3.077

.051

.321

101


.010

.707

.079

.096

101

Std. Predicted Value Standard Error of Predicted Value Adjusted Predicted Value Residual

Deleted Residual Stud. Deleted Residual


Table 90: Residuals Statistics backward regression – EDS12% Checking for outliers There were no outliers at three standard deviations and the Standardised Residual minimum and maximum do not exceed +_3 indicating that there are not significant outliers. Analysis with +- 2 standard deviations identified three outlying suburbs. They are Glen Alpine, Macquarie Links and Regents Park.

Case Number 42

62

SSC_NAME

Std. Residual

EDS12 %

Predicted Value

Glen Alpine

-2.297

-1.14006

.8622008

2.0022610 1

Macquarie Links

2.291

2.62049

.6232014

1.9972864 3

Old Guildford

2.836

2.62049

.1477647

2.4727231 3

72


Table 91: Casewise Diagnostics at + or – 2 Standard Deviations

544

Residual


Normality of the Error Term 20

Frequency

15

Mean =7.03E-17 Std. Dev. =0.959 N =101

10

5

0 -3

-2

-1

0

1

2

3


Figure 159: Histogram of EDS9M Regression Residual The histogram of the residuals approximates the curve of a normal distribution and confirms the assumption of normality of the error term.

1.0

Expected Cum Prob

0.8

0.6

0.4

0.2

0.0 0.0

0.2

0.4

0.6

0.8

1.0

Observed Cum Prob

Figure 160: Normal P-P Plot of Regression Standardised Residual The P-P plotted residuals follow the 45-degree line and confirms the assumption of normality.

545


Checks for Independence of the Error Term

3

Old Guildford Macquarie Links


2

1

0

-1

R Sq Linear = 1.11E-16

-2

-3 -3

-2

-1

0

1

2

3

Regression Standardized Predicted Value

Figure 161: Scatter Plot of Regression Standardised Residual to Standardised Predicted Value Checks for the independence of the error term was undertaken using a plot of residuals by the predicted values. There is a good scatter with no relationship identified. Further checks were undertaken by plotting the standardised residuals against the significant variables. All plots were similar to that of Entropy below. 3.00000

Standardized Residual

2.00000

1.00000

0.00000

-1.00000

-2.00000 R Sq Linear = 1.11E-16

-3.00000 -2.00000

-1.00000

0.00000

1.00000

2.00000

3.0000

ENTROPY

Figure 162: Scatter Plot of Standardised Residual to Standardised Variable Entropy

546


Identification of Influential Points A plot of the Cook’s Distance against Centered Leverage value was undertaken. The following Scatter Plot identifies two points far to the right of the others. The suburb of Chullora has high leverage and high influence. Its high leverage gives it extra weigh in the computation of the regression line, and the high influence indicates that it did affect the slop of the regression line.

4.00000

Chullora

Cook's Distance

3.00000

2.00000

Blairmount

1.00000

0.00000

Bankstown Macquarie Links Hammondville Moorebank Glen Alpine Currans Hill Yennora Leumeah IngleburnCanley Heights Airds Camden Canley Vale Ashcroft BonnyriggCarramar Glenfield Wattle Grove Camden South Cabramatta West Bass Hill Claymore 0.00000

0.20000

0.40000

0.60000

0.8000


Figure 163: Scatter Plot of Cook’s Distance to Centered Leverage Value Note: Results of the analysis of EDS > 9 (Imputed missing data) are presented in the summary table below. The full results and diagnostics are not presented. Note: Results of the analysis of EDS > 9 (Imputed missing data) are presented in the summary table below. The full results and diagnostics are not presented.

547


548

Appendix I: Ecological Bayes Bivariate

Group Level Bayesian CAR Bivariate Analysis Variable

mean

sd

MC error

2.5%

Median

97.5%

No Covariate ZGDENSITY

Pd 23.404

0.09085

0.02604

0.000052

0.03979

0.091

DIC 565.468

0.142

18.318

561.917

Sig

ZGENTROPY

0.1171

0.03565

0.001105

0.04475

0.1173

0.1854

17.593

564.483

Sig

ZGSIMPSON

0.1489

0.03296

0.001016

0.08368

0.1494

0.2151

13.222

560.061

Sig

ZGMALY

-0.04259

0.01996

0.000028

-0.08226

-0.0423

-0.003647

23.994

564.131

Sig

ZGVOLYPC

-0.1632

0.04121

0.001055

-0.2438

-0.1638

-0.08221

15.460

561.749

Sig

ZGUHVRATE

-0.009803

0.0316

0.000043

-.07124

-0.009914

0.05204

25.404

567.171

NS

ZGVISRATE

-0.01171

0.05891

0.001203

-0.127

-0.01071

0.1016

25.430

567.276

NS

ZGPOORPC

0.0747

0.02842

0.000052

0.01808

0.07488

0.1297

21.602

564.252

Sig

ZGICE

-0.08148

0.02858

0.000047

-0.138

-0.08131

-0.02626

20.796

563.678

Sig

ZGSMOKPC

0.01835

0.04827

0.000087

-0.07613

0.018

0.1138

24.935

567.294

NS

ZGPLPREGPC

0.1207

0.03542

0.000597

0.05243

0.1208

0.19

18.828

561.784

Sig

ZGACCPUBPC

0.04258

0.03033

0.000039

-0.018

0.04288

0.101

24.513

566.397

NS

ZGBREASTPC

0.08696

0.04355

0.000068

0.000074

0.08695

0.1729

22.781

565.391

Sig

ZGNSPTPC

0.09624

0.02284

0.000041

0.05113

0.09604

0.1415

5.907

560.012

Sig

ZGNRGRETPC

0.1231

0.03607

0.000053

0.05262

0.1234

0.1945

18.237

562.061

Sig

ZGOWNEDPC

-0.06459

0.02706

0.000034

-0.1179

-0.06473

-0.01137

22.968

564.055

Sig

ZGRENTPC

0.05984

0.0276

0.000032

0.005643

0.05972

0.114

23.162

564.836

Sig

ZGUNEMPLR

0.0842

0.02617

0.000038

0.03258

0.08429

0.1355

19.834

562.289

Sig

ZGDADD5R

-0.007428

0.0332

0.000054

-0.07143

-0.007947

0.059

25.427

567.023

NS

ZGDADD1R

0.01445

0.03757

0.000054

-0.059

0.0146

0.088

25.703

567.018

NS

ZGVACANPC

0.03729

0.02753

0.000047

-0.01697

0.03734

0.09

24.484

566.439

NS

ZGAPPPC

0.02724

0.01918

0.000032

-0.01075

0.0272

0.06495

24.575

566.794

NS

ZGHOUSEPC

-0.04423

0.02091

0.000029

-0.08423

-0.04448

-0.002423

23.109

565.345

ZGAPHIPC

0.02773

0.01911

0.000033

-0.0104

0.02772

0.065

24.545

566.963

NS

ZGOCC3R

0.09529

0.02894

0.000048

0.03817

0.09567

0.1503

19.681

561.909

Sig

ZGOCC1R

-0.08923

0.03594

0.000063

-0.1592

-0.08928

-0.01833

20.926

564.747

Sig

ZGONEPFPC

0.05921

0.03000

0.000038

-1.577

0.05912

0.1181

23.082

565.160

NS

ZGSCH8LPC

0.1014

0.03139

0.000077

0.039

0.101

0.164

18.680

562.840

Sig

ZGIRSD

-0.08535

0.02715

0.000040

-0.1377

-0.08538

-0.03107

20.14

562.097

Sig

ZGIRSDDEC

-0.1162

0.03036

0.000084

-0.1735

-0.1167

-0.05529

13.635

564.481

Sig

ZGCRIMR

0.05893

0.03347

0.000045

-0.00564

0.0583

0.1265

24.91

564.863

NS

Table 92: Bivariate spatial CAR models for percent EDS >9

549


For EDS >9 the following variables were assessed as significant: the IRSD, ICE, entropy, Simpson and Maly indexes, population density, and percentage of volunteers, poor, unplanned pregnancy, breast feeding, no social support, no regret leaving, owned accommodation, rented accommodation, living in a house, unemployment, high occupational class, low occupational class, and schooling to year 8.

Variable

mean

sd

MC error

2.5%

Median

97.5%

No Covariate ZGDENSITY ZGENTROPY ZGSIMPSON ZGMALY ZGVOLYPC ZGUHVRATE ZGVISRATE ZGPOORPC ZGICE ZGSMOKPC ZGPLPREGPC ZGACCPUBPC ZGBREASTPC ZGNSPTPC ZGNRGRETPC ZGOWNEDPC ZGRENTPC ZGUNEMPLR ZGDADD5R ZGDADD1R ZGVACANPC ZGAPPPC ZGHOUSEPC ZGAPHIPC ZGOCC3R ZGOCC1R ZGONEPFPC ZGSCH8LPC ZGIRSD ZGIRSDDEC ZGCRIMR

0.1304 0.1846 0.2314 -0.07817 -0.2496 0.0056 0.01246 0.1316 -0.1477 0.03699 0.184 0.05833 0.1207 0.1433 0.2175 -0.1128 0.1064 0.1401 0.01454 0.06019 0.1003 0.0755 -0.09322 0.07539 0.152 -0.1275 0.09993 0.1378 -0.1351 -0.1903 0.1134

0.03855 0.05215 0.05114 0.02764 0.06033 0.04786 0.08612 0.03963 0.0396 0.07117 0.05111 0.04383 0.06109 0.03102 0.05151 0.03894 0.03873 0.03517 0.04994 0.05641 0.03869 0.02636 0.02859 0.02624 0.04108 0.05205 0.04215 0.04478 0.03784 0.04365 0.04617

0.000086 0.001643 0.001624 0.000038 0.001307 0.000063 0.002108 0.000084 0.0000958 0.001271 0.001015 0.000055 0.001125 0.000067 0.000089 0.0000635 0.000058 0.000068 0.000084 0.000089 0.000071 0.000043 0.000055 0.000041 0.000071 0.001156 0.000072 0.001028 0.000086 0.001352 0.00006312

0.05386 0.07951 0.1284 -0.1326 -0.3688 -0.08612 -0.1585 0.05096 -0.2225 -0.09899 0.08263 -0.02998 0.000080 0.08106 0.1147 -0.1883 0.02869 0.07134 -0.08144 -0.05096 0.02514 0.02353 -0.1483 0.02315 0.07262 -0.2285 0.01687 0.04909 -0.2059 -0.2729 0.0204

0.1305 0.1846 0.2313 -0.07796 -0.2491 0.005266 0.01226 0.1325 -0.1485 0.03779 0.1846 0.05952 0.1209 0.1434 0.2172 -0.1136 0.1066 0.1407 0.01346 0.06072 0.1003 0.07522 -0.09345 0.07535 0.1527 -0.1278 0.1001 0.1376 -0.1361 -0.192 0.1144

0.206 0.286 0.3305 -0.02388 -0.131 0.0989 0.1798 0.207 -0.06825 0.1784 0.2836 0.1412 0.2376 0.2039 0.3178 -0.03546 0.1814 0.2087 0.1152 0.1715 0.1759 0.127 -0.03768 0.1265 0.2328 -0.02457 0.181 0.2248 -0.05622 -0.09867 0.2023

Pd

DIC

21.481

473.277

15.545 14.551 12.142 19.074 13.158 22.076 22.230 15.218 13.031 22.597 13.814 22.701 19.899 11.433 13.800 17.339 17.747 12.931 22.746 23.048 19.185 17.558 16.436 17.730 14.607 17.755 18.677 15.848 13.951 7.529 18.532

470.901 469.984 465.104 470.150 468.012 475.361 475.631 471.154 471.172 475.357 470.470 474.493 473.967 466.878 465.737 471.532 472.030 469.391 475.155 473.897 471.436 471.806 471.705 471.869 468.481 473.669 472.724 471.488 471.464 472.089 472.464

Table 93: Bivariate spatial CAR models for percent EDS >12 For EDS >12 all were assessed as significant except for the changes in address variables, nursing visit variables, and percent smoking or in public accommodation.

550

Sig Sig Sig Sig Sig NS NS Sig Sig NS Sig NS Sig Sig Sig Sig Sig Sig NS NS Sig Sig Sig Sig Sig Sig Sig Sig Sig Sig Sig


For those variables with the greatest fall in DIC in the Bivariate analysis I systematically added other variables that had been significant in the Bivariate analysis. The results of adding a second variable to “percent no support” are reported in Table 94. Two of the indices of ethnic integration or segregation (Entropy and Simpson Indexes) were significant as was the percent of volunteerism and IRSD decile. Interestingly some of the other variables were significant when added later (see final model).

Variable

mean

Pd

DIC

No Covariate

23.404

565.468

NSPT +

5.907

560.012

ZGDENSITY ZGENTROPY ZGSIMPSON ZGMALY ZGVOLYPC ZGUHVRATE ZGVISRATE ZGPOORPC ZGICE ZGSMOKPC ZGPLPREGPC ZGACCPUBPC ZGBREASTPC ZGNSPTPC ZGNRGRETPC ZGOWNEDPC ZGRENTPC ZGUNEMPLR ZGDADD5R ZGDADD1R ZGVACANPC ZGAPPPC ZGHOUSEPC ZGAPHIPC ZGOCC3R ZGOCC1R ZGONEPFPC ZGSCH8LPC ZGIRSD ZGIRSDDEC ZGCRIMR

sd

MC error

2.5%

Median

97.5%

0.04212 0.08008 0.09729 0.03253 -0.1016

0.032 0.03232 0.03829 0.02389 0.0244

0.000079 0.000094 0.000057 0.000091 0.000055

-0.02046 0.01514 0.02246 0.01617 -0.186

0.04191 0.0801 0.09675 0.03289 -0.1021

0.1063 0.1436 0.172 0.07774 -0.01374

17.700 13.194 13.192 14.174 11.890

559.650 557.988 557.853 562.479 558.864

0.01844 -0.02473

0.03191 0.02855

0.000053 0.000065

-0.04387 -0.09057

0.01831 -0.025

0.08129 0.04054

16.957 18.136

561.439 560.334

0.07401

0.2379

0.00006

-0.00198

0.07153

0.1434

12.854

560.365

0.02219 0.0679 -0.0197 0.01747 0.02992

0.02556 0.02575 0.02868 0.02806 0.03258

0.000062 0.000060 0.000044 0.000043 0.000061

-0.06472 -0.01298 -0.07601 -0.0382 -0.03451

0.02252 0.06845 -0.0197 0.01781 0.0298

0.1052 0.1468 0.03603 0.07174 0.09313

18.003 15.653 17.296 17.276 17.585

560.931 559.617 561.758 561.774 560.141

-0.003895 -0.01413 0.001286 -0.01379 0.04414 -0.04239 0.01663 0.02696 -0.02822 -0.07052 0.01901

0.02697 0.02012 0.023091 0.02004 0.03212 0.03578 0.03067 0.04044 0.03273 0.03309 0.03127

0.000050 0.000038 0.000047 0.000038 0.000066 0.000082 0.000057 0.001278 0.000069 0.000073 0.000061

-0.05599 -0.05384 -0.04351 -0.05313 -0.01963 -0.1121 -0.0432 -0.052 -0.09194 -0.1361 -0.04287

0.003636 -0.01398 0.001114 -0.01372 0.04401 -0.04295 0.01695 0.02678 -0.0286 -0.07102 0.01896

0.04895 0.02542 0.04683 0.02581 0.107 0.02894 0.07693 0.1069 0.03667 -0.004736 0.0804

19.201 16.492 18.324 16.563 13.997 15.676 17.844 18.001 17.439 12.745 18.231

560.650 561.847 560.578 561.893 562.219 562.350 562.098 560.490 559.976 559.543 560.419

NS Sig Sig NS Sig NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS NS Sig NS

Table 94: EDS >9 Spatial regression with %No Support and second covariate.

551


552

Appendix J: Ecological EDS > 9

Final Spatial Regression Model EDS >9 I have elected to report here in more detail the best fitting of the EDS >9 models namely that including the covariates:% No Support, Entropy Index, %living in apartments and %smoking. To assist the map decomposition process I have elected to use the full BYM model with both structured and unstructured random effects. Thus I will be able to visualise the full range of residuals. The WinBUGS Code is below. # Spatial model EDS9 for map decomposition full GYM model { for (i in 1:N) { EDS9_OBSER[i] ~ dpois(mu[i]); log(mu[i]) 9

Map Decomposition I will use “map decomposition” to visualise the results of the final model as reported by Law and Haining (2004). This approach examines the results of the full posterior distribution and will allow high and low areas of relative risk to be identified, as well as an assessment of the contribution made by the covariates, unstructured and spatially structured random effects. Figure 169 shows the map of RR for EDS > 9 in the final BYM model. The clustering of EDS > 9 can be seen in the northern suburbs in a similar pattern to that observed previously (SMR -Figure 63, Smoothed RR – Figure 64 and SatSCAN Clusters – Figure 65). Figures 170 to 173 show the RR for areas in the north are strongly driven by the covariates “% No Support”, and Entropy Index with a smaller protective effect from the covariate “% living in an apartment. The covariate % smoking is making a small contribution to the RR in areas in the south of the map. The map of residuals (not shown) is strongly dominated by the spatial component exp(s[i]) as shown in Figure 174. The unexplained spatial residual is strongest in the southern suburbs. The implication is that covariates that would eliminate the spatial residual have not been included in the model and possibly have not been identified among the candidate variables.

558


Figure 169: Map of the Relative Risk of EDS > 9 – Final Model (BYM)

559


Figure 170: Contribution by covariate “%No Support” – EDS >9 Final Model (BYM)

560


Figure 171: Contribution by covariate Entropy Index – EDS > 9 Final Model (BYM)

561


Figure 172: Protective contribution by covariate “%Living in Apartment” – EDS > 9 (BYM)

562


Figure 173: Contribution by covariate “%Smoking” – EDS >9 Final Model (BYM)

563


Figure 174: Contribution by spatially unexplained components – EDS > 9 (BYM)

564


Figure 175: Composition by unstructured unexplained components – EDS >9 (BYM)

565


566

Appendix K: Ecological EFA

Ecological Exploratory Factor Analysis 7.4.7.1 Extraction and rotation method For theoretical EFA where there is expected to be correlation between factors the most appropriate method of extraction is “common factor analysis” as opposed to PCA with oblique rotation, rather than orthogonal rotation. Consequently I used Principal-axis factoring with oblimm oblique rotation. I also undertook compared the outcomes using PCA and orthogonal rotation. The extracted factors were in all cases the same. 7.4.7.2 Assumptions and Diagnostics

Correlation Matrix Examination of the initial correlation matrix revealed high correlations between variables related to poverty and community entropy/integration. For example the Index of Relative Social Deprivation was also highly correlated with “sole parenthood” and “poverty”. A number of variables were mutually exclusive and thus “different address five years previously” was highly correlated with “same address five years previously”. Variables were removed and the correlation matrix reanalysed.

Bartlett’s Test of Sphericity Bartlett’s test is used to evaluate whether a correlation matrix is suitable for factor analysis by testing the hypothesis that the matrix is an identity matrix (a matrix in which all coefficients not in the diagonal are zeros. A low p value indicates that the identify matrix is rejected, which it is in this case. Kaiser-Meyer-Olkin Measure of Sampling Adequacy. Bartlett's Test of Approx. Chi-Square Sphericity df Sig.

.801 2151.753 190 .000

Table 98: Kaiser-Meyer-Olkin and Bartlett’s Tests

Kaiser-Meyer-Olkin Measure The Kaiser-Meyer-Olkin measure (KMO) is based on the principal that if variables share common factors, then partial correlations between pairs of variables should be small when the effects of the other variables are controlled. The KMO varies between zero and one with a KMO of 0.801 being described as “meritorious” (p 324, Munro 2001).

567

Appendix K: Ecological EFA

Anti-image correlation The anti-image correlation matrix did not contain any high values and thus there was predominantly unique rather than common covariance. This supports the undertaking of the factor analysis.

Measure of sampling adequacy The diagonals in the anti-image matrix are measures of sampling adequacy (MSA). Variables with low MSA (