Contents - Office for National Statistics

5 downloads 3 Views 2MB Size Report
Mar 30, 2010 - Self-rated health and mortality in the UK: results from the first ... 2008-based national population projections for the United Kingdom and. 91 ...... 1,775. 1,774. 1. 0.08%. * Net movements of Armed Forces and other smaller ...
Population Trends 139 Spring 2010

Contents In brief

1

Population Trends: readers’ views invited; Quick online access to older Population Trends Articles; Publication of revised population estimates and subnational population projections; Reference Data Tables; Population projections for Scottish areas (2008-based); Data visualisation and demography – popularising population statistics

Features

The ONS Longitudinal Study – a prestigious past and a bright future Shayla Goldring and Jim Newman

4

Self-rated health and mortality in the UK: results from the first comparative analysis of the England and Wales, Scotland and Northern Ireland Longitudinal Studies Harriet Young, Emily Grundy, Dermot O’Reilly, Paul Boyle

11

Do partnerships last? Comparing marriage and cohabitation using longitudinal census data Ben Wilson, Rachel Stuchbury

37

Households and families: Implications of changing census definitions for analyses using the ONS Longitudinal Study Emily Grundy, Rachel Stuchbury, Harriet Young

64

Ten year transitions in children’s experience of living in a workless household: variations by ethnic group Lucinda Platt

70

2008-based national population projections for the United Kingdom and constituent countries Emma Wright

91

This issue is available from 25 March 2010 at: www.statistics.gov.uk/populationtrends

Office for National Statistics

Population Trends

Spring 2010

In brief Population Trends: readers’ views invited At ONS we continually strive to maintain the quality of Population Trends as an important demographic journal. The views of our readership are important to us and we would welcome any comments and suggestions you have about the future scope and direction of the journal to ensure it remains fresh and pertinent while maintaining the high standards expected by our readership. Please email your comments and suggestions to: [email protected] Readers are also reminded that we always welcome submission of papers from external colleagues that are appropriate to the scope of the journal.

Quick online access to older Population Trends articles Readers interested in locating previously published articles may like to know that a searchable database is available. This online article search facility covers all ONS journals. To find an article it is possible to do a text search for keywords, journal title, article title, author’s name, issue number, and publication year. All articles published in Population Trends since Winter 1997 (issue no. 90) are available online. Using this free search facility, pdf files can be downloaded for each article. To use the service, go to: www.statistics.gov.uk/cci/articlesearch.asp

Publication of revised population estimates and subnational population projections On 30 March 2010 the ONS Centre for Demography (ONSCD) will be publishing a report summarising the outcome of methodological work undertaken since the release of indicative impacts to changes to the mid-2002 to mid-2008 local authority population estimates on 30 November 2009 and the responses received from users on the impact of the package of improvements that will be implemented in May 2010 when revised estimates are published. More detailed information, including data by age and sex, together with detailed papers on the revisions, will be published on 27 May 2010. Also on 27 May 2010, ONSCD will be publishing 2008-based Subnational Population Projections for local authorities in England, and the Welsh Assembly Government will be publishing 2008-based Subnational Population Projections for local authorities in Wales.

Reference Data Tables Population Trends and Health Statistics Quarterly have been developed as online publications. As part of the ‘web-only’ publication approach the content and format of all the reference data tables within these publications is being reviewed. To help ONS determine user requirements, proposed changes are outlined in a consultation document, which is available at: www.statistics.gov.uk/STATBASE/Product.asp?vlnk=15354 Comments from users are welcomed. Please email your responses and suggestions to: [email protected] Office for National Statistics

1

Population Trends

Spring 2010

Population projections for Scottish areas (2008-based) On 3 February 2010 the General Register Office for Scotland published its 2008-Based Population Projections for Scottish Areas. The report covers the period from 2008 to 2033, and the key points are: • Scotland’s population is expected to increase over the next 25 years, although this rise is projected to be unevenly spread across the country. • The population of 19 of the 32 council areas in Scotland is projected to increase, while the population in the other 13 is projected to decrease. The council areas with the greatest projected increase in population are East Lothian (+33 per cent) and Perth & Kinross (+27 per cent). Inverclyde (–18 per cent) and East Dunbartonshire (–13 per cent) have the largest projected decreases. • Every council area is projected to have more elderly people than today, though the scale of the increase will vary. • The number of children aged 0–15 is projected to decrease in 20 of the 32 council areas, with the largest percentage decreases in Shetland (–33 per cent) and Inverclyde (–29 per cent). The biggest increases are projected in East Lothian (+38 per cent) and Perth & Kinross (+24 per cent). • The population of working age (accounting for future changes to the state pension age) is projected to increase in 15 council areas and decrease in 17 – increasing the most in East Lothian (+29 per cent) and decreasing the most in Inverclyde (–26 per cent). • The population of pensionable age (accounting for future changes to the state pension age) is projected to increase in all council areas, the largest increases being projected in Aberdeenshire (+65 per cent) and West Lothian (+59 per cent), with the smallest increase in Dundee City (+8 per cent) and Glasgow City (+11 per cent). • It is hard to predict how many people might migrate to Scotland. The high migration projection shows what would happen if Scotland were to gain larger numbers through migration than expected. The populations in 26 councils would rise under this variant. The greatest increase is again projected in East Lothian (+38 per cent) and Perth & Kinross (+37 per cent) and the largest decrease again in Inverclyde (–14 per cent) and East Dunbartonshire (–11 per cent). • The low migration projection shows the population if Scotland were to gain smaller numbers through migration than expected. Populations are expected to rise in 15 councils under this variant. The greatest increase is again projected in East Lothian (+29 per cent) and Perth & Kinross (+25 per cent) and the largest decrease again in Inverclyde (–20 per cent) and East Dunbartonshire (–17 per cent). Further details can be found at: www.gro-scotland.gov.uk/statistics/publications-and-data/popproj/2008-based-pop-proj-scottishareas/index.html

Office for National Statistics

2

Population Trends

Spring 2010

Data visualisation and demography – popularising population statistics Population trends have long benefited from graphical presentation techniques – for example William Playfair, the father of quantitative data presentation, produced early population diagrams in the late eighteenth century. However, traditional data graphics usually reserve one whole dimension (typically the x axis) of a 2-d static graph to reveal change over time. Modern web technologies allow us to challenge convention, using richer means of graphical presentation to produce more engaging representations of change over time. The Data Visualisation Centre at ONS works closely with the ONS Centre for Demography to develop a range of animated and interactive data graphics aimed at revealing structural changes in the UK population over extended periods. These include an animated local authority map representing changes in subnational age structure in the UK from 1992 to 2031. This is a visual representation of work first reported on in Population Trends in June 20091. The classic demographic display –the population pyramid– has also received fresh treatment. For example, a new animated edition allows users to explore combined population estimates and projections from 1961 (England and Wales) or 1971 (UK) through to 2083. Users can click and drag across age bands on the graphic to define their own summary statistics on the fly. A further innovation is the twin-pyramid display, initially published for National Population Projections. This display allows users to visualise ONS’ variant population projections in parallel, visualising change over time, and dynamically overlaying the two images to provide easy comparisons of structural differences between the projections. The new graphics are all published in Adobe Flash format and allow the user to control not only the animation but interact with the graphical content to query the underlying data (which can be downloaded separately). They have ‘full screen’ functionality, making them ideal for lectures and presentations. The visualisations have all been designed as templates which can be reused and extended where appropriate. Further work is planned this year, refining and adding functionality to the existing visualisations, and reusing them, where appropriate, with other datasets. This approach reflects and reinforces the move of ONS publications away from print to online formats. It is anticipated that these new graphics will be the first of a new generation of data graphics optimised for web presentation.

Links Animated map of ageing, available at: www. statistics.gov.uk/ageingintheuk/default.htm Animated national population pyramid, available at: www.statistics.gov.uk/populationestimates/ flash_pyramid/default.htm Animated twin population projections, available at: www.statistics.gov.uk/ national projections/ flash_pyramid/projections.html

Reference 1 Blake, S. (2009) ‘Subnational patterns of population ageing’. Population Trends 136. Available at: www.statistics.gov.uk/articles/ population_trends/PT136SubnationalAgeing.pdf

Office for National Statistics

3

Population Trends 139

Spring 2010

The ONS Longitudinal Study – a prestigious past and a bright future Shayla Goldring and Jim Newman Office for National Statistics This issue of Population Trends includes a number of articles and reports resulting from research based on the ONS Longitudinal Study (ONS LS). They have been drawn together in one issue to highlight the value of this type of study for demographic research. 2009 marked the 35th anniversary of the establishment of the ONS LS. The study now contains data from the last four censuses (1971 to 2001), linked to vital events data since 1971, for a sample of one per cent of the population of England and Wales. More recently, sister studies have been established in Scotland and Northern Ireland. The Scottish Longitudinal Study (SLS) started with 1991 Census data and the Northern Ireland Longitudinal Study (NILS) started with 2001 Census data. The lead article in this issue comes from an exemplar project that was established to explore how to utilise the three studies to carry out UK-wide longitudinal analysis. Two different methods were used to analyse socio-economic and country level differences in health and mortality across the studies. The article summarises the results of this analysis, reports on the relative strengths of the different methods used, and draws attention to a number of new resources that have been developed by the project researchers as aids to using all three studies. This is an excellent example of collaborative working across the UK, involving researchers from the Centre for Longitudinal Study Information and User Support (CeLSIUS) at the London School of Hygiene and Tropical Medicine, the Longitudinal Studies Centre – Scotland (LSCS) at the University of St. Andrews and the Northern Ireland Longitudinal Study – Research Support Unit (NILS-RSU) at Queen’s University Belfast. The project also involved collaboration between ONS, the General Register Office for Scotland (GROS) and the Northern Ireland Statistics and Research Agency (NISRA) to ensure the secure transfer and handling of data from the three studies so that it could be brought together in one place for analysis. The other ONS LS-based articles and reports in this edition largely focus on research into issues related to families and households, as summarised below: • A collaborative project involving Ben Wilson (ONS) and Rachel Stuchbury (CeLSIUS) comparing the stability of partnerships involving marriage and cohabitation. • A project looking at transitions in children’s experience of living in a workless household and how this varies by ethnic group, submitted by Lucinda Platt (Institute for Social & Economic Research, University of Essex).

Office for National Statistics

4

Population Trends 139

Spring 2010

• An article on the effect of a change in the census definition of a child between 1991 and 2001 submitted by Emily Grundy, Rachel Stuchbury and Harriet Young (CeLSIUS). The remainder of this introductory article will focus on the ONS LS, its history and some examples of its use, and gives a summary of planned developments over the coming years. Please refer to the contact details at the end of the article if you require further information on any of the three longitudinal studies.

Contents Introduction to the ONS LS................................................................................................................ 6 Research using the ONS LS.............................................................................................................. 6 Future plans for the ONS LS.............................................................................................................. 8 References....................................................................................................................................... 10

Office for National Statistics

5

Population Trends 139

Spring 2010

Introduction to the ONS LS Longitudinal data sets are based on repeated measurements of a sample population. They allow us to answer questions about how a particular cohort of people changes over time, and to explore reasons for change. In addition, the ONS LS allows users to look at and compare the experiences of different cohorts at different points in time. This enables users to separate age, period and cohort effects in their analysis. Cohort analysis was identified as a priority in developing the analysis of mortality by William Farr, a noted epidemiologist who was appointed the first ‘Compiler of Abstracts’ at the newly established General Register Office for England and Wales in 1839. At this time the analysis of information collected by statistical offices was mainly cross-sectional as a result of the limited data available. Farr was the first to combine information from a national census (1861) and the death registers to look at the occupation of men, their age at and cause of death. The ONS LS was established in 1974 by taking a sample of records from the 1971 Census for England and Wales of all those born on one of four dates of birth. This original sample has been continuously augmented since 1971 with new members. New members enter the study through one of the following three routes if born on one of the four dates of birth: • completion of a census form • birth registration through the civil registration service; or • registration as a patient with a doctor. Information from the 1971, 1981, 1991 and 2001 Censuses has been linked, along with information on events such as births, deaths, immigration, emigration and cancer registrations for study members. More than half a million study members have been identified at each of the four censuses, and the study now includes information on more than one million different individuals. The ONS LS is a study – not a survey. Its strength lies in the re-use of data that have already been collected for other purposes, significantly reducing the effects typically associated with respondent burden. As a result, both retention and response rates are relatively high.

Research using the ONS LS The study was originally set up primarily to meet the need for better data on occupational mortality and fertility patterns. Data at the time were inadequate for the study of occupational mortality rates. In order to provide evidence for a causal relationship between occupation and mortality, information on occupation is needed for a period well before the onset of illness and death. In addition, information on an individual’s characteristics such as employment status, area of residence, qualifications and general health would be needed for some years before death to use as control variables, as these may also have an influence on mortality. It was also accepted that there was a need for more detailed information on fertility patterns, in particular changes in the spacing of births, and the part that social and economic characteristics play in family formation. The ONS LS addresses these needs, and many more, by linking existing census and vital event data. The strengths of the study include: Office for National Statistics

6

Population Trends 139

Spring 2010

• the robustness of the sample size, around 500,000 at any one census • the relatively high rates of retention and response • the range and stability of the information available for analysis over time, from censuses and vital events • the inclusion of census information on co-residents of study members • the availability of sister studies in Scotland and Northern Ireland for those interested in a UKwide perspective; and • the services of dedicated user support teams (contact details for these teams are provided at the end of this article). The ONS LS enables analysis of a wide range of key sub-groups and topics of policy interest. Since its inception, the study has been used to address research questions including studies of social mobility, ageing and migration. Studies that make the fullest use of the data link social, occupational and demographic information at successive censuses to data about fertility, mortality, and cancer incidence and survival. Some examples of recent research that used the ONS LS follow below. 1

In the field of fertility, the study has recently been used to explore lifelong childlessness , a topic which has received little attention given the decline in fertility experienced since the baby boom. This research investigated the degree to which socio-economic characteristics of women and, where present, their partners were related to female lifelong childlessness. The study measured the extent to which women who remained childless throughout their life course were distinctive from those who became mothers, and therefore improved our understanding of childlessness among women in England and Wales. The researchers on this project, Martina Portanti and Simon Whitworth from ONS, won the inaugural Neville Butler Memorial Prize in 2009, awarded by the Economic and Social Research Council for excellence in the analysis of longitudinal data. It received a great deal of media attention, as demonstrated by the following newspaper headlines:

“One in five women stays childless because of modern lifestyle”, Daily Telegraph “Fifth of women childless as careers take precedence, study shows”, The Times

Recent work in the field of mortality includes a project led by Dr David Pevalin from the School of Health and Human Sciences at Essex University. Dr Pevalin’s research analysed social inequalities in avoidable mortality, looking to test empirically the theory of social conditions as fundamental causes of disease. Findings from this project were presented at the 2009 conference of the British Society for Population Studies. The information available on co-residents from the census, and also vital events such as the registration of the death of a spouse, are very important for analyses of partnership formation and dissolution. This information means that it is possible to look at the characteristics of both partners in a relationship and use them or the differences between them to study the factors relating to the stability of different partnership types over time. This is the approach used by Ben Wilson and Rachel Stuchbury in their paper ‘Do partnerships last? Comparing marriage and cohabitation using the ONS Longitudinal Study’, included in this issue of Population Trends.

Office for National Statistics

7

Population Trends 139

Spring 2010

The study also has internal uses within ONS, such as quality assurance of other data sources. It played a key role in the assessment and adjustment of population estimates based on the 2001 Census. Analysis using the study highlighted a shortfall in men aged 25 to 34 in the Census. The 2001 mid-year population estimates for 68 local authorities were adjusted as a result of this 2 analysis. For more information on this analysis, refer to Section 7 of Series LS no.10 and the 3 Census 2001 Quality Report for England and Wales . Planning for the 2011 Census reflects the importance of the part the study played after 2001. This time the ONS LS will be a key source of data used in carrying out quality assurance of census data. This will allow information from the study to be considered alongside other quality assurance material before any population estimates are published.

Future plans for the ONS LS Linking data from the 2011 Census With the 2011 Census only a year away, plans for incorporating the next set of census data into the ONS LS are well advanced. As a result of this work, the study will contain linked data from five successive censuses. The new census data will be available from the study in autumn 2013. The addition of 2011 Census data will enable users to study transitions in people’s caring responsibilities for the first time. This isn’t currently possible as the census question on caring was first asked in 2001. Many users will also want to update previous analyses such as • socio-economic and/or ethnic differences in mortality, life expectancy, cancer incidence, fertility and migration behaviour; and • transitions over time in topic areas such as occupational and social mobility, household composition and partnerships.

UK-wide longitudinal study research As noted earlier, the establishment of longitudinal studies in Scotland and Northern Ireland means that a longitudinal study infrastructure now exists that enables researchers to take a UK-wide view, or to draw comparisons between different regions and countries across the whole of the UK. While there are a number of differences in the structure, content and operation of the three studies, the basic principle behind each of them is essentially the same. That is, to link the wide range of information collected at each census with data from subsequent events, most notably those relating to fertility and mortality. The exemplar project reported in this issue of Population Trends has involved both researchers and the statistical offices working through a number of issues that required resolution to allow this research to take place. These are reported on in a technical working paper on the CeLSIUS 4 website . One significant outcome of this work is that a provisional working model has been established for anyone wishing to conduct research across the three studies. As part of a wider review of user needs of the ONS LS, ONS will be gauging the demand for a more permanent solution that allows UK-wide research to take place. Further work on this will be led by ONS and will, of course, involve very close collaboration with colleagues at GROS and NISRA. In the meantime, any researchers interested in exploring this as an option should contact their nearest user support team (contact Office for National Statistics

8

Population Trends 139

Spring 2010

details for these teams are provided at the end of this article). Any proposed projects will be considered by all three statistical offices on a case-by-case basis.

Linking additional administrative data The richness of information in the ONS LS comes from matching census data with administrative data over time. These administrative data are currently limited to birth and death related data from the civil registration service, cancer data from the cancer registries and demographic data from the patient registration system. The last Longitudinal Study review in 1998 recommended that the linkage of administrative data held by other government departments should be considered. The passing of the Statistics and Registration Service Act 2007, which came into effect in April 2008, provides the legal framework through which ONS can seek access to any data held by other government departments. ONS and the Department for Work and Pensions (DWP) are currently working closely to make a case for linking unemployment related benefit data to the study. This is the first attempt to use the new legislation to extend the content of the ONS LS. It is intended that this will be the first of a number of new linkages that will enrich the data available through the study. The aim of linking additional data is to extend the range of research topics that can be explored through the study. This will enhance the value of the study for existing users, as well as reach out to new users carrying out research in areas that the study cannot currently address. Users will be consulted to identify which additional data are most in demand. This will form part of a wider user review that will allow ONS to prioritise this alongside other development activity.

Using the ONS LS ONS actively promotes use of the ONS LS while maintaining the confidentiality of the individuals in the sample. ONS LS records available for analysis are anonymised but the database contains individual-level data that have not been aggregated or disguised. To ensure confidentiality, these microdata can only be accessed at ONS sites and can only be accessed from a secure area known as the Virtual Microdata Laboratory (VML). Support officers are available to help users extract and use the data. For further information, or for an informal discussion about using the ONS LS, government and other non-academic users should contact the Microdata Analysis and User Support team at ONS. Tel: 01633 455844 email: [email protected] Website: www.ons.gov.uk/about/who-we-are/our-services/longitudinal-study Academic users should contact the CeLSIUS team at the London School of Hygiene and Tropical Medicine. Tel: 020 7299 4634 email: celsius[email protected] Website: www.celsius.lshtm.ac.uk

Office for National Statistics

9

Population Trends 139

Spring 2010

For further information about the SLS, users should contact the Longitudinal Studies Centre – Scotland at the University of St. Andrews. Tel: 01334 463992 email: [email protected] Website: www.lscs.ac.uk/sls/ For further information about the NILS, users should contact the Northern Ireland Longitudinal Study – Research Support Unit (NILS-RSU) at Queen’s University Belfast. Tel: 028 9082 8210 or 028 9034 8199 email: [email protected] Website: www.qub.ac.uk/research-centres/NILSResearchSupportUnit/

References 1 Portanti, M and Whitworth, S (2009) A comparison of the characteristics of childless women and mothers in the ONS Longitudinal Study, Population Trends 136, Summer 2009, pp 10–20. Available at: www.statistics.gov.uk/downloads/theme_population/Popular-Trends136.pdf 2 Blackwell, L, Lynch, K, Smith, J and Goldblatt, P (2003) Longitudinal Study 1971–2001: Completeness of Census Linkage, Series LS no. 10, September 2003. Available at: www.statistics.gov.uk/downloads/theme_population/LS_no10.pdf 3 Census 2001: Quality Report for England and Wales, 2005. Available at: www.statistics.gov.uk/StatBase/Product.asp?vlnk=14212 4 Young, H (2009) Technical Working Paper: Guide to parallel and combined analysis of the ONS LS, SLS and NILS, July 2009. Available at: www.celsius.lshtm.ac.uk/UKLS/Guide%20 to%20parallel%20and%20combined%20LS%20analysis.doc

Office for National Statistics

10

Population Trends 139

Spring 2010

Self-rated health and mortality in the UK: results from the first comparative analysis of the England and Wales, Scotland, and Northern Ireland Longitudinal Studies Harriet Young, Emily Grundy London School of Hygiene & Tropical Medicine Dermot O’Reilly Queen’s University Belfast Paul Boyle University of St Andrews Previous studies have shown that self-reported health indicators are predictive of subsequent mortaity, but that this association varies between populations and population sub-groups. For example, self-reported health is less predictive of mortality at older ages, has a stronger association with mortality for men than for women and is more predictive of mortality for those of lower than those of higher socio-economic status, particularly among middle aged working adults This article explores this association using individual level, rather than ecological, data to see whether there are differences between the constituent countries of the UK in the relationship between self-reported health and subsequent mortality, and to investigate socio-economic inequalities in mortality more generally. Data are used from the three Census based longitudinal studies now available for England and Wales, Scotland and Northern Ireland.

Office for National Statistics

11

Population Trends 139

Spring 2010

Contents Introduction....................................................................................................................................... 14 Previous research on associations between self-reported health and mortality.............................. 14 Methods............................................................................................................................................ 15 Descriptive results............................................................................................................................ 19 Multivariate results........................................................................................................................... 20 Mortality............................................................................................................................................ 22 Summary of results.......................................................................................................................... 23 Discussion........................................................................................................................................ 24 Strengths and weaknesses of each analysis strategy...................................................................... 24 Acknowledgements.......................................................................................................................... 28 References....................................................................................................................................... 36

List of figures Figure 1 Percentage of the population aged 35–74 with fair or poor self-rated health by age group, gender and country, ONS LS, SLS, NILS, 2001....................................... 18 Figure 2 Mortality rate by gender and country for those aged 35–49, ONS LS, SLS, NILS, 2001.................................................................................................................. 20 Figure 3 Mortality rate by gender and country for those aged 50–64, ONS LS, SLS, NILS, 2001.................................................................................................................. 20 Figure 4 Mortality rate by gender and country for those aged 65–74, ONS LS, SLS, NILS, 2001.................................................................................................................. 21 Figure 5 Mortality rate by gender and country for those aged 35–74, ONS LS, SLS, NILS, 2001.................................................................................................................. 21

Office for National Statistics

12

Population Trends 139

Spring 2010

List of tables Table 1 Variables and categories used in individual level and aggregated datasets, ONS LS, SLS, NILS 2001................................................................................................... 16 Table 2 Socio-demographic and socio-economic characteristics of the population aged 35–74 in England and Wales, Scotland and Northern Ireland, ONS LS, SLS, NILS 2001................................................................................................................... 17 Table 3 Odds Ratios from logistic regression analysis of variations in the proportion of the population aged 35–74 with poor or fair self-rated health in 2001 by socio-demographic and socio-economic characteristics in England and Wales, Scotland and Northern Ireland. ONS LS, SLS, NILS 2001, using parallel datasets... 25 Table 4 Odds Ratios from logistic regression analysis of variations in the proportion of the population aged 35–74 with poor or fair self-rated health in 2001 by socio‑demographic and socio-economic characteristics in England and Wales, Scotland and Northern Ireland and for all countries combined. ONS LS, SLS, NILS 2001 using combined aggregated datasets....................................................... 26 Table 5 Rate ratios of mortality for the population aged 35–74 by socio-demographic and socio‑economic characteristics and health status in England and Wales, Scotland and Northern Ireland. ONS LS, SLS, NILS 2001 using parallel datasets.... 29 Table 6 Rate ratios of mortality for the population aged 35–74 by socio-demographic and socio-economic characteristics and health status in England and Wales, Scotland and Northern Ireland, and for all countries combined. ONS LS, SLS, NILS 2001 using combined aggregated datasets....................................................... 32

Office for National Statistics

13

Population Trends 139

Spring 2010

Introduction There are now three census based record linkage studies covering all constituent parts of the UK. The oldest of these, the Office for National Statistics Longitudinal Study (ONS LS) which covers England and Wales, was established in the mid 1970s and includes individual level information from the 1971, 1981, 1991 and 2001 Censuses. The Northern Ireland Longitudinal Study (NILS) and the Scottish Longitudinal Study (SLS) were launched in 2006 and 2007 respectively. The SLS includes information from the 1991 and 2001 Censuses and NILS data from the 2001 Census. All three studies include linked data from vital registration systems, including mortality. This means that for the first time there is the potential to analyse differentials between the constituent elements of the UK, using information from large representative longitudinal studies including individual level information from both census and vital registration sources. All three sources are subject to stringent disclosure control safeguards and it is currently not possible to combine individual level data from them to create a UK dataset. However, comparative analysis may be carried out in two ways: firstly, by conducting separate parallel individual level analyses of the three studies and comparing results; and secondly, by appending datasets of aggregated counts of individual level data from each study and then analysing this combined dataset. In this paper we show results from using both methods to analyse socio-economic and country level differences in health and mortality. This is an important topic because of research and policy interest in health inequalities in the UK, and indications from previous research using ecological data that patterns of reporting health may differ between the constituent countries of the UK.1 We examine the strengths and weaknesses of each method for addressing this question and discuss the issues involved in working with the three datasets together.

Previous research on associations between self-reported health and mortality Previous studies have shown that self-reported health indicators are predictive of subsequent mortality,2,3 but that this association varies between populations and population sub-groups. For example, self-reported health is less predictive of mortality at older ages;4 has a stronger association with mortality for men than for women;2 and is more predictive of mortality for those of lower than those of higher socio-economic status, particularly among middle aged working adults.4 Variations in reporting of self-rated health over time,5 and by geographic region,6,7,8 including by constituent country of the UK, have also been reported. Analysis of ecological associations using area level data has shown that for a given level of health, mortality rates are higher in Scotland than in Northern Ireland or Wales, an association that persists after control for socio-economic status.1 Thus the Scottish population has the highest mortality rates of the constituent countries of the UK, England the lowest, with Northern Ireland and Wales in between. However, on the evidence of self-reported health data, the population of Northern Ireland is less healthy than that of Scotland.1,9 In this study, we are able to explore this association using individual level, rather than ecological, data to see whether there are differences between the constituent countries of the UK in the relationship between self-reported health and subsequent mortality, and to investigate socioeconomic inequalities in mortality more generally.

Office for National Statistics

14

Population Trends 139

Spring 2010

Methods Data We use data from the three census based longitudinal studies now available for England and Wales, Scotland, and Northern Ireland. The ONS LS is a record linkage study of approximately one per cent of the population of England and Wales enumerated at the 1971 Census (some 500,000 people); sample members were selected on the basis of four birthdays in the year. Record linkage has been used to add information from subsequent censuses (1981, 1991, 2001) and data from vital registration sources including births, to sample mothers and deaths of sample members and their spouses.10 While losing emigrants and the deceased, the sample has been maintained by the recruitment of new births and immigrants born on LS birthdays and so remains nationally representative. The SLS is a 5.3 per cent representative sample of the Scottish population based on 20 birthdays in the year. A sample of approximately 265,000 SLS members was identified from the 1991 Census, with information linked in from the 2001 Census and other sources, including vital events, cancer registrations and hospital episodes.11 The NILS is also modelled on the ONS LS and includes approximately 500,000 sample members (around 28 per cent of the population of Northern Ireland). As with the ONS LS and the SLS, the sample is maintained by recruitment of new births and immigrants born on the 104 NILS birthdays. The NILS sample differs slightly from the ONS LS and SLS in that the initial sample was drawn from the Health Card Registration System and then linked to the census, whereas in the other two studies the initial sample was drawn from the census. Northern Ireland has a second census-based dataset that links the 2001 Census returns for the entire enumerated population to subsequently registered mortality data. However, the smaller NILS dataset was used for this study to maximise comparability with the other UK longitudinal studies. All three studies have associated user support services, which facilitate use of the data for authorised researchers subject to disclosure control procedures. Further details of the data sets and these support services are available elsewhere.12 Access to anonymised individual level data is only possible in the respective statistical office safe setting (ONS for the ONS LS, The General Register Office for Scotland for the SLS, and the Northern Ireland Statistics and Research Agency for the NILS).

Dataset development This study is based on analyses of those aged 35–74 at the 2001 Census and their mortality from the time of the 2001 Census until 30 June 2006. This age range was chosen because in younger groups levels of poor health and rates of mortality are very low, and in age groups 75 and over fewer indicators of socio-economic status are available in the data sets. We excluded those living in communal establishments, students not at their term time address and those lacking information on self-rated health or marital status in the 2001 Census. Proportions excluded because of non‑response to these questions in the census accounted for 1.3 per cent of the ONS LS sample, 1.4 per cent of the SLS sample and 3.2 per cent of the NILS sample. We created datasets for both individual level and aggregated analyses. For the individual level analysis, we constructed equivalent separate datasets for the ONS LS, SLS and NILS.

Office for National Statistics

15

Population Trends 139

Spring 2010

Table 1 Variables and categories used in individual level and aggregated datasets, ONS LS, SLS, NILS 2001 Variable

Variable categories Individual level datasets

Aggregated datasets

Self rated health

Good Fairly good or not good

Good Fairly good or not good

Gender

Male Female

Male Female

Age/Age group

Age–single years

35–49 50–64 65–74

Marital status

Married Separated or divorced Widowed Never married Upper secondary or degree Lower secondary None Other* (ONS LS only) Missing Manager or professional Intermediate ** Lower *** Never worked, unemployed, student, other Missing Owner occupier Social rental Private rental or other Missing Yes No Missing

Married Not married

Highest educational qualification

NS-SEC

Housing tenure

Car access









Socio-economic status score ****



0 (Highest) 1 2 3 4 5 (Lowest) Missing data

Country



England and Wales Scotland Northern Ireland

Notes: * This category includes City and Guilds, RSA/OCR and BTEC/Edexcel qualifications which cover qualifications from entry to degree level. ** This group includes intermediate occupations, small employers and own account workers. *** This group includes lower supervisory, technical, semi-routine and routine occupations.

For the aggregated analysis, we created aggregated count datasets for each LS and then combined them. In aggregated datasets such as these, cells comprise counts of individuals with a particular set of characteristics, (for example, being female, living in owner occupied housing and aged 35–49), rather than individuals themselves. Disclosure control guidelines meant that Office for National Statistics

16

Population Trends 139

Table 2

Spring 2010

Socio-demographic and socio-economic characteristics of the population aged 35–74 in England and Wales, Scotland and Northern Ireland, ONS LS, SLS, NILS 2001 LS Sample

Variable

Categories

Age (years)

Mean

ONS LS

SLS

NILS

52.2

52.1

51.7

0.022

0.032

0.025

35–49

44.5

45.3

47.0

50–64

37.4

36.8

36.2

65–74

18.0

17.9

16.8

Men

48.8

47.9

48.3

Women

51.2

52.1

51.7

Married

69.3

68.1

71.5

Separated or divorced

14.1

13.9

10.4

5.9

6.9

6.6

Never married

10.7

11.0

11.5

Highest educational qualification (per cent)

Upper secondary or degree

22.5

35.2

18.1

Lower secondary None Other Missing

28.5 34.9 8.7 5.5

19.6 40.1 – 5.2

23.7 51.0 – 7.3

NS-SEC (per cent)

Manager or professional

30.0

28.9

25.2

Intermediate occupations, small employers and own account

19.8

18.9

19.8

Lower supervisory, technical, semi-routine and routine

33.5

39.7

35.1

3.8

3.4

5.4

Missing

12.9

9.2

14.6

Owner

78.4

72.3

78.2

Social housing tenant

13.3

20.7

14.2

Private housing tenant and other

5.9

5.1

4.0

Missing

2.4

1.9

3.6

Car

83.9

77.5

82.7

No car

14.5

21.2

14.8

Missing

1.6

1.2

2.5

Standard error Age group (per cent)

Gender (per cent)

Marital status (per cent)

Widowed

Never worked, unemployed, student, other

Housing tenure (per cent)

Car access (per cent)

Socio-economic score

Mean (excluding those with missing values)

2.4

2.5

2.7

0.004

0.006

0.004

0 – Least disadvantaged

13.6

18.5

11.5

1

13.4

10.7

9.7

2

14.7

12.8

12.3

3

16.5

13.6

14.9

4

15.4

16.0

19.2

9.3

15.6

12.0

17.2

12.8

20.5

100

100

100

254,918

122,753

192,251

Standard error Socio-economic score (per cent)

5 – Most disadvantaged Missing Total (per cent) Number in analysis

 

Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

17

Population Trends 139

Spring 2010

cell counts of less than three were not permissible.13 For this reason in the aggregated analysis we used age groups rather than single year of age, combined marital status categories and created a socio-economic score derived from several variables rather than using each variable separately. This score was derived from separate indicators as follows: car access (0), no car access (1); home owner (0), private or social housing tenant (1); highest educational qualification upper secondary or degree (0), lower secondary or other (1), none (2); manager or professional (0), intermediate occupations (1), lower occupations, never worked, unemployed and students (2). Higher scores thus indicate a greater level of disadvantage. The main advantage of using the aggregated data set was that we could also include a variable indicating country (England and Wales, Scotland, or Northern Ireland) and compare effects across these directly.

Variables used in the analysis In all analyses we dichotomised self-rated health into a variable, distinguishing those who reported good health from those reporting ‘fairly good’ (termed ‘fair’ in some of the text below) or ‘not good’ health (hereafter referred to as ‘poor’ health). Mortality was measured from the census date, 29 April 2001, until 30 June 2006, the latest date that mortality data was available in all three data sources, giving five years and two months of follow-up. Table 1 shows the variable categories used in the individual and aggregated analysis. Demographic variables comprised single year of age, or age group, gender and marital status. Indicators of socio-economic status included individual-level highest educational qualification and National Statistics Socio-Economic Classification (NS SEC), derived from information on occupation and employment status, and two household-level variables, housing tenure and household access to one or more cars or vans. Variables and categories of variables were identical

Figure 1

Percentage of the population aged 35–74 with fair or poor self-rated health by age group, gender and country, ONS LS, SLS, NILS, 2001

70 England and Wales Scotland Northern Ireland

60

50

40

30

20

10

0 35–49

50–64

65–74

Men

35–74

35–49

50–64

65–74

35–74

Women

Source: ONS LS, SLS, NILS 2001

Office for National Statistics

18

Population Trends 139

Spring 2010

in all three sources with the exception of highest educational qualification. The ONS LS education variable included an additional category of ‘other’ which the SLS and NILS did not have.

Statistical methods We undertook preliminary descriptive analyses of the three samples using the individual level datasets. We used multivariate logistic regression to analyse differentials in self-reported health by socio-demographic characteristics using the individual level datasets, and by socio-demographic characteristics and country using the aggregated dataset. In the latter analysis we also present results for each country separately, in order to allow comparison between the two methods. Survival analysis, using Poisson regression, was undertaken to investigate associations between self-rated health and socio-demographic characteristics with subsequent mortality. Known emigrants were excluded from date of leaving the respective study. In both analyses of self-rated health and mortality we present results from models controlling for age and sex (Model 1), and results from models additionally controlling for socio-demographic characteristics (Model 2). In the aggregated analysis, country was included in both models. In the mortality analysis we also show results from a third model including self-rated health. All analysis was carried out in the statistical office safe settings and produced in accordance with disclosure control guidelines.

Descriptive results Socio-demographic sample characteristics were broadly similar for England and Wales, Scotland and Northern Ireland (Table 2). The samples were similar in age and gender distribution, except that the Northern Ireland sample was slightly younger and included slightly more married and fewer divorced members. Differences between the three study samples in the distribution of sample members by educational level reflect both the separate identification of those with ‘other’ qualifications in England and Wales, and the different educational system in Scotland. Scotland had the highest proportion in the highest education category at 39 per cent, compared with 25 per cent in England and Wales, and 21 per cent in Northern Ireland. In the Northern Ireland sample, 51 per cent had none of the educational qualifications asked about, compared with 40 per cent of the Scottish sample, and 35 per cent of those in England and Wales. The Northern Ireland sample also included a slightly lower proportion in managerial and professional occupations and a slightly higher proportion in the category of never worked, unemployed, students or other. The proportion in lower supervisory, technical, semi-routine or routine occupations was largest in Scotland. In England and Wales, and Northern Ireland, 78 per cent of the sample were owner-occupiers compared with 72 per cent in Scotland, where a larger proportion lived in social housing. Those in Scotland were also slightly less likely to have access to a car or van. For the socio-economic score, used in the aggregated dataset analysis, the NILS sample had the highest proportion with missing values at 20 per cent, compared with 17 per cent in England and Wales and 13 per cent in Scotland (this illustrates one of the main disadvantages of using summary scores such as this – the high proportion with missing values on at least one of the variables used to construct it). The mean socio-economic score was lowest (representing a lower mean level of disadvantage) in England and Wales at 2.4, and highest in Northern Ireland with a score of 2.7. Scotland had the highest proportion of the sample in both the least and most disadvantaged categories. Figure 1 shows the proportions with fairly good or not good self-rated health by gender, age group and country. These proportions were higher among women than men and higher in Northern Ireland than in Scotland or England and Wales. Office for National Statistics

19

Population Trends 139

Figure 2

Spring 2010

Mortality rate by gender and country for those aged 35–49, ONS LS, SLS, NILS, 2001

Rate per 1000 2.5 2 1.5 1 0.5 0 England and Wales

Figure 3

Scotland

N. Ireland

Male

England and Wales

Scotland

N. Ireland

Female

Mortality rate by gender and country for those aged 50–64, ONS LS, SLS, NILS, 2001

Rate per 1000 12 10 8 6 4 2 0 England and Wales

Scotland

Male

N. Ireland

England and Wales

Scotland

N. Ireland

Female

Multivariate results Self-rated health Table 3 and Table 4 show results from logistic regression analysis of differentials in the proportions reporting not good or fairly good self-rated health. In both individual level (Table 3) and aggregated analysis (Table 4), the odds of poorer self-rated health increased with age, and were significantly higher for women than men, although the gender difference was smaller once marital status and socio-economic status were controlled (Model 2). Inclusion of single year of age in the individual level models was a better control than in aggregated models which only included three age groups, as confirmed by a comparison of r-squared values for Model 1 individual level versus aggregated dataset analysis (r = 0.042 for individual analysis and r = 0.037 for aggregated analysis, for Scotland). Unmarried people were more likely to report poor or fair self-rated health

Office for National Statistics

20

Population Trends 139

Figure 4

Spring 2010

Mortality rate by gender and country for those aged 65–74, ONS LS, SLS, NILS, 2001

Rate per 1000 35 30 25 20 15 10 5 0 England and Wales

Figure 5

Scotland

N. Ireland

Male

England and Wales

Scotland

N. Ireland

Female

Mortality rate by gender and country for those aged 35–74, ONS LS, SLS, NILS, 2001

Rate per 1000 12 10 8 6 4 2 0 England and Wales

Scotland

Male

N. Ireland

England and Wales

Scotland

N. Ireland

Female

Source: ONS LS, SLS, NILS 2001

than the married. In the individual level analysis, in which we were able to distinguish between unmarried groups, we found that the separated, divorced and never married, but not the widowed, were significantly more likely to report not good or fairly good health than the married. In England and Wales, and Scotland the widowed were in fact marginally less likely to report not good or fairly good health than the married (Odds Ratio (OR) for England and Wales 0.96, 95 per cent confidence interval (CI) 0.92–0.99). In all countries, those living in social housing, with no car, with no recorded educational qualification and in lower status occupations or not employed were the most likely to report not good or fairly good health. Reported health differentials by tenure appeared weaker in England and Wales than Scotland or Northern Ireland, whereas health differentials by NS-SEC appeared stronger in England and Wales than the other countries. For example, in England and Wales the odds of reporting not good or fairly good health among those who had never worked were 89 per cent higher than among managers or professionals

Office for National Statistics

21

Population Trends 139

Spring 2010

(CI 1.80–1.99), whereas the equivalent figure for Scotland was 55 per cent (CI 1.44–1.67). Differentials in health status by educational level appeared smaller in Scotland than in England and Wales or Northern Ireland. In general, those with missing data were more likely than the most advantaged reference category to report not good or fairly good health, but did not appear to have the worst health. Results from analysis of the aggregated datasets (Table 4) showed that in each country increasing socio-economic score (indicating a higher level of disadvantage) was associated with poorer reported health. This association appeared to be the strongest in Northern Ireland, where those in the most disadvantaged category had 5.4 times the odds of reporting not good or fairly good self-rated health than the least disadvantaged (CI 5.19–5.66). In England and Wales the equivalent ratio was 4.4 (CI 4.20–4.52) and in Scotland, 4.7 (4.47–4.89). After adjusting for age and gender (Table 4, Model 2), those in Northern Ireland were 10 per cent more likely to report not good or fairly good health (CI 1.09–1.11) than those in England and Wales, but there was no difference in this regard between Scotland and England and Wales. However, after additionally adjusting for marital status and socio-economic score (Model 2), the odds of reporting not good or fairly good self-rated health were slightly lower in Scotland than in England and Wales (OR 0.96, CI 0.95–0.97).

Mortality Figures 2, 3, 4 and 5 show mortality rates (deaths/person years of exposure) by country, age group and gender. In all age groups, men had higher rates of death than women. Those in Scotland had higher mortality rates than those in England and Wales or Northern Ireland, although in the youngest age group, in which the numbers of deaths observed were lowest, country differentials were small and not statistically significant. Age and sex standardisation demonstrated that for those aged 35–74, mortality rates in Scotland were 24 per cent higher than in England and Wales, and Northern Ireland’s mortality rate was three per cent higher. The main aim of the mortality analysis was to examine the association between health status and subsequent mortality in the three countries. Results show risks of death relative to a reference category. First, we briefly describe associations between other co-variates and mortality. In all countries rate ratios of mortality increased with age, and were higher for men than for women, a difference that increased once marital status and socio-economic status were controlled for (Table 5). Although widows and widowers were no more likely to report not good or fairly good health than the married, in all countries their risks of death were higher. Indeed in England and Wales, relative risk ratios for the widowed were as high as for the separated, divorced and never married. Consistent with the analysis of variations in self-rated health, mortality was highest for: tenants in social housing; those with no educational qualifications; and for those who had never worked, were unemployed, students or unclassified. Analysis of separate country aggregated datasets showed that there was a stronger association between socio-economic score and mortality in Northern Ireland than in the other countries. After control for self-rated health status, the association between socio-economic status and mortality weakened in all models and for all countries, but remained significant. In other words, while strongly related to survival, variation in health status only partly explained the association between socio-economic status and mortality.

Office for National Statistics

22

Population Trends 139

Spring 2010

Analysis of the combined country aggregated dataset demonstrated that, after controlling for age group and gender (Table 6), the Scottish sample had significantly higher risks of death in the 5 years and two months following the 2001 Census than those in England and Wales (RR 1.23, CI 1.19–1.27). In Northern Ireland, mortality risks were not significantly different from England and Wales (RR 1.01 CI 0.98–1.05). After control for socio-economic and marital status, the ratio for Scotland decreased marginally to 1.19 (CI 1.15–1.23) and the rate ratio for Northern Ireland fell to 0.95 (CI 0.92–0.98) indicating a significantly lower risk of death than in England and Wales (after control for marital status and socio-economic status). Additional control for self-rated health status (Model 3, all countries) did not alter the differences between countries in terms of mortality risks. Those reporting not good or fairly good health in 2001 were more than twice as likely to die in the follow up period than those reporting good health, after controlling for socio-demographic and socio-economic factors (Model 3, Tables 5 and 6). However there was some variation in the association found using the different analysis strategies, with rate ratios associated with reporting poor or fair health being 7–9 per cent higher in the analysis of the individual level data than in the aggregated dataset. This is probably because of poorer control for socio-demographic and socio‑economic factors in the analysis of the aggregated data, because of the need to use collapsed and less detailed indicators (age group rather than single year of age, two rather than four categories of marital status, and socio-economic score instead of separate socio-economic indicators). The association also varied by country. Using both analysis strategies we found that the association between health status and mortality was stronger in Scotland, after control for all other factors (aggregated analysis RR 3.01, CI 2.81–3.22) than in England and Wales (RR 2.57 CI 2.45–2.70) or Northern Ireland (RR 2.69 CI 2.54–2.86).

Summary of results Consistent with previous studies, these results showed that in all constituent countries of the UK, women were more likely than men to report not good or fairly good self-rated health, but were less likely to die in the follow up period. The never-married, divorced and separated were also more likely to report not good or fairly good health. All unmarried groups, including the widowed, were more likely to die in the follow up period than the married. Living in social housing, not having a car, having no educational qualifications and having never worked or being unemployed were all associated with higher levels of self-reported not good or fairly good health and with mortality, as was overall worse socio-economic score. There was some variation in the strength of these associations by country. Analysis using the socio-economic status score, for example, suggested that socio-economic differentials in health and mortality were larger in Northern Ireland than in Scotland or England and Wales. We found a strong association between reporting of not good or fairly good health and mortality in all countries. This association appeared stronger in Scotland than Northern Ireland or England and Wales. This reflects our finding that members of the Scottish sample were no more likely to report not good or fairly good health than those in England and Wales, but that they had higher relative risks of death. This might indicate variations in pre-death health status in different parts of the UK or differences in the thresholds at which people in different parts of the UK report not having good health, or a combination of both. This would account both for the apparently lower risks of poorer health in Scotland, despite higher mortality, and the stronger association between self-rated health and mortality in Scotland.

Office for National Statistics

23

Population Trends 139

Spring 2010

Discussion In this article, we explored different strategies for comparative analysis of the ONS LS, the SLS and the NILS. All three studies have a very similar design and, even though each country has its own census form, most questions are identical and there is UK-wide co-ordination on census form development, data collection and data processing.14,15 Registration of deaths and processing of mortality data are also co-ordinated and comparable. There are, however, some minor differences in categories used which need consideration, namely the inclusion of an additional educational qualification category in the England and Wales Census. There are also differences in the distribution of the populations by educational and housing tenure indicators, reflecting the fact that in Scotland upper secondary level qualifications are gained a year earlier than in England, Wales or Northern Ireland and that the social housing stock (relative to population size) is far larger. These differences may explain why differentials in Scotland ,in health by education appeared weaker and by housing tenure stronger, than in England and Wales or Northern Ireland. These country differences in education and housing tenure also influenced the comparability of the socio-economic score used in the aggregated datasets, which was based on all four socioeconomic indicators. For example Scotland had the highest proportion in the least disadvantaged category of the socio-economic score, which is likely to have been in part a result of the large proportion in the highest education category. Therefore, care must be taken in interpreting country differences, especially by socio-economic status. The other factor affecting comparability of results between countries is the differing proportions of non-respondents for the socio-economic status variables. This was a particular problem when combining socio-economic variables to produce the score used in analysis of the aggregated data set in which the proportions with missing data ranged from 13 per cent in Scotland to 20 per cent in Northern Ireland.

Strengths and weaknesses of each analysis strategy Development of the individual level datasets involved standard application procedures, and so they were quicker and easier to prepare and use than the datasets for the aggregated analysis. There were no limits on the variables and categories used in the individual level datasets because all analysis was carried out in the safe setting for each longitudinal study. Preparation of the aggregated datasets was much more time consuming and logistically complex. It took time to obtain approval for release of aggregated NILS and SLS datasets from their respective safe settings to the ONS safe setting, where analysis of the aggregated data set was undertaken, and for the statistical offices to put into place secure data transfer systems. Data set preparation also took much longer than for the individual level datasets, because of the iterative process necessary to ensure that all datasets met disclosure control protocols of each longitudinal study and ensure that they were also identical in terms of the variables and categories included. Statistically, the individual level datasets provided more detailed, richer information than the aggregated datasets, including individual year of age instead of three age groups, four marital status groups instead of only two, and separate socio-economic variables instead of a combined socio-economic score. We therefore obtained more detailed country comparisons of the associations between different socio-economic and demographic indicators associations using the individual level datasets, and variables (particularly age) were more completely controlled than in

Office for National Statistics

24

 

 

Widowed

Never married

   

0.090

1.63

1.89

1.34

1.11

1.00

1.48

1.46

1.72

1.16

1.00

1.07

1.49

1.00

1.21

1.25

1.84

1.00

1.08

0.96

1.20

1.00

 

***

***

***

***

***

***

***

***

***

***

***

***

***

*

***

 

Model 1

1.04

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.20

1.00

0.042

 

***

***

Office for National Statistics

Source: Analysis of ONS, GROS, NISRA

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.17,1.23

 

1.04,1.05

Model 2

0.101

1.53

1.55

1.34

1.11

1.00

1.42

1.60

1.23

1.00

1.20

1.54

1.00

1.38

1.40

1.99

1.00

1.10

0.97

1.31

1.00

1.13

1.00

1.04

 

***

***

***

***

***

***

***

**

***

***

***

***

***

***

***

***

 

Model 1

1.05

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

1.19

1.00

0.050

 

***

***

 

1.16,1.21

1.05,1.05

Model 2

0.110

1.50

1.62

1.39

1.12

1.00

1.49

1.88

1.30

1.00

0.83

1.45

1.00

1.29

1.41

2.10

1.00

1.11

1.01

1.35

1.00

1.13

1.00

1.04

 

***

***

***

***

***

***

***

***

***

***

***

***

***

***

***

***

 

 

1.45,1.56

1.54,1.70

1.35,1.44

1.08,1.15

 

1.42,1.56

 

1.82,1.95

1.26,1.35

 

0.78,0.89

1.40,1.49

 

1.22,1.37

1.34,1.48

2.03,2.17

 

1.08,1.15

0.97,1.06

1.30,1.39

 

1.11,1.15

 

1.04,1.04

Odds Sign Confidence ratio limits

Northern Ireland

Odds Sign Confidence ratio limits

  192,251

1.45,1.61

1.44,1.67

1.29,1.38

1.07,1.15

 

1.33,1.52

 

1.54,1.65

1.18,1.27

 

1.07,1.34

1.48,1.59

 

1.26,1.51

1.32,1.48

1.92,2.06

 

1.06,1.15

0.92,1.02

1.26,1.36

 

1.10,1.16

 

1.04,1.04

Odds Sign Confidence ratio limits

Scotland

Odds Sign Confidence ratio limits

  122,753

1.58,1.69

1.80,1.99

1.31,1.38

1.08,1.14

 

1.42,1.55

1.41,1.51

1.68,1.77

1.13,1.19

 

0.99,1.15

1.45,1.53

 

1.14,1.28

1.21,1.30

1.79,1.89

 

1.05,1.11

0.92,0.99

1.17,1.23

 

1.07,1.11

1.09

1.04,1.04  

***

***

1.00

1.04

Odds Sign Confidence ratio limits

Model 2

* p < 0.05 ** p < 0.01 *** p < 0.001. Model 1: Age. Model 2: Additionally includes marital status and socio-economic score

R2

   

0.040

 

 

Never worked, unemployed, student, other

Missing

 

 

 

254,918

 

Lower supervisory, technical, semi-routine and routine

Number in analysis

 

 

Intermediate occupations, small employers and own account

NSSEC (Reference: manager or professional)

 

 

 

Other

Missing

 

   

 

 

Lower secondary

 

 

 

 

 

 

None

Education (Reference: upper secondary or degree)

Missing

No

 

 

 

 

Car access (Reference: yes)

Missing

   

 

 

Social housing tenant

Private rental and other

 

 

 

 

1.15,1.19

 

1.04,1.05

 

 

***

***

 

Housing tenure (Reference: owner occupier)

 

Separated or divorced

 

1.17

Women

Marital status (Reference: married)

1.00

1.05

Odds Sign Confidence ratio limits

Model 1

England & Wales

Odds Ratios from logistic regression analysis of variations in the proportion of the population aged 35–74 with poor or fair self-rated health in 2001 by socio-demographic and socio-economic characteristics in England and Wales, Scotland and Northern Ireland. ONS LS, SLS, NILS 2001, using parallel datasets

Gender (Reference: men)

Age

 

Table 3

Population Trends 139 Spring 2010

25

65–74

3.02

Missing

R2

Number in analysis

Northern Ireland 0.073

4.36

5 (most disadvantaged)

0.037

2.70

4

254,918

1.86

3

Scotland 122,753

***

***

***

Sign

1.17,1.23

3.15,3.36

1.86,1.96

Confidence limits

1.33

1.00

1.12

1.00

2.53

1.75

Odds ratio

***

***

***

***

***

2.92,3.12

4.20,4.52

2.62,2.79

1.80,1.92

1.46,1.56

0.037

0.085

3.30

4.67

2.67

1.96

1.60

1.34

1.51

Country (Reference: England and Wales)

1.20

1.00

3.25

1.91

Odds ratio

1.23

1.19,1.28

1.19,1.23

1.06,1.09

2.66,2.79

1.73,1.80

Confidence limits

Model 1

1.00 ***

***

***

***

***

Sign

Model 2

Scotland

1.00

1.21

1.00

1.08

1.17

2.72 1.00

1.15,1.19

3.30,3.46

1.77

Odds ratio

1.00 ***

***

3.38

1.85,1.92

Confidence limits

2

Socio-economic score (Reference: least disadvantaged) 1

Not married

Marital status (Reference: married)

Women

Gender (Reference: men)

***

Sign

1.88

Odds ratio

Model 1

England and Wales

***

***

***

***

***

***

***

***

***

***

Sign

Model 2

3.15,3.45

4.47,4.89

2.56,2.79

1.87,2.05

1.52,1.67

1.28,1.41

1.30,1.37

1.10,1.15

2.44,2.62

1.71,1.80

Confidence limits

Odds Ratios from logistic regression analysis of variations in the proportion of the population aged 35–74 with poor or fair self-rated health in 2001 by socio-demographic and socio-economic characteristics in England and Wales, Scotland and Northern Ireland and for all countries combined. ONS LS, SLS, NILS 2001 using combined aggregated datasets

Age group (Reference: 35–49) 50–64

Table 4

Population Trends 139 Spring 2010

Office for National Statistics

26

Continued

***

3.75

2.93 5.42 3.13

4

5 (most disadvantaged)

Missing

Source: Analysis of ONS LS, SLS and NILS

Model 2: Additionally includes marital status and socio-economic score

Model 1: Age (and country for all areas combined)

* p < 0.05 ** p < 0.01 *** p < 0.001

R2 0.046

0.040

569,922

1.10

Number in analysis

0.99

3.01,3.25

5.19,5.66

2.82,3.04

2.00,2.17

1.62,1.77

Northern Ireland

***

***

***

***

***

Scotland

0.090

2.08

3

Country (Reference: England and Wales) ***

***

***

***

Sign

0.98,1.01

1.17,1.19

3.42,3.52

1.96,2.01

Confidence limits

1.28

1.00

1.10

1.00

2.76

1.84

Odds ratio

1.09,1.11

0.081

1.01

0.96

1.00

3.09

4.77

2.76

1.95

1.58

1.27

1.18

1.00

3.47

1.99

Odds ratio

1.28

1.22,1.34

1.31,1.37

1.09,1.13

2.88,3.04

1.96,2.05

Confidence limits

Model 1

1.00 ***

***

***

***

***

Sign

Model 2

All UK countries

1.00

1.34

1.69

192,251

1.11 1.00

1.19

2.96

2.01

Odds ratio

1.00 1.17,1.21

3.65,3.85

2.13,2.22

Confidence limits

1.00 ***

***

Sign

2.18

Odds ratio

Model 1

Northern Ireland

2

Socio-economic score (Reference: least disadvantaged) 1

Not married

Marital status (Reference: married)

Women

Gender (Reference: men)

65–74

Age group (Reference: 35–49) 50–64

Table 4

***

***

***

***

***

***

***

***

***

***

***

Sign

Model 2

0.99,1.02

0.95,0.97

3.02,3.16

4.66,4.88

2.70,2.82

1.90,1.99

1.55,1.62

1.24,1.30

1.26,1.30

1.08,1.11

2.72,2.81

1.82,1.87

Confidence limits

Population Trends 139 Spring 2010

Office for National Statistics

27

Population Trends 139

Spring 2010

the aggregated dataset analyses, as confirmed by the r-squared values. Additionally, it was only possible to carry out an exploration of the characteristics of non-respondents to certain census questions using individual level and not aggregated datasets, because of small numbers that would have precluded clearance of such an aggregated dataset. There are therefore a number of advantages to using the individual level datasets. However, the major drawback was the difficulty in formally ascertaining country differences in the outcomes of interest. Using the combined aggregated datasets, we were easily able to ascertain country differences in health and mortality controlling for all co-variates and so add considerably to our knowledge of UK inequalities in health and mortality, and associations between self-rated health and mortality. In summary, the individual level datasets provided much richer data with more variables and less time taken for dataset development (although for this project, this involved travel to three UK locations). However there was no easy way to make statistical comparisons between the countries. The combined aggregated datasets were logistically much more challenging and time consuming to prepare, had less variable detail, but enabled direct analysis of country comparisons. Both methods therefore have benefits, and the choice is likely to depend on the focus of research. Stringent disclosure control procedures on cell release of data from statistical office safe settings also means that this strategy would not be suitable for those wishing to analyse rare outcomes or more detailed variable categories. Although it is not possible at present, the ability to combine subsets of individual level data from the three studies would combine the benefits of both of the methods currently possible – there is no question that this approach would be scientifically stronger. Given that the census offices pass census data between them, we would hope that it should be possible to develop relevant protocols and legal agreements to make the passing of longitudinal study data a future possibility. Finally, in the course of this project we developed a number of resources, including a technical working paper, comparative data dictionary and a comparative overview of database structure that we hope will be useful for others wishing to pursue UK comparative analyses. These are available via the web sites of all three user support services.

Acknowledgements The research reported here was funded by the Economic and Social Research Council, grant reference RES-348-25-0013.The permission of the Office for National Statistics to use the Longitudinal Study is gratefully acknowledged, as is the help provided by staff of the Longitudinal Studies Centre – Scotland (LSCS); the Northern Ireland Longitudinal Study (NILS) and the Centre for Longitudinal Study Information and User Support (CeLSIUS) service. The LSCS is supported by the ESRC Census of Population Programme (Award Ref: RES-161-25-0001-01), the Scottish Funding Council, the Chief Scientist’s Office and the Scottish Government; the NILS is funded by the Department of Health, Social Services and Public Health, and the Research and Development Office of the Health and Personal Services in Northern Ireland. CeLSIUS, is supported by the ESRC Census of Population Programme (Award Ref: RES-348-25-0004). The authors alone are responsible for the interpretation of the data. Census output is Crown copyright and is reproduced with the permission of the Controller of HMSO and the Queen’s Printer for Scotland.

Office for National Statistics

28

Population Trends 139

Table 5

Spring 2010

Rate ratios of mortality for the population aged 35–74 by socio-demographic and socio‑economic characteristics and health status in England and Wales, Scotland and Northern Ireland. ONS LS, SLS, NILS 2001 using parallel datasets England & Wales Model 1

Model 2

Model 3

Rate Sign Confidence Rate Sign Confidence Rate Sign Confidence ratio limits ratio limits ratio limits Age

1.10

***

1.10,1.11

1.09

***

1.09,1.10

1.09

***

1.08,1.09

0.65

***

0.62,0.68

0.58

***

0.55,0.60

0.58

***

0.55,0.60

1.26

***

1.18,1.34

1.23

***

1.16,1.31

Gender (Reference: men) Women Marital status (Reference: married) Separated or divorced Widowed

1.23

***

1.14,1.32

1.24

***

1.16,1.33

Never married

1.24

***

1.15,1.34

1.26

***

1.16,1.36

1.51

***

1.43,1.60

1.38

***

1.30,1.46

Housing tenure (Reference: owner occupier) Social housing tenant Private housing tenant and other

1.25

***

1.14,1.37

1.19

***

1.09,1.30

Missing

1.39

***

1.23,1.56

1.34

***

1.19,1.51

No

1.49

***

1.41,1.58

1.40

***

1.32,1.48

Missing

1.17

*

1.01,1.36

1.17

*

1.00,1.35

1.00,1.19

1.06

***

1.30,1.52

1.26

***

1.17,1.36

Car access (Reference: yes)

Education (Reference: upper secondary or degree) Lower secondary

1.09

None

1.41

0.98,1.16

Other

1.27

***

1.16,1.40

1.17

**

1.07,1.29

Missing

1.50

***

1.36,1.65

1.37

***

1.24,1.51

0.97,1.13

1.02

0.95,1.10 0.98,1.12

NSSEC (Reference: manager or professional) Intermediate occupations, small employers and own account

1.05

Lower supervisory, technical, semi-routine and routine

1.11

**

1.04,1.19

1.05

Never worked, unemployed, student, other

1.34

***

1.21,1.50

1.21

***

1.09,1.35

Missing

1.21

***

1.12,1.31

1.11

**

1.03,1.20

2.38

***

2.26,2.50

Self-rated health (Reference: good health) Fair or poor health Total person years analysed R2

1,251,009 0.09

0.11

0.12

Office for National Statistics

29

Population Trends 139

Table 5

Spring 2010

Continued Scotland Model 1

Model 2

Model 3

Rate Sign Confidence Rate Sign Confidence Rate Sign Confidence ratio limits ratio limits ratio limits Age

1.11

***

1.10,1.11

1.10

***

1.09,1.10

1.09

***

1.09,1.09

0.67

***

0.63,0.71

0.61

***

0.57,0.64

0.60

***

0.57,0.64

Gender (Reference: men) Women Marital status (Reference: married) Separated or divorced

1.36

***

1.25,1.48

1.30

***

1.20,1.42

Widowed

1.14

**

1.05,1.25

1.15

**

1.05,1.25

Never married

1.28

***

1.16,1.41

1.29

***

1.17,1.42

Housing tenure (Reference: owner occupier) Social housing tenant

1.52

***

1.42,1.63

1.34

***

1.25,1.44

Private housing tenant and other

1.36

***

1.21,1.54

1.26

***

1.12,1.43

Missing

1.44

***

1.22,1.69

1.34

***

1.14,1.58

No

1.40

***

1.30,1.50

1.28

***

Missing

1.25

*

1.02,1.53

1.19

0.97,1.46

0.95,1.18

0.99

0.89,1.11

Car access (Reference: yes) 1.19,1.37

Education (Reference: upper secondary or degree) Lower secondary

1.06

None

1.32

***

1.21,1.45

1.18

***

1.07,1.29

1.38

***

1.21,1.58

1.26

***

1.11,1.43

0.92,1.14

1.02

Other Missing NSSEC (Reference: manager or professional) Intermediate occupations, small employers and own account

1.03

0.92,1.14

Lower supervisory, technical, semi-routine and routine

1.20

***

1.09,1.32

1.13

*

1.03,1.24

Never worked, unemployed, student, other

1.36

***

1.16,1.61

1.27

**

1.08,1.49

Missing

1.26

***

1.13,1.41

1.17

**

1.05,1.31

2.82

***

2.63,3.01

Self-rated health (Reference: good health) Fair or poor health Total person years analysed R2

597,711 0.10

0.13

0.15

Office for National Statistics

30

Population Trends 139

Table 5

Spring 2010

Continued Northern Ireland Model 1

Model 2

Model 3

Rate Sign Confidence Rate Sign Confidence Rate Sign Confidence ratio limits ratio limits ratio limits Age

1.11

***

1.10,1.11

1.10

***

1.09,1.10

1.09

***

1.09,1.09

0.64

***

0.61,0.67

0.58

***

0.55,0.61

0.57

***

0.54,0.60

Gender (Reference: men) Women Marital status (Reference: married) Separated or divorced

1.28

***

1.18,1.40

1.21

***

1.11,1.32

Widowed

1.13

**

1.05,1.23

1.13

**

1.04,1.22

Never married

1.26

***

1.16,1.36

1.25

***

1.16,1.35

Housing tenure (Reference: owner occupier) Social housing tenant

1.53

***

1.43,1.63

1.36

***

1.27,1.46

Private housing tenant and other

1.23

***

1.09,1.39

1.16

*

1.02,1.31

Missing

1.27

***

1.13,1.44

1.21

**

1.07,1.37

No

1.47

***

***

Missing

1.12

Car access (Reference: yes) 1.38,1.57

1.39

0.96,1.30

1.16

0.99,1.34

1.30,1.48

0.98,1.24

Education (Reference: upper secondary or degree) Lower secondary

1.16

*

1.03,1.31

1.10

None

1.39

***

1.25,1.54

1.20

***

1.08,1.33

1.49

***

1.31,1.69

1.32

***

1.16,1.50

0.97,1.18

1.04

Other Missing NSSEC (Reference: manager or professional) Intermediate occupations, small employers and own account

1.07

0.95,1.15

Lower supervisory, technical, semi-routine and routine

1.18

***

1.08,1.29

1.09

*

1.00,1.19

Never worked, unemployed, student, other

1.43

***

1.27,1.62

1.31

***

1.16,1.48

Missing

1.27

***

1.15,1.39

1.16

**

1.05,1.27

2.50

***

2.35,2.66

Self-rated health (Reference: good health) Fair or poor health Total person years analysed R2

928,238 0.11

0.13

0.15

* p < 0.05 ** p < 0.01 *** p < 0.001 Model 1: Age. Model 2: Additionally includes marital status and socio-economic score. Model 3: Additionally includes health status indicator Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

31

Population Trends 139

Table 6

Spring 2010

Rate ratios of mortality for the population aged 35–74 by socio-demographic and socio-economic characteristics and health status in England and Wales, Scotland and Northern Ireland, and for all countries combined. ONS LS, SLS, NILS 2001 using combined aggregated datasets England & Wales Model 1

Model 2

Model 3

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

50–64

3.78

***

3.50,4.09

3.69

***

3.41,3.99

3.28

***

3.04,3.55

65–74

13.02

***

12.08,14.03

11.19

***

10.37,12.07

9.17

***

8.50,9.90

***

0.56,0.61

Age group (Reference: 35–49)

Gender (Reference: men) Women

1.00 0.65

1.00 ***

Marital status (Reference: married)

0.62,0.68

0.58

1.00 ***

0.56,0.61

1.00

Not married

0.58 1.00

1.42

***

1.36,1.48

1.38

***

1.32,1.45

1

1.26

***

1.12,1.41

1.20

**

1.07,1.35

2

1.39

***

1.25,1.55

1.26

***

1.13,1.41

Socio-economic score (Reference: least disadvantaged)

3

1.57

***

1.41,1.74

1.35

***

1.22,1.50

4

1.94

***

1.76,2.15

1.56

***

1.41,1.73

5 (most disadvantaged)

2.93

***

2.65,3.25

2.16

***

1.95,2.40

Missing

2.48

***

2.25,2.73

1.94

***

1.76,2.14

2.57

***

2.45,2.70

Self-rated health (Reference: good health) Fair or poor health Country (Reference: England & Wales) Scotland Northern Ireland Total person years analysed

1,251,009

* p < 0.05 ** p < 0.01 *** p < 0.001 Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

32

Population Trends 139

Table 6

Spring 2010

Continued Scotland Model 1

Model 2

Model 3

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

50–64

4.25

***

3.83,4.72

3.98

***

3.58,4.42

3.52

***

3.17,3.91

65–74

13.97

***

12.63,15.45

11.23

***

10.14,12.45

9.18

***

8.29,10.18

***

0.57,0.64

***

1.30,1.46

Age group (Reference: 35–49)

Gender (Reference: men) Women

1.00 0.67

1.00 ***

Marital status (Reference: married)

0.64,0.71

0.61

1.00 ***

0.58,0.65

1.00

Not married

0.61 1.00

1.46

***

1.38,1.55

1.38

1

1.17

*

1.00,1.38

1.10

2

1.41

***

1.23,1.63

1.24

Socio-economic score (Reference: least disadvantaged) 0.94,1.29 **

1.08,1.43

3

1.61

***

1.41,1.84

1.36

***

1.19,1.56

4

1.94

***

1.72,2.19

1.51

***

1.34,1.71

5 (most disadvantaged)

2.97

***

2.65,3.33

2.06

***

1.83,2.31

Missing

2.70

***

2.41,3.04

2.01

***

1.79,2.26

3.01

***

2.81,3.22

Self-rated health (Reference: good health) Fair or poor health Country (Reference: England & Wales) Scotland Northern Ireland Total person years analysed

597,711

* p < 0.05 ** p < 0.01 *** p < 0.001 Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

33

Population Trends 139

Table 6

Spring 2010

Continued Northern Ireland Model 1

Model 2

Model 3

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

50–64

3.90

***

3.57,4.26

3.69

***

3.38,4.04

3.20

***

2.92,3.50

65–74

13.54

***

12.43,14.74

11.27

***

10.34,12.29

8.99

***

8.24,9.81

***

0.54,0.60

***

1.28,1.42

Age group (Reference: 35–49)

Gender (Reference: men) Women

1.00 0.64

1.00 ***

Marital status (Reference: married)

0.61,0.68

0.58

1.00 ***

0.55,0.61

1.00

Not married

0.57 1.00

1.41

***

1.34,1.49

1.35

1

1.26

**

1.06,1.50

1.18

2

1.60

***

1.37,1.87

1.40

Socio-economic score (Reference: least disadvantaged) 0.99,1.41 ***

1.20,1.64

3

1.80

***

1.56,2.08

1.49

***

1.29,1.72

4

2.12

***

1.84,2.43

1.62

***

1.41,1.86

5 (most disadvantaged)

3.44

***

2.99,3.95

2.37

***

2.06,2.72

Missing

2.71

***

2.37,3.10

2.03

***

1.77,2.33

2.69

***

2.54,2.86

Self-rated health (Reference: good health) Fair or poor health Country (Reference: England & Wales) Scotland Northern Ireland Total person years analysed

942,434

* p < 0.05 ** p < 0.01 *** p < 0.001 Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

34

Population Trends 139

Table 6

Spring 2010

Continued All Model 1

Model 2

Model 3

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

Rate Sign Confidence ratio limits

50–64

3.93

***

3.74,4.14

3.76

***

3.57,3.96

3.31

***

3.14,3.49

65–74

13.41

***

12.77,14.09

11.24

***

10.70,11.82

9.12

***

8.67,9.59

***

0.64,0.67

0.59

***

0.57,0.60

0.58

***

0.57,0.60

1.43

***

1.39,1.47

1.37

***

1.33,1.41

1

1.25

***

1.15,1.35

1.18

***

1.09,1.28

2

1.45

***

1.34,1.56

1.29

***

1.20,1.39

Age group (Reference: 35–49)

Gender (Reference: men) Women

1.00 0.65

Marital status (Reference: married) Not married Socio-economic score (Reference: least disadvantaged)

3

1.64

***

1.52,1.76

1.39

***

1.29,1.49

4

1.98

***

1.85,2.12

1.56

***

1.46,1.67

5 (most disadvantaged)

3.06

***

2.87,3.28

2.18

***

2.04,2.33

Missing

2.57

***

2.41,2.75

1.97

***

1.85,2.10

2.71

***

2.62,2.80

Self-rated health (Reference: good health) Fair or poor health Country (Reference: England & Wales) Scotland Northern Ireland Total person years analysed

1.00 1.23

***

1.01

1.19,1.27

1.19

***

1.15,1.23

1.2

***

1.16,1.24

0.98,1.05

0.95

**

0.92,0.98

0.94

***

0.91,0.97

2,791,153

* p < 0.05 ** p < 0.01 *** p < 0.001 Source: Analysis of ONS LS, SLS and NILS

Office for National Statistics

35

Population Trends 139

Spring 2010

References 1 O’Reilly D, Rosato M et al. (2005) ‘Self reported health and mortality: ecological analysis based on electoral wards across the United Kingdom’. British Medical Journal 331: 938–9. 2 Idler E and Benyamini Y (1997) ‘Self-rated health and mortality: a review of twenty-seven community studies’. Journal of Health and Social Behaviour 38: 21–37. 3 DeSalvo K B, Bloser N et al. (2005). ‘Mortality Prediction with a Single General Self-Rated Health Question: A Meta-Analysis’. Journal of General Internal Medicine 21(3): 267–275. 4 Singh-Manoux A, Dugravot A et al. (2007) ‘The association between self-rated health and mortality in different socioeconomic groups in the GAZEL cohort study’. International Journal of Epidemiology 36: 1222–1228. 5 Mitchell R (2005) ‘Commentary: The decline of death – how do we measure and interpret changes in self-reported health across cultures and time ?’ International Journal of Epidemiology 34: 306–308. 6 Rees P (1993) ‘Counting people: past, present and future’. University of Leeds. Review 36: 247–273. 7 Boyle P J, Gatrell A C et al. (1999) ‘Self-reported limiting long term illness, relative deprivation, and population stability in England and Wales’. Social Science and Medicine 49: 791–9. 8 Bardage C, Pluijm S et al. (2005) ‘Self-rated health among older adults: a cross national comparison’. European Journal of Ageing 2: 149–158. 9 Breakwell C and Bajekal M (2006) ‘Health expectancies in the UK and its constituent countries, 2001’. Health Statistics Quarterly 29: 18–25. 10 Hattersley L and Creeser R (1995) ‘Longitudinal Study 1971–1991. History, organisation and quality of data’. Series LS no. 7. London HMSO. 11 Boyle P J, Feijten P, Feng Z, Hattersley L, Huang Z, Nolan J and Raab G (2008) ‘Cohort Profile: The Scottish Longitudinal Study (SLS)’. International Journal of Epidemiology 38: 385–392. 12 More information available at: http://census.ac.uk/ 13 Under the NILS Disclosure Control Policy, outputs containing tabular data with counts lower than ten are not released from the secure setting. However, as an exception and to facilitate analysis by the research team, it was agreed that NILS would securely transfer data with counts lower than ten to the ONS LS secure setting, though the final product contains no counts lower than ten. 14 UK Census Committee (1999) ‘The 2001 Census of Population’. UK Census Committee, HMSO. 15 Office for National Statistics (2005) ‘Census 2001 General report for England and Wales’, HMSO. Available at: www.statistics.gov.uk/census2001/cn_143.asp

Office for National Statistics

36

Population Trends 139

Spring 2010

Do partnerships last? Comparing marriage and cohabitation using longitudinal census data Ben Wilson Office for National Statistics Rachel Stuchbury Office for National Statistics and CeLSIUS (Centre for Longitudinal Study Information and User Support), London School of Hygiene & Tropical Medicine

Abstract The stability of couple partnerships is of continual interest to policy makers and many users of official statistics. This research used a sample of adults (from the Office for National Statistics Longitudinal Study) who were in a partnership (married or cohabiting) in the 1991 Census of England and Wales, and then explored whether these individuals were living with the same partner in 2001. Marital partnerships were found to be more stable, even when additional factors were taken into account. Of adults aged 16 to 54, around four in five adults (82 per cent) that were married in 1991 were living with the same partner in 2001. The equivalent figure for adults cohabiting in 1991 was around three in five (61 per cent), of whom around two-thirds (of those remaining with the same partner) had converted their cohabitation to a marriage by 2001. Long-running partnership stability was also found to vary according to the socio‑demographic characteristics of individuals and their partners and a summary of these variations is discussed.

Office for National Statistics

37

Population Trends 139

Spring 2010

Contents Abstract............................................................................................................................................ 37 Introduction....................................................................................................................................... 40 Previous research and different sources of data.............................................................................. 41 Analysis............................................................................................................................................ 42 Results............................................................................................................................................. 45 Changes in partnership status: cohabitation.................................................................................... 46 Comparing marriage and cohabitation............................................................................................. 47 Factors associated with stability....................................................................................................... 47 The influence of multiple factors....................................................................................................... 49 Further modelling of partnership outcomes...................................................................................... 53 Discussion........................................................................................................................................ 53 Key Findings..................................................................................................................................... 54 Acknowledgements.......................................................................................................................... 54 Appendix.......................................................................................................................................... 55 References....................................................................................................................................... 59

List of figures Figure 1

Changes in partnership status.................................................................................... 40

Office for National Statistics

38

Population Trends 139

Spring 2010

List of tables Table 1

Partnership status and legal marital status................................................................. 43

Table 2 Partnership status by sex, percentage in each age group in 1991............................. 43 Table 3 Partnership status in 2001 by age in 1991 (percentage in each age group).............. 44 Table 4 Partnership status in 2001 by partnership status in 1991 (percentages)................... 45 Table 5 Partnership status in 2001 by partnership status and age in 1991 (percentages)..... 46 Table 6

Probability of having the same partner in 2001.......................................................... 48

Table A1 Whether enumerated in the 2001 Census by de facto status in 1991........................ 55 Table A2

Whether enumerated in the 2001 Census by age in 1991......................................... 56

Table A3a Partnership status by sex, percentage in each age group in 1991............................. 57 Table A3b Partnership status by sex, percentage in each age group in 1990/91........................ 57 Table A4 Probability of being married to same partner in 2001 (if cohabiting in 1991)............. 58

Office for National Statistics

39

Population Trends 139

Spring 2010

Changes in partnership status

Figure 1

B

Not in a partnership

Married

E D

F

A

C Cohabiting

Formation Dissolution A: Cohabitation formation (residential partnership, not with a spouse) B: Marriage – without prior cohabitation (may or may not live with spouse) C: Marriage – with prior cohabitation (may or may not live with spouse) D: Cohabitation dissolution (end of residential partnership) E: Divorce/Separation (end of marriage and/or living with spouse) F: Divorce/Separation (moving in with a new partner, may still be legally married)

Introduction There have been notable changes in UK partnership behaviour over the last 40 years. Divorce rates rose considerably during the 1970s1, remained broadly stable after the mid-1980s, and more recently have fallen since 20042. At the same time, there has been a long-term fall in marriage rates since the beginning of the 1970s, and a steady increase in the proportion of adults cohabiting3. For unmarried men in Great Britain aged 16 to 59, the proportion cohabiting increased from 11 per cent in 1986 to 27 per cent in 2007. There was a similar change for equivalent unmarried women, from 13 per cent to 28 per cent4,5. This change in partnership behaviour is likely to persist. According to demographic projections, the long-term rise in cohabitation will continue, with the number of cohabiting couples in England and Wales projected to rise from 2.25 million in 2007 to 3.70 million in 20316. The same figures show that the proportion of the adult population that is legally married is projected to fall from 49 per cent in 2007 to 41 per cent by 20317. Official statistics provide considerable information on the estimated and projected population by partnership status. However, there is limited comparative information on the stability of different partnerships8. Furthermore, although the characteristics of married and cohabiting couples are available from various sources3, information on the factors associated with stability is also limited, largely due to a lack of suitable data (discussed later in this article). Information about partnership stability is important for many different users of official statistics. For example, discussions about the legal rights of cohabiting couples might be informed by comparing the stability of marriage and cohabitation9. This comparison also has implications for policy areas concerning children in different family types. Knowledge of partnership stability therefore informs policy connected with fertility, education, poverty, and any aspect of child welfare (including maintenance and contact with parents). In addition, as the prevalence of cohabitation and divorce has increased at older ages10, it is of interest to consider the impact that changes in partnership stability might have on older people. The UK is an ageing society11, and any changes in older people’s partnership histories or those of their progeny may affect family networks, care arrangements, or retirement income. From a research perspective, it is of great interest to discover how far the predictive power of marital status (for morbidity, mortality, socio-economic wellbeing Office for National Statistics

40

Population Trends 139

Spring 2010

and other outcomes) can also be attributed to cohabitation status (and for whom). For all of these topics, it is not just stability that is of interest, but also the extent to which cohabitation transitions differ from marital transitions.

Previous research and different sources of data The study of partnership stability ideally requires data on partnership formation, dissolution, and transformation (from cohabitation to marriage). Cohabitation may end when two partners cease to live together (dissolution) or when two partners decide to marry (formation), but a marriage will only end when it dissolves (see Figure 1)12. In this case, any analysis must take account of those who cohabit and then marry. Considering all this, two ways to gather information on stability (or partnership transitions) are by: 1. collecting retrospective partnership histories, and 2. using prospective longitudinal data13 It is also desirable that marriage can be reliably distinguished from cohabitation, and that the results should be valid for the whole population14. The General Household Survey (GHS) has included annual questions on partnership history – including cohabitation – since 1979 (for women), and 1986 (for men)15. Research using this source shows that in Great Britain there have been long-run increases (since the 1950s) in the proportion of married women cohabiting before marriage16. Among those cohabiting in their first union, a majority will marry their partner, although this proportion declines for more recent first unions17. Current cohabitations, that is, those cohabiting at the time of the survey, tend to have begun more recently than current marriages (although this compares partnerships that are not yet completed)18. Nevertheless, it should be noted that the median duration of cohabitation increased between 1979 and 199519. There are issues with research (such as that quoted above) using partnership history data. Marriage and cohabitation histories from cross-sectional data (such as the GHS) have the disadvantage that it is only possible to examine the partners by their characteristics at one point in time. Also, retrospective history data can suffer from respondent recall problems, which are known to be more likely with informal events such as the start or end of a cohabiting relationship20. On the other hand, partnership stability can be researched using longitudinal birth cohort studies21, although it takes several decades before the subjects themselves have acquired sufficient experience of partnerships. It is possible to examine parental partnerships in birth cohort studies. For example, results from the Millennium Cohort Study (MCS) showed that children living with both their natural parents at nine months were more much likely to remain so at five years if the parents were married to each other at nine months rather than cohabiting22. Of course, this result does not consider partnerships where neither partner has children in the household, and like other birth cohort studies it is only valid for a single cohort of children born between 2000 and 2002. Longitudinal data where the panel is continuously refreshed can offer a reliable sample for the whole population in any year. The British Household Panel Survey (BHPS) is one such source, and has the advantage that partnership histories have been collected from most respondents. Previous research has combined these histories with data from different waves of the survey to analyse partnership transitions. For example, it has been estimated that within 10 years about three-fifths of

Office for National Statistics

41

Population Trends 139

Spring 2010

first cohabitations turn into marriage, while just under a third dissolve23. The BHPS has also been used to show that cohabiting couples are more likely than married couples to separate24. One problem with the BHPS is its relatively small sample size. This is the case particularly when looking at the cohabiting population (which is much smaller than the married population). An alternative source (used for the research reported in this article), is the Office for National Statistics (ONS) Longitudinal Study (LS). This has a much larger sample, one per cent of the population, and has been used in previous research to explore partnership stability25. This research showed that adults in couples (either married or cohabiting in 1991) who had a dependent child in the household (in 1991) were more likely to be lone parents in 2001 compared with couples who had no dependent children in the household (in 1991). They were also less likely to be ‘not in a family’ (that is. not partnered or a lone parent). Other research using the LS has shown that only a fifth of cohabiting adults in 1991 were still cohabiting with the same partner in 2001 (although a further two-fifths had married their 1991 partner)26. The research in this article follows on from this analysis to compare cohabiting and married partnerships, and to explore the factors associated with stability. Unfortunately, apart from information on dissolutions due to widowhood, the LS only contains partnership information for respondents every 10 years (for more information on the LS see the section Analysis below). This means that it is not possible to know exactly when partnerships start or end, or to consider each individual’s amount of exposure to the different partnership states. It also means that some partnerships can be missed altogether because they begin and end between two censuses. Of course, even when data are collected annually, changes within the year may be missed27, and this should be considered when interpreting the results presented here and elsewhere. Thus the term ‘stability’ is used here to refer to long-term changes in partnership status, and the results only apply to a selected cohort of individuals (those enumerated at the 1991 and 2001 censuses of England and Wales). Bearing these restrictions in mind, the questions addressed by this research are: • What proportion of individuals remain with the same partner over a 10-year period? • What are the differences between the stability of marriage and cohabitation? • What are the characteristics associated with partnership stability? • To what extent does cohabitation end in marriage, and what are the associated factors?

Analysis This research uses the ONS Longitudinal Study (LS) to explore what happened to a cohort of individuals who were married or cohabiting in 1991. It examines their partnership status 10 years later in 2001, whether they are still living with the same partner, and what factors are associated with changes in partnership. As with all of the LS results in this article, the data are for England and Wales. The LS sample is selected by birthday, and continually replenished as new members with LS birthdays are born or migrate into England or Wales. Data comprise linked census records from 1971, 1981, 1991 and 2001 for sample members plus census records for those in their household at each census. Data from vital events are also added, including birth or death of a sample member, births and deaths of children to sample mothers and widowhoods to sample members. Vital event information on marriage and divorce registration cannot be included in the

Office for National Statistics

42

Population Trends 139

Table 1

Spring 2010

Partnership status and legal marital status

Longitudinal sample, England and Wales, All adults aged 16+ in 1991 Partnership status

1991

Living with a partner

%

2001

194,092

61

194,712

61

213,554

Married and living with spouse Cohabiting – single

%

220,117

12,343

4

14,251

4

Cohabiting – married (separated)

1,077

0

243

0

Cohabiting – divorced

5,653

2

10,166

3

Cohabiting – widowed

389

0

745

0

Not living with a partner

104,979

Single Married (separated)

98,416

67,811

21

35,280

11

6,302

2

2,832

1

Divorced

14,425

5

27,921

9

Widowed

16,441

5

32,383

10

318,533

100

318,533

100

Total

Note: These frequencies are for the same sample of individuals in 1991 and 2001 Source: ONS Longitudinal Study (authors’ analysis)

LS, as date of birth, the key variable for matching data sources, is not asked on the registration forms. In addition, since cohabitation (formation or dissolution) is not registered in any way there is no corresponding way of including inter-censal information on cohabitation. To begin with, a sub-sample of the LS was taken, giving over 435,000 adults (aged 16 and over) who were enumerated at the 1991 Census28. After removing those living in communal establishments and visitors to private households in 1991, the sample was reduced to 417,000. It was further reduced by the selection of those who were also enumerated at the 2001 Census.

Table 2 Partnership status by sex, percentage in each age group in 1991 Longitudinal sample, England and Wales 16–24

25–34

35–49

50–59

Total (16–59)

Lone

45

22

20

12

100

Cohabiting

33

40

23

5

100

5

27

46

23

100

Lone

49

25

18

8

100

Cohabiting

20

44

29

7

100

2

24

48

26

100

Women

Married1 Men

Married1 1 Married and living with spouse

Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

43

Population Trends 139

Spring 2010

These numbered 318,533 and formed the sample for this study, referred to henceforth as the ‘longitudinal sample’29. Table A1 in the Appendix shows the initial sub-sample by partnership status at 1991 and whether they were present at the 2001 Census. Over three-quarters of adults present in 1991 were also present in 2001, with 14 per cent having died or embarked between 1991 and 2001, and the remaining 11 per cent ‘missing’. The latter represent all individuals unaccounted for in the 2001 Census. There are many possible reasons for this, but the most likely are non-response in the 2001 Census or migration to a location outside England and Wales (without notifying a General Practitioner)30. Compared with women, men were more likely to be missing in 2001. This was particularly the case for men who were cohabiting or not living with a partner in 1991. Compared with married women, married men were more likely to have died or embarked. Around 97 per cent of the 60,000 deaths and embarkations (of men and women) were deaths, so it is likely that this largely reflects the fact that a marriage is more likely to end by the death of the male partner rather than the female partner31. There are also variations in whether initial sub-sample members were ‘missing in 2001’ by age (see Table A2 in the Appendix). Partnership status variables for 1991 and 2001 were constructed for this analysis. It should be noted that they were intended to represent actual partnerships in the household, so adults were only classified as married if the spouse was present in the household at census, and the same of course applied to cohabitation. A few spouses and partners will not have been recorded by the census (in 1991 or 2001), and therefore both married and cohabiting adults will be slightly undercounted in favour of people not living with a partner. Since there was no direct question about cohabitation in the 1991 Census and no household relationship grid, partnership status was derived from information about relationships in the family and household (as explained below). This means that there will also be a slight tendency throughout this research to undercount those cohabiting32. Partnership status in 1991 was derived from the LS member’s position in the family33, the relationship of other household members to the LS member, and the sex, age and marital status of all household members. In 2001 it was derived from the same factors in 2001, as well

Table 3 Partnership status in 2001 by age in 1991 (percentage in each age group) Longitudinal sample, England and Wales, All adults cohabiting in 1991 Partnership status in 2001 With the same partner

16–24

25–34

35–44

45–54

55–64

65+ Total (16+)

51

62

67

70

67

51

61

Cohabiting with the same partner

11

20

33

38

42

35

23

Married to the same partner

41

43

34

32

25

16

39

49

38

33

30

33

49

39

Partnership has ended Cohabiting with a new partner

13

8

6

4

3

1

8

Married to a new partner

15

10

6

4

4

1

10

Not living with a partner

21

20

21

22

27

46

21

100

100

100

100

100

100

100

All individuals in age group Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

44

Population Trends 139

Spring 2010

as the LS member’s partnership status in 1991, and the widowhood records in the LS for 1991 to 2001. Other people in an LS member’s household are not linked from census to census, so there is no cross-census identifier for them. The sex, date of birth, marital status and relationship to LS member of the LS member’s partner from 1991, were used to determine whether that person was still in the LS member’s household 10 years later.

Results Table 1 provides a summary of partnership status for 1991 and 2001 respectively. In both years, around two thirds of adults are living with a partner. These may be different individuals in different years (the table does not show changes in individual partnership status). Nevertheless, the table indicates that partnership is more common than not living with a partner, and that the majority of partners are married. In 2001 there are larger proportions of divorced and widowed adults not living with a partner, but this is to be expected given the fact that the sample is older in 200134. Before investigating changes in individual partnership status, it is worth looking more closely at the distribution of sample members by partnership status in 1991. Table 2 shows that in 1991, cohabiting men and women tended to be younger than those who were married and living with their spouse. Lone adults (that is not in a partnership) tended to be younger still. The raw data from Table 2 was also compared with published GHS data for 1990/199135. Tables A3a and A3b (in the Appendix) provide a summary of the comparison, which shows that the adult population by partnership status has a similar age distribution for both sources (LS and GHS). It may therefore be assumed that the sample is broadly representative of the 1991 adult population (by age and partnership status), despite the fact that non-response will affect both sources, and non-response may be different for the GHS and the 1991 Census. (For information on adults not responding to the 2001 Census that were excluded from this sample, see Appendix Tables A1 and A2.) There are additional issues that may affect both sources, but the comparison provides verification that cohabiting adults were successfully identified from the 1991 Census.

Table 4 Partnership status in 2001 by partnership status in 1991 (percentages) Longitudinal sample, England and Wales, All adults aged 16 to 54 in 1991 Partnership status in 2001

Cohabiting in 1991

Married in 1991

All partnerships in 1991

With the same partner

61

82

79

Cohabiting with the same partner

22

0

3

Married to the same partner

39

82

77

39

18

21

Partnership has ended Cohabiting with a new partner

9

3

4

Married to a new partner

10

5

5

Not living with a partner

21

10

12

100

100

100

Total Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

45

Population Trends 139

Spring 2010

Table 5 Partnership status in 2001 by partnership status and age in 1991 (percentages) Longitudinal sample, England and Wales, All adults aged 16 to 54 in 1991 Partnership status in 2001

16–24

25–34

35–44

45–54

Total (16–54)

Married in 1991 With the same partner

64

77

84

87

82

With a new partner

19

12

7

4

8

Not with a partner Total

17

12

9

9

10

100

100

100

100

100

51

62

67

70

61

Cohabiting in 1991 With the same partner With a new partner

27

18

12

8

18

Not with a partner

21

20

21

22

21

100

100

100

100

100

Total

Source: ONS Longitudinal Study (authors’ analysis)

Changes in partnership status: cohabitation As indicated in Figure 1, cohabiting partnerships may end due to marriage, separation or death, whereas marriages end in separation (and/or divorce) or death. To consider this additional complexity, Table 3 shows only the population that were cohabiting in 1991, and what their partnership status was in 2001. Of all cohabiting adults in 1991, 61 per cent were living with the same partner in 2001 – 23 per cent cohabiting and 39 per cent married. Another way to summarise this is that over the 10-year period, almost two in five cohabiting partners separated, and almost two in five married their partner, while the remainder were still cohabiting. Table 3 also shows considerable variation by age. Cohabitants aged 45 to 54 years were most likely to remain with the same partner (compared with other age groups). The youngest cohabitants aged 16 to 24, and the oldest aged 65 and over were the most likely to have separated. However, although the youngest age group were the most likely to be living with a new partner (married or cohabiting), the oldest were the most likely not to be in a partnership. These differences no doubt reflect the influence of mortality at older ages. In addition, cohabitation among the young might be expected to be more transient, and this is reflected in both the high level of separation (cohabitation as a trial relationship) and the high level of cohabitants that marry (cohabitation as a precursor to marriage). At ages over 35, the higher proportions of cohabitants that remain in a cohabiting relationship with the same partner may be indicative of cohabitation as a substitute for marriage at these ages (although it is not possible to state this with certainty). Further analysis was carried out looking at the differences between male and female cohabitants. Overall and at all ages female cohabitants were found to be more likely to have separated from their partner over the 10 years compared with male cohabitants. They were also more likely not to be living with a partner in 2001 (24 per cent, compared with 17 per cent for men), a fact that is partially explained by mortality differentials between the sexes, and the likelihood that a male partner will on average be older than the female partner36.

Office for National Statistics

46

Population Trends 139

Spring 2010

Comparing marriage and cohabitation Considering the above results, it is possible to compare the stability of couples who were cohabiting in 1991 with those who were married (Table 4). For this comparison the age group (in 1991) has been restricted to 16 to 54-years-olds. This restriction does not materially affect the distribution of partnership outcomes (as illustrated by comparing the total column in Table 3 with the cohabiting column in Table 4). However, it does allow widowhood to be largely discounted as a reason for partnership dissolution, which is important given the younger mean age of cohabiting adults compared with the married population. Table 4 shows that adults aged 16 to 54 in 1991 were more likely to be living with the same partner in 2001 if they were married. Around four in five married adults (82 per cent) were living with the same partner in 2001, compared with around three in five cohabiting adults (61 per cent). Of those that were no longer living with the same partner (having been married or cohabiting), a little more than half were not living with any partner at all. The remainder were living with a new partner, with a slightly higher likelihood of being married rather than cohabiting. Table 3 showed variations in the stability of cohabitations by age, and Table 5 shows similar results for all partnerships in 1991. Previous research has shown that adults who marry at younger ages are more likely to divorce, and the results in Table 5 do not contradict this finding37. However, it should be remembered that the duration of existing partnerships in 1991 is not known, either for marriage or for cohabitation. Importantly, the effects of age are similar for both marriage and cohabitation, with young adults in partnerships in 1991 more likely to be separated from their partner in 2001. Despite the general finding that marriage is more stable than cohabitation, it is interesting to note that the youngest married adults (aged 16 to 24 in 1991) were less likely to be living with the same partner in 2001 compared with older cohabiting adults (aged 45 to 54). Despite this, marriages were more stable when comparing partnerships in each age band. As with those cohabiting adults that separated, married adults that separated were more likely to be living with a new partner if they were young (aged 16 to 24), and more likely to live without a partner if they were older (aged 35 to 54).

Factors associated with stability Table 5 shows the influence of a single factor – age on partnership stability. However, it is likely that other socio-demographic factors will influence whether individuals remain with the same partner. These other factors may also explain the variation by age. For example, younger partnerships may be less stable, but this may be because young people are more likely to have other risk factors associated with instability. Reviewing the results of previous research, it is difficult to prepare an exhaustive list of potential factors, partly because factors vary over time and according to which population is being studied. In addition, much research focuses on marital stability (partly because of data constraints), and caution should be exercised when considering the similarity of marital and cohabiting stability. With this in mind, it is useful to mention a review published by the Lord Chancellor’s Department, which stated that socio-demographic factors affecting marital stability may be placed in three groups: characteristics of the individual’s parents, marital factors (demographic factors associated with the couples’ partnership history and childbearing experience), and the individual’s own socio-economic characteristics38. Office for National Statistics

47

Population Trends 139

Table 6

Spring 2010

Probability of having the same partner in 2001

Longitudinal sample, England and Wales, Adults aged 16–54 and in partnerships in 1991 (Model 3 & 4 are sub-samples) MODEL 1: individual characteristics (n = 156,739) Variable

Age in 1991

MODEL 2: MODEL 3: including partner cohabiting couples characteristics (in 1991) only (n = 156,739) (n = 18,501)

MODEL 4: women only (and if they had a baby) (n = 82,467)

Odds ratio1

Sig. level2

Odds ratio1

Sig. level2

Odds ratio1

Sig. level2

Odds ratio1

Sig. level2

1.05

***

1.05

***

1.04

***

1.06

***

0.95

***

0.97

***

0.96

***

Age gap in absolute years (0 = man 2 years older) Married in 1991 Cohabiting (reference category)

1.00

n/a

1.00

n/a

1.00

n/a

Married

1.83

***

1.73

***

1.78

***

Female (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

Male

1.11

***

1.11

***

1.13

***

No (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

Yes

1.07

***

1.07

***

1.12

***

1.05

**

Yes (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

No

1.25

***

1.10

***

1.16

*

1.10

**

Single or widowed or married (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

Remarried or divorced (or married if cohabiting)

0.62

***

0.73

***

0.79

***

0.72

***

Degree or higher

1.38

***

1.21

***

1.18

**

1.12

**

Other professional or vocational qualification

1.21

***

1.13

***

1.14

*

1.15

***

No degree or professional qualification (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

One: professional

1.20

***

1.12

**

1.12

1.15

Two: managerial or technical

1.05

***

0.98

0.97

0.95

*

Three: skilled non-manual

1.20

***

1.11

***

1.12

**

1.12

***

Three: skilled manual

1.14

***

1.09

***

1.11

**

1.01

Four: part-skilled, unskilled, other (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

Unemployed (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

1.00

n/a

Not economically active

1.31

***

1.25

***

1.14

*

1.25

***

Self-employed

1.33

***

1.21

***

1.20

**

1.22

***

Employed

1.38

***

1.24

***

1.26

***

1.16

***

Single or widowed or married (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

Remarried or divorced (or married if cohabiting)

0.90

***

1.04

0.92

***

Yes (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

No

1.61

***

1.17

*

1.33

***

Sex

Dependent children in household in 1991

Has limiting long term illness in 1991

Previous dissolution (marital status in 1991)

Higher qualifications in 1991

Social class (Registrar General’s) in 1991

Economic activity in 1991

Partner: previous dissolution (marital status in 1991)

Partner: has limiting long term illness in 1991

Office for National Statistics

48

Population Trends 139

Table 6

Spring 2010

Continued

Longitudinal sample, England and Wales, Adults aged 16–54 and in partnerships in 1991 (Model 3 & 4 are sub-samples) MODEL 1: individual characteristics (n = 156,739) Variable

Odds ratio1

Sig. level2

MODEL 2: MODEL 3: including partner cohabiting couples characteristics (in 1991) only (n = 156,739) (n = 18,501) Sig. level2

MODEL 4: women only (and if they had a baby) (n = 82,467)

Odds ratio1

Sig. level2

Odds ratio1

Odds ratio1

Sig. level2

Degree or higher

1.23

***

1.12

1.32

***

Other professional or vocational qualification

1.16

***

1.33

***

1.17

***

No degree or professional qualification (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

One: professional

1.19

***

1.19

*

1.22

***

Two: managerial or technical

1.09

***

1.08

1.20

***

Three: skilled non-manual

1.10

***

1.07

1.07

*

Three: skilled manual

1.09

***

1.02

1.15

***

Four: part-skilled, unskilled, other (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

Unemployed (ref.)

1.00

n/a

1.00

n/a

1.00

n/a

Not economically active

1.35

***

1.33

***

0.84

***

Self-employed

1.35

***

1.55

***

1.43

***

Employed

1.43

***

1.56

***

1.57

***

No (ref.)

1.00

n/a

Yes

1.28

***

Partner: highest qualification in 1991

Partner: social class (Registrar General’s) in 1991

Partner: economic activity in 1991

Had a baby between 1991 and 2001

Note: For Registrar General’s social class, other includes armed forces and missing 1 Reference categories are shown with an odds ratio of 1.00 2 * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level n/a = reference category (significance is not applicable) Source: ONS Longitudinal Study (authors’ analysis)

In the case of this research, the limits of the LS data mean that it is not possible to explore either parental characteristics or some of the marital factors, such as age at marriage39. The same can be said for psychological factors, such as behavioural and emotional problems, or wider social factors (such as the effects of legislation on divorce and the rights of cohabiting couples). A final restriction relates to unavailable socio-economic characteristics that would ideally be of interest, such as income and religious belief40.

The influence of multiple factors The next stage of this research uses logistic regression to create four models. Each of these models explores the influence of multiple factors on a single outcome. that is whether an individual who is partnered in 1991 remains with the same partner in 200141 (for an example of logistic regression using the LS, see the online training module42). Office for National Statistics

49

Population Trends 139

Spring 2010

The first model explores the effect of individual characteristics; the second extends this to include the characteristics of their partner; the third looks at 1991 cohabiting adults in isolation (that is the model excludes those who were married in 1991); and the fourth looks at women only – both married and cohabiting in 1991. It was decided to use 1991 data for all explanatory variables so that circumstances prior to the outcome were being investigated. Using 1991 data, the following individual factors were investigated: • age – which indicates birth cohort and will be correlated with length of partnership up to 1991 • whether married or cohabiting – one of the main factors of interest • whether dependent children were in the household. In 1991 a dependent child was a child aged under 16 years, or a never married, economically inactive, full-time student aged under 19 years • limiting long-term illness – to measure health • marital status – indicating previous marital dissolution • highest qualification – to measure socio-economic potential43, • social class – to measure socio-economic circumstances, and • employment status – to measure economic circumstances Partner characteristics included the same variables used to measure individual factors. Age of partner was not included because this was measured by looking at absolute age difference between partners44. Sex of the LS member was also included for all models except the fourth, which looked at women only45. To investigate the influence of childbirth on stability in the fourth model, a variable was added showing the effect of whether women gave birth to a living child between 1991 and 2001. This was the only factor using data from between the two censuses, and was made possible because annual birth registrations are linked to individual data in the LS. The results of all four models are shown in Table 6, which compares the influence of multiple factors on stability. Table 6 also shows the effect of a single factor, for example age, when other factors are held constant, that is, net of other factors46. In all the models, a reference category is chosen for each categorical variable. The other categories of this factor are then interpreted in comparison to the reference category. Therefore the reference category itself has an odds ratio of one. For example, in Model 1 the odds ratio for adults with no limiting long-term illness in 1991 is 1.24. This means that the odds of remaining with the same partner in 2001 are 1.24 times higher for those without a limiting long-term illness (compared with those who do have a limiting long-term illness), all other factors being equal47. For the two continuous variables, age and age difference, an odds ratio shows the effect of a change in one unit, that is one year48.

Model 1 Model 1 shows the likelihood of an individual remaining with the same partner in 2001 according to individual factors. The model includes both men and women, aged 16 to 54 in 1991, who were either married or cohabiting in 1991. Notable results are as follows: • Marriage remains more stable than cohabitation after controlling for individual factors. Those who were married were more likely to remain with the same partner (the odds of remaining with the same partner if you were married in 1991 were 1.83 times the odds if you were cohabiting).

Office for National Statistics

50

Population Trends 139

Spring 2010

• Adults were less likely to remain with the same partner if, in 1991, they were: -- younger -- cohabiting -- had no dependent children living in the household -- had a limiting long-term illness -- had previous experience of partnership dissolution -- had no higher qualifications -- had a low social class, or -- unemployed • The fact that there is a significant difference between men and women suggests that the sample may be affected by attrition. That is, given that there were equal numbers of men and women in the population of opposite-sex residential partnerships in 1991, there should be no sex differences. According to the model, men have more stable partnerships, but they are also more likely to be missing from the sample (see Appendix Table A1). This suggests that men in less stable partnerships may be more likely to be missing from the sample49. Two points are worth mentioning when interpreting these results. The first is that possible selection effects should be considered. For example, those adults who are more likely to have stable relationships may also be more likely to marry (rather than cohabit). The married and cohabiting populations have different characteristics, and it may be these different characteristics, rather than the partnership arrangements themselves, that result in the differences in stability. Without a more refined model, it is not possible to be certain about the impact of selection effects on these results. The second point worth mentioning is that all of the factors in the model are significant at the 1 per cent level. However, in some respects this is unsurprising given the very large sample size (almost 157,000 adults).

Model 2 Model 2 is the same as Model 1, but also includes characteristics of each individual’s partner in 1991. Notable results are as follows: • The inclusion of partner’s characteristics does not materially affect the difference in stability between married and cohabiting partnerships • Most of the individual factors remain broadly the same (in magnitude and direction). However, the effect of limiting long-term illness is reduced, and the effect of social class becomes less clear50 • A larger age difference between partners reduces the likelihood of remaining with the same partner in 2001 • Partner’s characteristics are all significant and are similar in direction to individual factors. Adults were less likely to remain with a partner who in 1991 had: -- a limiting long-term illness -- previous experience of marital dissolution Office for National Statistics

51

Population Trends 139

Spring 2010

-- no higher qualifications -- a low social class, or -- was unemployed It is worth considering that there will be some correlation between an individual’s sociodemographic characteristics and their partner’s. As such, the effect of some of these factors may be overstated and would be reduced by the inclusion of interaction effects.

Model 3 Model 3 is the same as Model 2, but excludes all adults who were married in 1991. In other words, it includes only those who were cohabiting in 1991. Notable results are as follows: • Individual factors that remain highly significant and increase the likelihood of stability are: -- being older -- the presence of dependent children -- no experience of previous marital dissolution -- economic activity also remains fairly significant with a relatively strong effect – being employed increases the likelihood of stability. • For partner’s characteristics, age difference and partner’s socio-economic activity remain highly significant. That is to say, being employed or self-employed, and having a smaller age difference increase the likelihood of stability. • Partly due to the smaller sample size, many of the factors reduce in magnitude and become far less significant (or insignificant). There is a large fall in the effect of whether a partner has a limiting long-term illness, as well as a reduction in significance. Previous marital status and social class of partner also cease to be significant. Model 3 aims to show which factors are associated with cohabitation stability, in isolation from marriage. A model for married adults only is not shown because it is very similar to Model 2. This is partly due to the far larger number of married adults in the Model 2 sample. This means that data for cohabitants has a smaller influence on Model 2. Apart from the overall reduction in significance for many of the variables, the odds ratios for cohabiting adults (Model 3) are not very different from those in Model 2. This suggests that the factors influencing cohabitation stability are somewhat similar to those influencing marital stability, particularly those that remain significant in Model 3.

Model 4 Model 4 is the same as Model 2, but excludes men. In other words, it includes only women who were married or cohabiting in 1991. Notable results are as follows: • Compared with women who did not have a baby between 1991 and 2001, those that did have a baby were more likely to remain with the same partner in 2001 • Despite the introduction of this new childbirth factor, and a slight fall in the significance of some factors, the model for women only is very similar to the model for both men and women – Model 2. As with the model for both sexes, women who were not economically active were more likely than either working women or unemployed women to be with the same partner in 2001

Office for National Statistics

52

Population Trends 139

Spring 2010

• Apart from a considerable reduction in the effect of partner’s limiting long-term illness, the main difference is for partner’s economic activity. Women whose partners were not economically active were less likely to remain with the same partner, compared with those whose partners were unemployed.

Further modelling of partnership outcomes There is limited space in this article to discuss further modelling that was undertaken. However, one additional question is: ‘what are the characteristics of cohabiting adults that go on to marry their partners?’. Table A4 (in the Appendix) shows the results of an additional model with the outcome: ‘Was the cohabiting adult in 1991 married to the same partner in 2001?’ The sample for this model was the same as Model 3 – all cohabiting adults in 1991. A preliminary model was run for this new outcome, with all the factors in Model 3 used as covariates. Categories that were not significant were then either removed, or combined with other categories in the same variable. The results are shown in Table A4. It is interesting to note the different factors that are associated with whether cohabiting adults marry their partner (between 1991 and 2001). They are more likely to marry if they or their partner have experienced previous marital dissolution. They are less likely to marry if they or their partner are unemployed, or if dependent children are present in the household in 1991. In addition, limiting long-term illness is not significant for either an individual or their partner. Compared with the previous models, this suggests that the presence of dependent children increases the likelihood of remaining with the same partner, but reduces the likelihood of cohabiting couples becoming married (between 1991 and 2001). Experience of previous marital dissolution has the opposite effect, reducing the likelihood of remaining with the same partner, but increasing the likelihood of cohabiting couples becoming married (between 1991 and 2001). This suggests that factors may act in different directions when considering different types of change in partnership status (for example. formation versus dissolution). In this case, and for this cohort, couples who have children and have not experienced marital dissolution may be more likely to be cohabiting as a substitute for marriage. There may of course be other reasons for this difference, and it should also be noted that cohabiting couples with children are different from married couples with children51.

Discussion This research provides an overview of long-term partnership stability between 1991 and 2001. It shows that marriage was more stable than cohabitation, even when controlling for a variety of factors. Despite this difference, the majority (61 per cent) of cohabiting adults aged 16 to 54 were living with the same partner in 2001. Of those 1991 cohabitants that were living with the same partner, two thirds had married this partner by 2001. This suggests, at least for those cohabiting in 1991, that cohabitation may be (or rather, may have been), more likely to be a precursor to marriage, rather than a substitute. However, this conclusion might change if those that cohabit as a substitute to marriage are (or were) less likely to remain with the same partner. Although the exact timing and order of events are beyond the scope of this study, the stability of partnerships between 1991 and 2001 is shown to be associated with both the presence of children in the household and the birth of a child. In addition, looking at cohabiting adults in isolation, it Office for National Statistics

53

Population Trends 139

Spring 2010

appears that social factors which are known to be associated with marital stability (for example age, economic activity and previous experience of partnership dissolution) are also associated with cohabitation stability. Further research is required to elaborate these conclusions, in particular to measure partnership transitions that occur both within and beyond a ten year period52.

Key Findings • Of adults aged between 16 and 54 in 1991, around four in five married adults (82 per cent) were still living with the same partner in 2001, compared with around three in five cohabiting adults (61 per cent). • Marital partnerships were found to be more stable than cohabitations, even when additional factors were taken into account. After controlling for the characteristics of both individuals and their partners, married adults were more likely than cohabiting adults to remain with the same partner between 1991 and 2001. • Adults were less likely to remain with the same partner if, in 1991, they were younger, had no dependent children living in the household, had a limiting long-term illness, had previous experience of partnership dissolution, had no higher qualifications, or were unemployed. • Partner’s characteristics also have an impact upon partnership stability. Adults were less likely to remain with the same partner in 2001 if, in 1991, their partner had a limiting long-term illness, had previous experience of partnership dissolution, had no higher qualifications, had a low social class, or was unemployed. • Compared with women who did not have a baby between 1991 and 2001, those that did have a baby were more likely to remain with the same partner in 2001.

Acknowledgements The authors would like to thank all those who commented on this article and all members of the LS team who provided assistance with this project.

Office for National Statistics

54

Population Trends 139

Spring 2010

Appendix Table A1 Whether enumerated in the 2001 Census by de facto status in 1991 All adults (aged 16+) enumerated in the 1991 Census, England and Wales Count Partnership status in 1991

In the Dead or LS sample embarked1 in 2001

Percentages Missing In the Dead or in 2001 LS sample embarked1 in 2001

Missing in 2001

Males Married and living with spouse Cohabiting Not living with a partner In a communal establishment Visitor All males

  93,373

17,859

9,917

77

15

8

9,344

521

1,989

79

4

17

46,088

8,399

12,010

69

13

18

1,192

1,087

734

40

36

24

3,481

707

1,076

66

13

20

153,478

28,573

25,726

74

14

12

100,719

10,297

9,470

84

9

8

Females Married and living with spouse Cohabiting

10,118

326

1,312

86

3

11

Not living with a partner

58,891

16,846

9,419

69

20

11

In a communal establishment

1,047

2,992

503

23

66

11

Visitor

3,807

994

730

69

18

13

174,582

31,455

21,434

77

14

9

328,060

60,028

47,160

75

14

11

All females All men and women

1 This category combines those who died between 1991 and 2001 and those who migrated (out of England Wales).It should be noted that only known migrants are in the embarked category. Some in the “missing in 2001” category will be undeclared migrants. Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

55

Population Trends 139

Table A2

Spring 2010

Whether enumerated in the 2001 Census by age in 1991

All adults (aged 16+) enumerated in the 1991 Census, England and Wales Count Partnership status in 1991

In the Dead or LS sample embarked1 in 2001

Percentages Missing In the Dead or in 2001 LS sample embarked1 in 2001

Missing in 2001

16–34

 

Married and living with spouse

45,084

655

6,197

87

1

12

Cohabiting

13,025

213

2,316

84

1

15

Not living with a partner

60,543

1,150

14,756

79

2

19

1,160

64

745

59

3

38

In a communal establishment Visitor All adults aged 16–34

4,591

108

1,365

76

2

23

124,403

2,190

25,379

82

1

17

111,292

5,822

9,979

88

5

8

35–59 Married and living with spouse Cohabiting Not living with a partner In a communal establishment Visitor All adults aged 35–59

5,892

283

894

83

4

13

24,951

2,283

4,022

80

7

13

655

147

275

61

14

26

1,534

166

281

77

8

14

144,324

8,701

15,451

86

5

9

37,716

21,679

3,211

60

35

5

545

351

91

55

36

9

19,485

21,812

2,651

44

50

6

424

3,868

217

9

86

5

60+ Married and living with spouse Cohabiting Not living with a partner In a communal establishment Visitor All adults aged 60+ All adults 16+

1,163

1,427

160

42

52

6

59,333

49,137

6,330

52

43

6

328,060

60,028

47,160

75

14

11

1 This category combines those who died between 1991 and 2001 and those who migrated (out of England Wales). It should be noted that only known migrants are in the embarked category. Some in the “missing in 2001” category will be undeclared migrants. Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

56

Population Trends 139

Spring 2010

Table A3a Partnership status by sex, percentage in each age group in 1991 Longitudinal sample, England and Wales 16–24

25–34

35–49

50–59

Total (16–59)

73

27

18

21

32

12

11

5

2

7

Women Lone Cohabiting Married1 All women Men Lone Cohabiting Married1 All men

14

62

77

77

60

100

100

100

100

100

86

33

16

14

33

8

12

6

3

7

7

55

78

83

59

100

100

100

100

100

1 Married and living with spouse. Source: ONS Longitudinal Study (authors’ analysis)

Table A3b Partnership status by sex, percentage in each age group in 1990/91 Cross-sectional sample, Great Britain 16–24

25–34

35–49

50–59

Total (16–59)

Lone

70

26

18

21

31

Cohabiting

14

10

5

2

7

Married1

16

64

77

77

62

100

100

100

100

100

86

30

16

16

33

7

12

5

2

7

Women

All women Men Lone Cohabiting Married

1

All men

7

58

79

82

60

100

100

100

100

100

1 Married and living with spouse. Source: General Household Survey (GHS); 1990 and 1991 combined

Office for National Statistics

57

Population Trends 139

Spring 2010

Table A4 Probability of being married to same partner in 2001 (if cohabiting in 1991) Longitudinal sample, England and Wales, All cohabiting adults (aged 16–54) in 1991 Variable

Odds ratio1

Significance level2

Age in 1991

0.98

***

Age gap in absolute years

0.97

***

Female (ref.)

1.00

n/a

Male

1.18

***

No (ref.)

1.00

n/a

Yes

0.84

***

Single or widowed or married (ref.)

1.00

n/a

Remarried or divorced (or married if cohabiting)

1.14

***

No qualifications after age 18 (ref.)

1.00

n/a

Has qualifications after age 18

1.14

***

Professional, managerial, technical or skilled non-manual

1.18

***

Skilled manual, part-skilled, unskilled, other (ref.)

1.00

n/a

Unemployed (ref.)

1.00

n/a

Not economically active

1.19

**

Self-employed

1.28

***

Employed

1.43

***

Single or widowed or married (ref.)

1.00

n/a

Remarried or divorced (or married if cohabiting)

1.15

***

No qualifications after age 18 (ref.)

1.00

n/a

Has qualifications after age 18

1.17

***

Sex

Dependent children in household in 1991

Previous dissolution (marital status in 1991)

Qualifications after age 18 (in 1991)

Social class (Registrar General’s) in 1991

Economic activity in 1991

Partner: previous dissolution (marital status in 1991)

Partner: qualifications after age 18 (in 1991)

Partner: social class (Registrar General’s) in 1991 Professional, managerial, technical or skilled non-manual

1.17

***

Skilled manual, part-skilled, unskilled, other (ref.)

1.00

n/a

Unemployed (ref.)

1.00

n/a

Not economically active

1.40

***

Self-employed

1.58

***

Employed

1.74

***

Partner: economic activity in 1991

Note: For Registrar General’s social class, other includes armed forces and missing. 1 Reference categories are shown with an odds ratio of 1.00. 2 * significant at the 10% level; ** significant at the 5% level; *** significant at the 1% level. n/a = reference category (significance is not applicable). Source: ONS Longitudinal Study (authors’ analysis)

Office for National Statistics

58

Population Trends 139

Spring 2010

References 1 This rise is often attributed to changing legislation (the Divorce Reform Act 1969 and Matrimonial Causes Act 1973) and changing attitudes in society. Considering the long-term trend, and ignoring minor fluctuations, this increase can be seen as a step-change. For more information see: Smallwood S & Wilson B (2008) ‘The proportion of marriages ending in divorce’, Population Trends 131, pp. 28–36. Available at: www.statistics.gov.uk/downloads/theme_population/Population_Trends_131_web.pdf 2 In 2007 the provisional divorce rate in England and Wales fell to 11.9 divorcing people per 1,000 married population, compared with the 2006 figure of 12.2. The divorce rate is at its lowest level since 1981. See also: www.statistics.gov.uk/cci/nugget.asp?id=170 3 For example see: Smallwood S & Wilson B (2007) ‘Understanding recent trends in marriage’. Population Trends 128, pp. 24–32. Available at: www.statistics.gov.uk /downloads/theme_population/ PopulationTrends128.pdf and Wilson B (2009) ‘Estimating the cohabiting population’. Population Trends 136, pp. 21–27. Available at: www.statistics.gov.uk/ downloads/theme_population/Popular-Trends136.pdf 4 Both figures are from the GHS. For 1986 results see: OCPS (1989) General Household Survey 1986 (Series GHS no.16), London: HMSO. For 2007 results, see: Results from the General Household Survey (GHS), 2007 (Table 5) available at: www.statistics.gov.uk/StatBase/ Product.asp?vlnk=5756&Pos=&ColRank=1&Rank=256 5 In addition there has been a long-term increase in adults living alone, and an increase in lone parent families. For more information see Social Trends 39: Chapter 2 Households and families, available at: www.statistics.gov.uk/downloads/theme_social/Social_Trends39/ST39_Ch02.pdf 6 Office for National Statistics (2007) 2006-based marital status projections. Available at: www.statistics.gov.uk/pdfdir/marr0309.pdf 7 The proportion of adults who have never married is projected to rise from 34 per cent to 42 per cent. It should be noted that some of these will be cohabiting. Therefore there is an overlap with the projected numbers of cohabitants. 8 Although there is good information on marriage and divorce, statistics on the formation and dissolution of cohabiting partnerships are not collected routinely. In order to consider partnership stability adequately, it is desirable to have comparative information on partnership transitions. These transitions are important because they go beyond stock estimates at a given time point, to suggest how (and why) partnership estimates change over time. In some respects, this can be considered equivalent to the importance of births, deaths and migration when considering changes in the population. (Of course, mortality and migration may also change an individual’s partnership status.) 9 For example, see the report published to Parliament by the Law Commission on 31 July 2007. Available at: www.lawcom.gov.uk/cohabitation.htm 10 Wilson B (2009) ‘Estimating the cohabiting population’. Population Trends 136, pp. 21–27. Available at: www.statistics.gov.uk/downloads/theme_population/Popular-Trends136.pdf

Office for National Statistics

59

Population Trends 139

Spring 2010

11 Dunnell K (2008) ‘Ageing and Mortality in the UK – National Statistician’s Annual Article on the Population’. Population Trends 134, pp. 6–23. Available at: www.statistics.gov.uk/downloads/ theme_population/Population-Trends-134.pdf 12 As far as legal status is concerned, a marriage ends in either death or divorce, however it is also important to note that couples often separate prior to divorce (that is there is a residential dissolution prior to the legal decree). Separated individuals may therefore begin to cohabit with a new partner prior to divorce (which is one of several explanations why a married couple might not be living together). 13 Although marriage and divorce statistics have been collected by the registration system (and the courts) for over a century, there is currently no requirement for cohabiting couples to register the formation or dissolution of their partnerships. As such, there are limited sources of information on partnership transitions. It is not possible to use simple cross-sectional surveys because we need to explore changes in individual partnerships over time. 14 It is also important that cohabitation can be distinguished from simply sharing accommodation. In addition, any attempt to identify cohabitants can be affected by misreporting. For example, prevailing social attitudes have (at least in the past) attached a stigma to cohabitation. 15 The coverage of topics has been developed and extended over the years: initially in 1971 a few questions were addressed to women aged between 18 and 44; additional subjects – including cohabitation – were introduced in 1979; and the age range was extended, firstly going up to age 49, and then from 16 to 59 in 1986, when men were first asked questions on cohabitation. For more information (and the source of the previous sentence) see: Haskey J (2001) ‘Cohabitation in Great Britain: past, present and future trends – and attitudes’, Population Trends 103, pp. 4–25. 16 Haskey J (2001) ‘Cohabitation in Great Britain: past, present and future trends – and attitudes’. Population Trends 103, TSO London, pp. 4–25. 17 Haskey J (1999) ‘Cohabitational and marital histories of adults in Great Britain’. Population Trends 96, TSO London, pp. 13–24. 18 Haskey J (2001) ‘Cohabiting couples in Great Britain: accommodation sharing, tenure and property ownership’. Population Trends 103, TSO London, pp. 26–36. 19 Murphy M (2000) ‘The evolution of cohabitation in Britain, 1960–95’. Population Studies 54(1), pp. 43–56. 20 Lilly R (2000) ‘Developing questions on cohabitation histories for the General Household Survey’. Survey Methodology Bulletin 46 (January), ONS, pp. 15–22. Available at: www.statistics.gov.uk/ssd/ssmb/smb_46.pdf 21 Berrington A and Diamond I (2000) ‘Marriage or cohabitation: a competing risks analysis of first-partnership formation among the 1958 British birth cohort’. Journal of the Royal Statistical Society: Series A (Statistics in Society) 163(2), pp. 127–151. 22 Calderwood L (2008) Chapter Three: Family Demographics. Millennium Cohort Study Third Survey: A User’s Guide to Initial Findings, by Hansen K & Joshi H (eds.), Centre for Longitudinal Studies, Institute of Education, University of London, pp. 22–50.

Office for National Statistics

60

Population Trends 139

Spring 2010

23 Ermisch J and Francesconi M (2000) ‘Cohabitation in Great Britain: Not for Long, but Here to Stay’. Journal of the Royal Statistical Society: Series A (Statistics in Society) 163(2), pp. 153–171. 24 Buck N and Ermisch J (1995) ‘Cohabitation in Britain’, in Changing Britain: Newsletter of the ESRC Population and Household Change Research Programme 3, pp. 3–5, October 1995. 25 Clarke L and Buxton J (2006) ‘Cohabitation: Changes over the 1990s and longitudinal evidence on transitions in status’. Presentation at 2006 BSPS Annual Conference. 26 CeLSIUS (2008) Downloadable tables from the ONS Longitudinal Study. Available at: www.celsius.lshtm.ac.uk/download/wt020400.html 27 Wolf DA and Gill TM (2009) ‘Modelling transition rates using panel current-status data: How serious is the bias?’ Demography 46(2), May 2009: pp. 371–386. 28 Essentially, this was all adults in the LS that were both present in 1991, and aged 16 or over in 1991. 29 The date of extraction for the sample was June 2009 (LSLOAD62). 30 Embarkation is only flagged when an individual notifies their GP. 31 For deaths by marital status see DR Table 4 (ONS), available at: www.statistics.gov.uk/downloads/ theme_health/DR2007/DR_07_2007.pdf 32 No direct question was asked about cohabitation in the 1991 Census, although marital status was asked. This means that a cohabiting partnership involving an LS member must be identified using the relationship questions on the census form. Because only relationship to the head of household was collected in 1991, in complex households or where the LS member is not the head of household some partnerships are likely to have been missed. Moreover, for people who were enumerated at an address which was not their usual place of residence, marital status will be known but whether they were cohabiting will not be known. 33 Strictly speaking, the Minimal Household Unit (MHU), which is a subdivision of the Census category ‘family’. A MHU comprises either an unmarried individual, or a lone parent with his/her dependent children, or a couple (married or cohabiting) with their dependent children. 34 Being an adult present at both censuses is the criterion for inclusion in the sample. As such, there will be no sample members in 2001 aged between 16 and 25 (since they are under 16 in 1991). 35 OPCS (1993) General Household Survey 1991 (Series GHS no. 22), HMSO London. 36 For a distribution of age differences at marriage see: Wilson B and Smallwood S (2008) ‘Age differences at marriage and divorce’. Population Trends 132, pp.17–25, available at: www.statistics.gov.uk/downloads/theme_population/Population_trends_132.pdf 37 For an example with recent results see: Smallwood S and Wilson B (2008) ‘The proportion of marriages ending in divorce’. Population Trends 131, pp. 28–34, available at: www.statistics. gov.uk/downloads/theme_population/Population_Trends_131_web.pdf 38 Clarke L and Berrington A (1999) ‘Socio-demographic predictors of divorce’. Published in: Simons J (ed.) High divorce rates: The state of the evidence on reasons and remedies: Reviews of the evidence on the causes of marital breakdown and the effectiveness of policies Office for National Statistics

61

Population Trends 139

Spring 2010

and services intended to reduce its incidence. (Lord Chancellor’s Department Research Series, 1 2/99) London. 39 As the initial LS sample ages, it will be increasingly possible to explore the influence of parental characteristics. 40 Although the LS contains information on religion, it was decided not to include this because information was only available in 2001 and even then the question was not compulsory. 41 Many models were created to test partnership stability, but the four most important are shown in this article. 42 See: Online training module for users of the ONS Longitudinal Study. The logistic regression example starts at the below link. Follow links at the bottom of the page to continue the example. Use: www.celsius.lshtm.ac.uk/modules/analysis/an030200.html 43 It is worth noting that in 1991, only information on degree and professional qualifications was collected, not information on school qualifications. 44 Adjusted for ‘normal’ age difference so that zero represents a man two years older than his female partner. 45 The LS is not a household based sample, which means that non-response is at the individual, rather than the household level. It was therefore deemed important to consider differences by sex, which might link to any non-response issues. 46 Table 5 (which looks at a gross relationship) does not hold any other factors constant when considering stability and age. In fact, Table 5 does not consider the influence of any factors other than age. When interpreting both statistics, it is important to remember that neither one is more accurate, but that they each offer a different perspective on the same results. For more information see: Murphy M (1985) ‘Demographic and socio-economic influences on recent British marital breakdown patterns’. Population Studies 39, 441–460 as cited in Clarke L and Berrington A (1999). 47 Alternatively, those without a limiting long-term illness in 1991 are 24 per cent more likely to remain with the same partner between 1991 and 2001 compared with those who have a limiting long-term illness in 1991, all other factors being equal. The last part of this statement (all other factors being equal) means that the effect of limiting long-term illness on partnership stability (for this sample) has been shown controlling for all the other factors in the model (age, qualifications etc). It is important to note that any factors not in the model are not considered. As such, any variations in stability by limiting long-term illness may be explained by these (exogenous) excluded factors. 48 For example, in Model 1 the odds ratio for age difference is 0.95. This means that for every additional year of absolute age difference between partners, the odds of remaining with the same partner between 1991 and 2001 are 0.95 (or 5 per cent lower). Absolute age difference is the total age difference irrespective of which partner is older. 49 Some of the difference between men and women will reflect the typical partnership age gap where the man is on average 2 to 3 years older than the woman. Some older men will therefore fall above the 16–54 age range when women in an equivalent partnership will not. However, the effect of age difference was investigated and found to explain only a minority of the difference between men and women. Office for National Statistics

62

Population Trends 139

Spring 2010

50 In particular, the odds ratio for the managerial or technical class ceases to be either material or significant. 51 A number of selection effects might be considered here, and further research would be required in order to draw more definitive conclusions. 52 For example, further research is needed to explore the effect of partners that separate and then reform their partnership with the same person (including those that are married and not living together at any given point).

Office for National Statistics

63

Population Trends 139

Spring 2010

Households and families: Implications of changing census definitions for analyses using the ONS Longitudinal Study Emily Grundy, Rachel Stuchbury and Harriet Young Centre for Longitudinal Study Information and User Support (CeLSIUS), London School of Hygiene & Tropical Medicine

Abstract The ONS Longitudinal Study (LS) includes information from the 1971, 1981, 1991 and 2011 censuses. This article explains definitional differences over time, and their implications for household and family classifications.

Office for National Statistics

64

Population Trends 139

Spring 2010

Contents Abstract............................................................................................................................................ 64 Introduction....................................................................................................................................... 66 What is a child?................................................................................................................................ 66 Family definitions.............................................................................................................................. 66 The impact of changes in 2001........................................................................................................ 67 References and key publications..................................................................................................... 68

List of tables Table 1 Distribution of ONS Longitudinal Study members by family/household type in 2001 using the 2001 and 1991 (and earlier) definitions of a child....................................... 68

Office for National Statistics

65

Population Trends 139

Spring 2010

Introduction The strengths of the ONS Longitudinal Study (LS) of England and Wales include the fact that information on all co-residents of LS sample members is available at each census point, together with information on family and household type (more information on family and household definitions and classifications is available in the CeLSIUS training module on households and 1 families) . The LS is thus a valuable resource for those interested in, for example, changes over the life course in the types of household people live in and those who want to compare distributions of household and family types at different time points in order to investigate period changes. For researchers interested in either approach, consistency of definitions is important. The ONS LS now includes information from the 1971, 1981, 1991 and 2001 Censuses. Changes 2 in definitions between the first three of these censuses were relatively minor , but in 2001 there was a more substantial change arising from a revised definition of a child. In this paper we explain this difference, its implications for household and family classifications, and offer a link to code (in STATA) which can be used by those wanting consistent definitions over time.

What is a child? In the 2001 UK Census a child was defined as an individual of any age or marital status, not themselves part of a co-residing couple or a parent, grand-parent or step-parent of anyone else in the household, who lived with one or both of their own parents. This differed from the definition used in previous censuses in which a child had to be never-married, as well as meeting the other criteria specified above. This change affects a number of the classifications that researchers may use, such as the statistical definition of a family. Prior to 2001, ONS defined a family as either a co-resident couple; a couple and never-married child(ren); a lone parent and never-married child(ren); or a grandparent and never-married child(ren) if the intervening generation was absent. Households refer to co-resident groups sharing common living space, or at least one meal a day, and may include one or more families, or none. In 2001 the change in the definition of a child meant that the definition of a family also changed, as did descriptions and definitions of households based on the families within them. In 1991 for example, a widowed mother and divorced daughter living together with no-one else would not have been classed as a family, and their household would not have been described as a family household. In 2001 however, the same two people would have been classed as a lone-parent family and their household as a lone-parent household. This change presents difficulties for those undertaking longitudinal analyses who may want to analyse changes in household and family status over census points, or for those who are interested in looking at period changes between different censuses. This article investigates and quantifies the impact of this change.

Family definitions In previous research on households, we have derived and used a variable describing the 3,4,5,6 household and family circumstances of LS members at the 1971, 1981 and 1991 Censuses . This takes into account the position of the LS member in the family and household in which they live and relationships with other family and household members. The variable, which we have named ‘housefam’, includes the following categories: living alone; couple only; couple and children; Office for National Statistics

66

Population Trends 139

Spring 2010

couple and others; couple and children and others; lone parent; lone parent and others; two or more families; not in a family but living with others; child (including adult children) in family; and living in a communal establishment (although in most previous work a collapsed version of this has been used). Constructing a 2001 version of this using the 2001 definitions of child and family will produce slightly different results from equivalent analyses using the earlier definition. For example, in 2001 a divorced female LS member living with her parents would have been classed as a ‘child in family’, where previously she would have been classed as ‘not in a family but living with others’. If the LS member was the mother of the divorced daughter in the same configuration in 2001 she would be classed as living in a ‘couple and child’ family/household, but in 1991 or earlier, as living in a ‘couple and others’ family/household. Fortunately the LS includes information on all those in sample members’ households and on intra-family and intra-household relationships, including in 2001 a full relationship grid. It is therefore possible to produce classifications for 2001 using the old (pre 2001) rather than the new child definition. Details of the algorithms, and relevant code 7 (in STATA) for doing this are available on the CeLSIUS web site

The impact of changes in 2001 The table below shows the distribution of LS members in 2001 by family/household type using alternative definitions of ‘housefam’ based on either the 2001 or the earlier definition of a child. In each case, if there is any imputed relationship in the household, the family/household type has been set to ‘unclassifiable’. Using the wider 2001 definition of a child obviously results in the numbers in categories which include a child being larger than when using the more restrictive 1991 definition. For example using the 2001 definition, 3.5 per cent of LS members were classed as living in a lone parent family, compared with 3.3 per cent using the earlier definition. This difference may seem slight, but differences in the numbers concerned are considerable, given that the LS is a one per cent sample of the population. It is recommended that the impact of these changes be considered when making any comparison between the 2001 Census and previous censuses for statistics relating to households, families and children.

Office for National Statistics

67

Population Trends 139

Spring 2010

Table 1 Distribution of ONS Longitudinal Study members by family/ household type in 2001 using the 2001 and 1991 (and earlier) definitions of a child 2001 definition

1991 definition

Number

%

Number

%

65,033

12.22

65,053

12.22

Couple only

120,830

22.70

120,830

22.70

Couple and children

107,386

20.17

106,095

19.93

3,721

0.70

5,098

0.96

Solitary

Couple and others Couple and child and others

4,658

0.88

5,207

0.98

18,495

3.47

17,725

3.33

Lone parent and others

1,939

0.36

2,058

0.39

2 or more families

4,284

0.80

4,326

0.81

Lone parent

Not in a family, with others Child in family (including adult children) Communal establishment Unclassifiable Total

10,850

2.04

12,270

2.31

127,400

23.93

125,954

23.66

7,922

1.49

7,922

1.49

59,773

11.23

59,773

11.23

532,311

100

532,311

100

Notes: 1 The definition of a family includes grandparent(s) living with a grandchild whose parents are not resident in the same household. We did not account for these families when making the new housefam variable. There were only approximately 40 such households who had an ever-married child in the same family, and so a decision was made to leave these families in their original categories. 2 We found that there were approximately 900 never-married children not in the same family as the LS member. We left these as they were, and assumed that they are likely to be never-married children with children of their own, who are therefore part of a separate family. 3 The table above excludes imputed values in the source variables. We have also derived versions of the housefam variables using imputed values. For further information on these and on general derivation of these variables, please 7 see our information pages on derived variables .

References and key publications 1 CeLSIUS (2009) CeLSIUS training module on households and families. Available at: www.jcelsius.lshtm.ac.uk/modules/hhfam/hf010000.html 2 Brasset-Grundy A (2003) ‘Researching households and families using the ONS Longitudinal Study’. LS User Guide 20. Institute of Education, University of London, London. Available at: www.celsius.lshtm.ac.uk/documents/userguide20.pdf 3 Grundy E (1987) ‘Household change and migration among the elderly in England and Wales’. Espace, Populations, Sociétés 1, 109–123. 4 Grundy E (1999) Household and family change in mid and later life in England and Wales. Published in McRae S (ed.) Changing Britain: Families and Households in the 1990s. Oxford University Press, Oxford. 5 Glaser K and Grundy E (1998) ‘Migration and household change in the population aged 65 and over, 1971–1991’. International Journal of Population Geography 4, 323–339.

Office for National Statistics

68

Population Trends 139

Spring 2010

6 Grundy E and Jitlal M (2007) ‘Socio-demographic variations in moves to institutional care 1991–2001’. Age and Ageing 36(4), 424–430. 7 CeLSIUS (2009) Derived variables: household composition. Available at: www.celsiusdev.lshtm.ac.uk/private/forclearance/derive/hhcomp.html

Office for National Statistics

69

Population Trends 139

Spring 2010

Ten year transitions in children’s experience of living in a workless household: variations by ethnic group Lucinda Platt Institute for Social and Economic Research, University of Essex

Abstract Over the last few decades, there has been an increase in the proportion of children growing up in workless households, that is households in which no adult member is in paid work. This proportion has stabilised, and has declined slightly in recent years. Worklessness among households with children is viewed as a cause for concern for two reasons: firstly, because children in workless households are much more likely to be growing up in poverty; secondly, because of concern that worklessness in families with children may be subject to intergenerational transmission. We know surprisingly little about children’s experience of household worklessness over time, particularly over their childhood as a whole, even though worklessness is heavily implicated in higher poverty risks. Children from most minority ethnic groups are at substantially higher risk of household worklessness than those from the majority. For some ethnic groups, children’s rates of living in a workless household are associated with high rates of lone parenthood. For others it is worklessness in couple parent families that predominates. This article uses the Office for National Statistics (ONS) Longitudinal Study to explore differences in risks of worklessness over time, among ethnic groups within a single cohort of children who are observed at two time points, 10 years apart.

Office for National Statistics

70

Population Trends 139

Spring 2010

Contents Abstract............................................................................................................................................ 70 Introduction....................................................................................................................................... 72 Data and study design...................................................................................................................... 77 Results............................................................................................................................................. 80 Discussion........................................................................................................................................ 85 Acknowledgments............................................................................................................................ 86 Appendix.......................................................................................................................................... 87 References....................................................................................................................................... 87

List of figures Figure 1 Proportions of children in workless households at 1991 (aged 0–5 years) and 2001 (aged 10–15 years) by ethnic group, England and Wales................................. 80 Figure 2 Movers and stayers, children in workless households 1991–2001, by ethnic group....................................................................................... 81 Figure A1

Employment status by gender and ethnicity (percentages)........................................ 87

List of tables Table 1 Recent estimates of proportions of children living in a workless household and living in a poor household by ethnic group (percentages).......................................... 74 Table 2 Children in lone parent families by age group and ethnic group; risk of living in a workless household for children in lone parent family................................................ 75 Table 3 Children aged 0–5 years in 1991 and observed aged 10–15 years in 2001 by ethnic group, England and Wales............................................................................... 78 Table 4 Relative chances of being in a workless households in 2001 conditional on 1991 workless household status, by ethnic group............................................................... 83

Office for National Statistics

71

Population Trends 139

Spring 2010

Introduction Children living in workless households The last decade has seen a growing research and policy interest in workless households, that is 1,2,3 households where no one of working age is in work. A particular concern has been the welfare 4,5,6 and future prospects for children in such households. Attention has focussed on the differential 5 risks of living in a workless household faced by children from different ethnic groups. We still know little about how worklessness is experienced over time, and how that may or may not differ by ethnicity. This is of particular concern, since children from many minority ethnic groups are at 7 relatively high risk of living in a workless household and of the poverty stemming from that. This article focuses on a particular cohort of children, born around the end of the 1980s, and investigates the currently unexplored question of whether risks of remaining in or moving into a workless household during their childhood are comparable for children from different ethnic groups. It describes absolute differences in risks of remaining in or entering a workless household across groups, and examines the extent to which any differences are mediated by household structure and characteristics. For the purposes of this article, a workless household is defined as one where 8 no adult member is in work. The increase of work work-rich and work work-poor households has been well documented. 3 Gregg and Wadsworth have shown how the share of workless households increased over the last two decades of the 20th century with some levelling off by 2001, and that this was the case for households with children as well as for all households. Indeed, in 1996 the UK had the highest proportion of workless households with children in the member countries of the Organisation for Economic Cooperation and Development (OECD). At the same time there has been a longstanding interest in, and concern with, the potential transmission of various forms of economic disadvantage 9 between generations, and with how children’s experience of such disadvantage, including worklessness, can have long term impacts. Children’s risks from living in a workless household tend to be higher when they are younger and 10,11 tends to decline with age, but experience of a workless household can have negative consequences at any age, and growing up in a persistently workless household is likely to be particularly detrimental to future outcomes. Children living in workless households face very high risks of living in poverty, and the associations 12 13 between childhood poverty and future outcomes, as well as their development, are well attested. Moreover, persistent poverty both tends to represent more severe poverty and to be associated 14 with more negative outcomes than short-term or transient poverty. Thus, to the extent that it implies long-term poverty, long-term worklessness is likely to be of particular concern in relation to children’s welfare. The timing can also have implications for children’s later life outcomes. Poorer outcomes associated with poverty and worklessness tend to be greater for younger rather than 10, 11 and it is also at younger ages that the risks of living in a workless household are older children 11 greatest. Nevertheless, Ermisch et al. have shown that the experience of worklessness in later childhood (11–15 years) is associated with increased chances of smoking and of psychological distress. While there has been substantial emphasis on the detrimental effects of long-term poverty or worklessness, increasing attention is being paid to the negative impacts of socio-economic 15,16,17,18 Instability provides the opportunity for periods of relative advantage instability in its own right. Office for National Statistics

72

Population Trends 139

Spring 2010

compared to remaining persistently below a given poverty threshold such as 60 per cent of median 19 equivalent household income, as used in UK low income statistics. However, those who fluctuate between states are likely to be in more marginal positions – on the borders of poverty or on the 16 margins of work , while the actual variation in circumstances may introduce its own costs, such as uncertainty, the need to reclaim benefits with the consequent possibility of periods without any support, the need to change arrangements for care, and so on. Therefore vulnerability to poverty, as evidenced by subsequent moves into worklessness, raises concerns for family welfare. In addition to the welfare implications of growing up in a workless household, the experience or persistence of worklessness among families with children, which typically implies a need for support by state benefits, has raised concerns about the extent to which there is intergenerational 9 transmission of worklessness and benefit dependence. Evidence from the US provides supporting 21 evidence for intergenerational transmission of ‘welfare’ participation, over and above income effects, though the exact mechanisms are not clearly understood. Both lack of role models and limited access to networks and opportunities for pursuing employment – or a combination of these – have been offered as explanations for intergenerational transmission. There are good reasons for being concerned about children’s rates of living in a workless household, their risks of being persistently in a workless household and their vulnerability to ending up in a workless household from a working household. However, there is little understanding of the extent to which these risks differ for children according to their ethnicity, and the factors implicated in differential risks. This is despite the fact that it is well known that there are higher risks of poverty for children from certain ethnic groups, and that the risks of being in a workless household also vary substantially by ethnicity, as the next section discusses. Although there are some indications 22 of differences in persistence and instability in economic circumstances across ethnic groups, we have little understanding of how transitions into workless households, or persistence in growing up in a workless household across childhood over an extended period, varies by ethnic group. The contribution of this article is to explore precisely these questions for a particular cohort of children of the same age and over the same period.

Ethnicity and worklessness Individual employment rates are well known to vary by ethnicity, particularly for women, with high rates of inactivity among Pakistani and Bangladeshi women, relatively high levels of participation among Black Caribbean women, relatively high rates of inactivity (compared to other men) among Bangladeshi men and above average unemployment risks for all minority groups (see Appendix, Figure A1). A substantial body of research has shown that some minority groups are disadvantaged in the labour market, even taking account of variation in qualifications and other 23 job-relevant characteristics . However, far less is known about the duration of unemployment or worklessness across ethnic groups even at the individual level; nor do individual levels of 24 employment – analyses of which abound – tell us about the same phenomenon as household 3 experience of work and non-work. Cross-sectional analyses indicate that there are clear differences in workless household rates by ethnic group; we also know that there are substantially higher rates of cross-sectional child poverty among all minority groups compared to the majority. Table 1 shows that for most recent estimates, children’s risks of living in a workless household were particularly high for Black African children and lowest for Indian children. There are substantial differences between the groups and the rates

Office for National Statistics

73

Population Trends 139

Spring 2010

Table 1 Recent estimates of proportions of children living in a workless household and living in a poor household by ethnic group (percentages) Children’s household workless (per cent) by ethnic group of child

Children’s household poverty (per cent) by ethnic group of head of household

2009

2001

2004/05–2006/07

2001/02–2003/04

White

15.4

14.8

20

20

Mixed

27.9

28.6





Indian

8.6

10.4

27

28

25.1

34.8

54/58

59/72

Black Caribbean

30

27.4

26

31

Black African

42

50.1

35

38

Other Black

26.8

36.8





Chinese

13.9

23.4





Pakistani/Bangladeshi

Notes: ‘–’ = figures not available due to small sample sizes. Children refers too children aged under 16. Sources: Column 1: ONS Statistical Bulletin ‘Work and worklessness among households 2009’, Table 3(iv) (from Labour Force Survey) UK data; Column 2: Platt 2009, Table 2.1 (from Family Resources Survey), data for Great Britain

for minority groups are significantly different from those for the White majority. Given that, in this article, rates of living in a workless household are considered for 1991 and 2001, Table 1 also illustrates the rates that pertained across groups in 2001. While the rates are rather different for some groups at the earlier period, the ranking is very similar for the two time points, with the major change being the reversal of the relative positions of Black Caribbean and Pakistani/Bangladeshi children. It can also be seen from the right hand panel of Table 1 that the ranking of workless household risks does not map precisely onto poverty rates. Worklessness is of concern in part because it 6 brings high risks of poverty, but poverty is not fully accounted for by worklessness. As Nickell pointed out in his discussion of children and workless households, 53 per cent of poor children lived in workless households in 2000/01, and those living in workless households had a 70 per cent chance of being poor. However, there is not a complete overlap. Nevertheless, worklessness may have implications for future welfare over and above the material deprivation that it is likely to bring. It is known that family structure varies substantially between groups. For example, Black Caribbean and Black African children experience high rates of lone parenthood, and children from South Asian groups are much less likely to live in a lone parent family (see Table 2). The trend with age is towards higher risks of living in a lone parent family, but this is counteracted by the greater likelihood of lone parents with older children being in work. Recent policy changes are intended 25 to enhance this pattern. Since we know that family structure, in particular lone parenthood, is heavily implicated in risks of worklessness, we might therefore expect that such variations in family structure would influence absolute risks of worklessness, despite the greater propensity of Black Caribbean lone parents to be in employment compared to other lone parents. This is reflected in Table 2, where children’s risks of living in a workless household, given that they are in a lone parent family, is shown.

Office for National Statistics

74

Population Trends 139

Spring 2010

Table 2 Children in lone parent families by age group and ethnic group; risk of living in a workless household for children in lone parent family Ethnic group

Percentage in lone parent family All children [C.I.]

Age 0–5 [C. I.]

Age 10–15 [C. I.]

Risk of household worklessness in lone parent family: percent [C. I.]

White British

24 [24–25]

21 [21–22]

27 [26–27]

45 [44–46]

37,362

White & Black Caribbean

52 [44–61]





58 [47–69]

141

White & Asian

16 [11–22]







166

10 [8–12]

8 [5–11]

12 [9–17]

51 [41–61]

887

16 [14–19]

11 [8–14]

21 [16–27]

62 [52–70]

923

11 [8–15]

7 [4–12]





417

Black Caribbean

56 [52–60]

54 [47–62]

58 [52–66]

39 [34–45]

583

Black African

46 [43–50]

40 [34–46]

51 [45–58

65 [59–70]

760

Indian Pakistani Bangladeshi

Number

Note: Figures are weighted. ‘–’ indicates that sample sizes are too small to allow for reliable estimates. Source: Family Resources Survey 2001/02–2006/07, pooled. Author’s analysis

Whether family structure does account for differences in overall risks of worklessness over time, when considering children of a comparable age and a common cohort, is a question addressed by this article. There are also variations in average family size according to ethnic group, with Pakistani and Bangladeshi families in particular having larger family sizes on average. Additional children may make moves out of worklessness more difficult, both as a result of the demands they make on parental time and as a result of the structure of benefits and the impact on marginal tax rates. On the other hand, as children grow up they may provide additional sources of labour market income 7 for families as they remain in the home. It is therefore not clear what the different chances are likely to be of moving into, or remaining in, a workless household over the childhood years, for different groups. What neither Table 1 nor Table 2 reveals are the risks of worklessness for children of particular 26 ages and family circumstances, nor is there information on risks of long-term worklessness, or the nature of transitions to and from worklessness. These are addressed in this article.

Aims of current analysis This article sets out to map the patterns of children’s workless household transitions, for children from different ethnic groups. It asks: • What are the differences in risks of worklessness for a single cohort of children according to their ethnic group? Office for National Statistics

75

Population Trends 139

Spring 2010

• What are the chances of remaining in a workless household (persistence) or ending up in a workless household after having been in a working household (entry), for children from different ethnic groups? • To what extent are experiences of workless household persistence and entry significantly different for children from particular minority groups, compared to White majority children? • To what extent are such differences mediated by family and household context? • And conversely, to what extent do greater risks appear to exist over and above the contribution of relevant household and family characteristics? The analysis is motivated by the implications raised, by the differential chances of remaining workless, for children from minority ethnic groups. In absolute terms, any differences in vulnerability to remaining in or entering a workless household may have implications for the future wellbeing of children from those groups, and therefore merits attention. Understanding the role of family and household characteristics can inform and reinforce strategies to address these areas. If those differences are largely mediated by household and family characteristics, such as the emphasis on moving lone parents into work, then there is less argument that policy should be differentiated to address the risks of different groups. Conversely, if there appear to be ethnic differences in children’s risks of staying in or moving into a workless household even after taking account of relevant family and household characteristics, 27 then such ethnic penalties in children’s risks of worklessness require further explanation, and possibly targeted intervention. It should be noted however, that the extent to which the long-term impacts of worklessness are themselves comparable across ethnic groups, including transmission of deprivation, is as yet untested and is an area for future research. The following analysis explores transitions into and out of workless households over a ten year interval by ethnic group, using a unique data set, the ONS Longitudinal Study (see Box one). It examines the risks of living in a workless household for a cohort of children born between 1986 and 1991, when they are young (0–5 years old) at the beginning of the 10 year window in 1991 and when they are older (10–15) at the other end of the observation window in 2001. It explores their chances of remaining in, moving into or moving out of a workless household between these two time points, and how those chances vary by ethnic group. It cannot be assumed that the households will have been workless throughout the whole period demarcated by the two measurement points. Indeed, we can expect substantial fluctuation in family and household circumstances. However, those who are continuously workless will be overrepresented at the second time point compared to those moving in and out. Moreover, it is relevant to observe that there is an association between worklessness at a ten year interval, even if there have been shorter moves out of worklessness within the period. The article estimates these chances, controlling for both household and family characteristics associated with the chances of living in a workless household, such as family composition, parental qualifications, access to a car and housing tenure. It also examines the contribution of changes in circumstances during the observation window, such as parental separation, change in family composition, or geographical mobility. By estimating models, both with and without these additional explanatory and control variables, it is possible to measure the extent to which family and household characteristics account for observed Office for National Statistics

76

Population Trends 139

Spring 2010

differences in patterns of worklessness between ethnic groups, or conversely, the extent to which residual ‘ethnic penalties’ remain.

Data and study design Data and sample This article makes use of an extract based on a cohort of LS members who were children aged 0–5 in 1991 and who were linked to their records in 2001 when aged 10–15. Information on the households and those enumerated in the households (the non-members in the data) at which the study members were living at either point in time also formed part of both extracts. The children had to be observed at two time points in order to be included in the sample. This means they will not precisely reflect the overall populations of children aged 0–5 in 1991 or aged 10–15 in 2001. However, those children who join the LS during the decade (via immigration or return) are not a concern of this analysis of transitions, and any potential bias stemming from systematic 28 differences in those observed at 1991 but not responding in 2001 is anticipated to be marginal.

Box one ONS Longitudinal Study data The ONS Longitudinal Study (LS) contains linked census and vital event data for one per cent of the population of England and Wales. Information from the 1971, 1981, 1991 and 2001 censuses has been linked across censuses as well as information on events such as births, deaths and cancer registrations. The original LS sample included 1971 Census information for people born on one of four selected dates in a calendar year. These four dates were used to update the sample at the 1981, 1991 and 2001 censuses and to add new members between censuses. New LS members enter the study through birth and immigration. Data are not usually linked to a member after their death or after de-registration from the NHS Central Register but these members’ records remain available for analysis. Census information is also included for all people enumerated in the same household as an LS member, but only information on LS members is linked over time.

Ethnic group Children’s ethnic group was allocated on the basis of their (non-imputed) ethnic group in 2001. Where ethnic group information was missing for 2001, the 1991 response and parental ethnic group were used to allocate ethnic group as far as possible. The approach for adding information from parents’ ethnic group was carried out on the basis of the observed patterns of parents’ and children’s ethnic group in the non-missing data. Therefore, where couple parents had the same ethnic group as each other, the child was given the ethnic group of the parents. Among the remainder, where two parents were from different white ethnic groups, the child was attributed White British ethnicity. Where the two parents were from different ethnic groups, these were mapped onto the appropriate mixed categories. This left some missing cases where only one parent was present. It is not possible to assume that lone parent and child share the same ethnic group, and so these few cases were excluded from the analysis. Table 3 shows the number of children included in the analysis by ethnic group. There were rather small numbers of children from some ethnic groups, rendering them unsuitable for detailed Office for National Statistics

77

Population Trends 139

Spring 2010

Table 3 Children aged 0–5 years in 1991 and observed aged 10–15 years in 2001 by ethnic group, England and Wales Ethnic group

Total in group

Per cent of sample

33,166

90.2

White Irish

119

0.3

White Other

304

0.8

White and Black Caribbean

394

1.1

75

0.2

White and Asian

228

0.6

Other Mixed Group

161

0.4

Indian

757

2.1

Pakistani

564

1.5

Bangladeshi

212

0.6

Other Asian

123

0.3

Black Caribbean

262

0.7

Black African

135

0.4

Other Black

97

0.3

113

0.3

63

0.2

36,773

100

394

1.1

37,167

100

White British

White and Black African

Chinese Other ethnic group Total Missing ethnic group Total including missing Source: ONS Longitudinal Study, author’s analysis

consideration, though they were included in the estimations for completeness. Small sample sizes were particularly an issue for White Irish, White and Black African, and Chinese children, as they were for the heterogeneous ‘other’ groups: Other Mixed, Other Asian, Black Other and Other. The illustration of results and the discussion therefore focus on the larger groups: White British, White Other, White and Black Caribbean, White and Asian, Indian, Pakistani, Bangladeshi, Black Caribbean and Black African.

Workless household For the purposes of this article, the definition of a workless household is that no member of the household was in paid work, either full-time or part-time. To construct the workless household variable, the non-members file was used, providing information on those co-resident with the LS member at each measurement point.

Additional explanatory and control variables The non-members file was also used alongside the members file to enable the construction of variables to indicate whether: • the sample member was living with both parents or just one at either time point • whether the co-resident parent(s) were UK born, and their educational level

Office for National Statistics

78

Population Trends 139

Spring 2010

• the age of the mother • whether the parents had experienced separation, widowhood or divorce within the decade • how many siblings were co-resident, whether this changed between the two time points, and whether there was a child aged under five in the household at the later time point Research suggests that all these are likely to influence the chances of adult household members being in work, and therefore the chances of a household being or becoming jobless. Household level variables on car ownership and housing tenure were also included, as was a measure of change in housing tenure. These variables are indicative of financial resources which may assist maintenance of family work and protect against adverse circumstances. They have been shown to be more directly related to employment outcomes. Access to a car, or at least possession of a driving licence, has been shown to be important in facilitating labour market 29 (re)-entry, including among lone parents. Housing tenure is known to be strongly associated 30 with employment status, as well as a range of other unfavourable outcomes. While the causal relationship and direction between housing tenure and other outcomes is hard to determine precisely, it does appear that living in social housing is not solely a consequence of disadvantage 23 in other domains, but may also shape outcomes. The analysis also included a measure of whether the family had experienced a geographical move between the two time points, and the distance moved. The economic variables are likely to be protective against joblessness, and geographical relocation may also imply a change in socioeconomic circumstances (including being associated with a move into work). The child’s own age and sex were also included. The variables included focused on the family (or parental) circumstances of the child and the more general household context. However, they were not exhaustive. This was partly for reasons of parsimony and the risks of overparameterising the model, given the small sample sizes of some ethnic groups, and partly to aid more direct interpretation. Other analysis of workless households (not focusing on ethnicity, and with a richer set of variables to choose from) has included a more 31 complex range of variables, but that can come at the risk of rendering individual variables hard to interpret. Key variables that may be relevant to consider in future analysis are regional effects and household size/number of adults.

Analytical approach Following inspection of the simple distributions of worklessness by ethnic group across the two time points and transitions between workless and non-workless states, binary logistic regression models were estimated for the probability of being in a workless household in 2001 conditioning on workless household status in 1991, and both with and without controlling for the household and family characteristics. This enabled entry, exit and persistence effectively to be summarised in a single model. By constructing a set of dummies that combined ethnic group and workless household status in 1991, the estimation allowed the association between workless household status in 1991 and 2002, that is, patterns of entry, exit and persistence, to vary by ethnic group, while still using the full estimation sample. Creating individual dummies for the combinations of ethnic group and household workless status avoids the problems of interpreting interaction effects 32 in a logit model, while not forcing the impact of prior worklessness to be constant across groups. Not allowing for interactions would mean, given the numerical dominance of the White majority group, that the effect of worklessness in 1991 on workless household status in 2001, would be

Office for National Statistics

79

Population Trends 139

Spring 2010

driven by the association for the white majority. In the results section, odds ratios are provided for combined ethnic group and 1991 workless households status effects. These capture entry rates relative to the reference category of White British children not living in a workless household in 1991, and persistence rates relative to the same reference group. Given that persistence rates for those workless in 1991 are likely to be higher for all ethnic groups, including the White majority, compared to White majority children not workless in 1991, evaluation of whether there are differences in persistence between minorities and the majority was attempted, by testing the equality of the coefficients for each minority dummy for those in workless households in 1991 with the coefficient for White majority children workless in 1991.

Results Patterns of children’s experience of workless households by ethnic group Figure 1 shows the simple proportions of the sample of children who experience worklessness at either time point. Overall, a substantial 21 per cent were living in a workless household in 1991 and 33 this declined to 17 per cent by 2001, a statistically significant change. Given that these are the same children who aged over the decade, this could be partly an age effect (that as the children become older, other members of their household become workers). For example, lone parents 34 become freed for work or older siblings still living in the household begin work. It could also be a structural effect related to the improvement in the economy and reduction of unemployment over time. The role of family and household characteristics in contributing to household worklessness and workless transitions is explored below. It is worth noting the very different rates of worklessness experienced across the groups. For most groups, except White Other and Pakistani children, there is a decline in household worklessness risks over time, though it is not statistically significant in all cases.

Figure 1 Proportions of children in workless households at 1991 (aged 0–5 years) and 2001 (aged 10–15 years) by ethnic group, England and Wales Percentages 60 51

Workless in 1991 Workless in 2001

50

49 45

40 34

34

26

24 24 20

21

19 17

36

36

32 27

30

44

18 15

13

12

10

0

All

White British

White Other

White and White and Black Asian Caribbean

Indian

Pakistani Bangladeshi

Black Caribbean

Black African

Source: ONS Longitudinal Study, author’s analysis

Office for National Statistics

80

Population Trends 139

Spring 2010

Figure 2 Movers and stayers, children in workless households 1991–2001, by ethnic group Percentages 9

8

8

8

12

11

10

13 25

8

11 11

4 7

26

27

9

11 17

17

9

15

18

18

10

21

14

19

27

72

73

22

80 64

65

40

53

51 In a workless household at both time points Moves into a workless household Moves out of a workless household Not in workless household at either time point All groups

White British

White Other

White and Black Caribbean

White and Asian

45

34

Indian

Pakistani

Bangladeshi

Black Caribbean

Black African

Source: ONS Longitudinal Study, author’s analysis

Figure 2 divides these overall risks of living in a workless household at the two time points by looking at the actual patterns of movement in and out of a workless household for the children at either end of the decade. It shows those who were in a workless household at neither time point, those who moved out of one over the decade (exits), those who moved into one over the decade (entries) and those who were living in a workless household at both time points (persistence). Is it that the majority of those who are living in workless households in 1991 were also living in workless households in 2001, or is there a significant movement between one time point and the next? It is clear from Figure 2 that experience of living in a workless household across the decade is a minority experience since 72 per cent of children lived in a working household at younger and older ages. Only nine per cent were living in a workless household at both ends of the decade. Of course, for the former group it cannot be assumed that they never experienced worklessness, nor that the latter group was continuously living in a workless household. For example, the extensive 35,36,37 dynamics in poverty was documented by Jenkins and others. 38

However, those with continuous experience will be over represented in either group. Aside from these two groups of ‘stayers’, however, a fifth of young children (20 per cent) have either moved out of or into a workless household by the time they are aged 10–15. There is additionally, substantial ethnic group variation in these patterns. With the exception of children in Indian families who are less likely to have experienced worklessness at either time point, the minority groups are all more likely than the white majority to have been in a workless household at one or other time, as well as to have experienced persistence, that is being in a Office for National Statistics

81

Population Trends 139

Spring 2010

workless household at both time points. The small sample sizes mean that not all the differences in persistence are statistically significant, but they differ significantly for Pakistani, Bangladeshi and Black African children compared to White British children. The proportions of all children who are persistently in workless households are particular high for White and Black Caribbean, Bangladeshi and Black African children: a quarter or more were living in a workless household at both time points. However, if we take the proportion in workless households as a proportion of all those workless at the first time point (that is the top green section over the top grey plus the first green sections combined) to be the persistence rate, that is the proportion of those children in workless households who are also in workless households at the second time point, we find a slightly different pattern. Indian children have the lowest persistence rates at 31 per cent, followed by 37 per cent for White and Asian children, 42 per cent for White British and Black Caribbean children, 48 per cent for White and Black Caribbean children, around 55 per cent for White Other, Pakistani and Bangladeshi children and to 58 per cent for Black African children. The patterns of transitions are also varied. Pakistani and Bangladeshi children have high rates of entry into worklessness: 17 and 18 per cent respectively of children from these groups moved into a workless household. As a proportion of those not workless at the first time point, this amounts to entry rates of 25 per cent and 35 per cent respectively, compared to only ten per cent for White British children. While in general there is a slight tendency of workless household rates to reduce with time (and age) – and this is particularly true for White and Black Caribbean children – Pakistani children (where moves into worklessness outweigh moves out of it) have in fact higher risks of living in a workless household at the later time point.

Estimating ‘ethnic penalties’ in children’s workless household persistence Table 4 shows the results from models estimating the impact of ethnic group and workless household status on chances of being in a workless household in 2001. In model 1 only the dummies created by interacting all ‘ethnic group’ categories with ‘1991 workless household status’ (in working household 1991 / in workless household 1991) were included, whereas model 2 also included the full set of household and family characteristics. Variables which were particularly strongly associated with persistence in or entry to workless households status in 2001 included family structure. ‘Presence of a father in 2001’, was found to be negatively and significantly associated with workless household status (odds ratio = 0.34). The variable ‘number and increase in siblings’ was also positively associated with living in a workless household in 2001. Parental qualifications at every level decreased chances of living in a workless household in 2001 relative to having no qualifications. Housing tenure was strongly associated with worklessness. Both private tenancy and social housing had odds ratios of over 5 relative to living in owner occupation. Change in housing tenure (that is into owner occupation) moderated this effect slightly as it was negatively associated with remaining or becoming workless by 2001 (odds ratio = 0.81). Car ownership in 1991 was also negatively associated with workless household status ten years later, consistent with expectations. While it is not possible to disentangle the causal relationships in every case, the indication is that prior household resources as well as parental qualifications and family structure are all implicated in children’s vulnerability to worklessness over time. These are areas that are already recognised as affecting children’s opportunities.

Office for National Statistics

82

Population Trends 139

Spring 2010

Table 4 Relative chances of being in a workless households in 2001 conditional on 1991 workless household status, by ethnic group Ethnic group

Household work status in 1991

Model 1 Simple model

Model 2 with household and family characteristics

In workless household 1991 (odds ratio) in working hh 1991 [relative entry compared to White British] (odds ratio)

7.9*** 1.4

2.14*** 1.70*

In workless hh 1991 [compared to White British not workless]

15.3***

4.43***

6.27(1)*

5.47(1)*

2.56*** 10.07*** 2.38(1)

1.42 1.89*** 0.53(1)

Reference: white not workless White British White other

Wald test of difference from White British workless [relative persistence compared to White British] (Chi2 (df)) White and Black Caribbean in working hh 1991 [relative entry] (odds ratio) In workless hh 1991 (odds ratio) Difference from White British workless [relative persistence] Chi2 (df) White and Asian

in working hh 1991 [relative entry] (odds ratio) In workless hh 1991 (odds ratio) Difference from White British workless [relative persistence] Chi2 (df)

1.26 5.98*** 0.99(1)

1.08 2.25* 0.02(1)

Indian

in working hh 1991 [relative entry] (odds ratio) Workless (odds ratio) Difference from White British workless [relative persistence] Chi2 (df)

0.79 4.95*** 3.69+

0.99 3.01*** 1.32(1)

Pakistani

in working hh 1991 [relative entry] (odds ratio)

3.25***

2.56***

14.15***

8.92***

10.98(1)***

38.96(1)***

In workless hh 1991 (odds ratio) Difference from White British workless [relative persistence] Chi2 (df) Bangladeshi

Black Caribbean

Black African

Number in analysis

in working hh 1991 [relative entry] (odds ratio)

5.05***

2.11**

In workless hh 1991 (odds ratio)

12.43***

2.41*

Difference from White British workless [relative persistence] Chi2 (df)

4.32(1)*

0.18(1)

in working hh 1991 [relative entry] (odds ratio)

1.95***

0.9

In workless hh 1991 (odds ratio)

7.92***

1.83*

Difference from White British workless [relative persistence] Chi2 (df)

0.00(1)

0.38(1)

in working hh 1991 [relative entry] (odds ratio)

2.83***

1.43

In workless hh 1991 (odds ratio)

9.30***

2.72*

Difference from White British workless [relative persistence] Chi2 (df)

0.24(1)

0.36(1)

33,051

P values: +