Estimating the Probability of Youth Unemployment - CiteSeerX

9 downloads 104 Views 58KB Size Report
Bradley (1990) stresses the importance of identifying the school-leavers at greatest risk of ...... Bradbury, Bruce, Pauline Garde and Joan Vipond (1986).
Estimating the Probability of Youth Unemployment

An Extended Essay for a BSocSc in Economics and Statistics1

Tarja K. Viitanen March, 1999

Abstract: This paper investigates the personal, regional, and family background characteristics of the unemployed youth using data from the General Household Survey. The main result noted is the different labour market experiences of males and females which may reflect the traditional genderspecific division of entry-level jobs. The variables which are found to be significant for both sexes include no qualifications, parental unemployment, and rented accommodation. Factor analysis and correlation analysis further highlight the importance of ‘good’ family background for a successful labour market entrance

I. INTRODUCTION

Knowledge of the relative importance of personal, regional and family background characteristics of the unemployed youth is crucial in designing effective policy. Current policy mainly concentrates on improving the employability of an individual through e.g. the Youth Training Scheme. The purpose of this paper is to investigate whether current policy emphasises the right issues by concentrating on the personal characteristics of the unemployed youth and especially whether parental background is more important in determining labour market outcomes. Many studies have found that spells of unemployment experienced during one’s early career have profound effects on future labour market outcomes and earnings. It has been shown that youth unemployment both reduces the probability of future employment (Ellwood 1982; Lynch 1989) and lowers wages throughout later work careers (Corcoran 1982; Ellwood 1982; Baker and Elias 1991; Gregory and Jukes 1997; Caspi et al. 1998). The human capital model is a labour market theory which assumes a unique set of abilities and acquired skills, or human capital. The model suggests that the age-earnings profile is upward-sloping and concave. This means that the optimal timing of human capital investment over the life-cycle is in one’s youth and that earnings increase with age but at a decreasing rate. Furthermore the model suggests that substantial investment in human capital should occur in the early years of labour market experience in the form of on-the-job training.

Therefore, spells of unemployment during one’s youth are

particularly costly as, first, the entire age-earnings profile of the worker will be depressed, and second, the human capital acquired through schooling is likely to depreciate. The dual labour market theory provides another reason why periods of unemployment may have longterm consequences. The theory suggests that early experiences of unemployment may lead to poor work habits and a weaker labour market attachment. The experience of unemployment may alter the attitudes of the young people as if they become discouraged about the chances of finding a job, the job

1

Word count excluding appendices: 4381

2

search behaviour is affected negatively. This can be interpreted as a discouraged worker effect which may eventually increase the adult unemployment rate. Therefore, knowledge of the determinants of unemployment is relevant to policy considerations as if personal characteristics antedate unemployment providing jobs may be insufficient. In this case the jobs would go to the most employable individuals, leaving those with adverse personal characteristics unemployed. Hence the argument is that personal and family background characteristics are important in determining labour market outcomes and that these factors should be taken into account when designing policy. In this paper the links between youth unemployment2 and personal, regional and family characteristics are studied using the General Household Survey 1993-6. The structure of the paper is as follows. Section II contains a literature review. Section III describes the data and the econometric methodology. Section IV present the empirical findings and Section V concludes.

II LITERATURE REVIEW

The early analyses of the characteristics of the young unemployed have primarily concentrated on examining the probability of the incidence of juvenile unemployment accounting for personal and regional characteristics. One of the first analyses of the characteristics of the unemployed youth is provided by Feldstein and Ellwood (1982). They find that for American males teenage unemployment is concentrated in a group that experiences long periods of unemployment. This group is characterised by low level of education and low family income. Similarly, Lynch (1987), using a longitudinal survey of London youths, finds individual characteristics such as ethnicity, parental unemployment, and educational qualifications significant in determining the likelihood of unemployment. These findings support the results of Main and Raffe (1983) for Scottish school leavers. They find that the young people who have sat “O” grades although failing (signal), possessed a part -time job, or did not have a truancy problem (proxy for attitudinal factors) are at a lower risk of unemployment at leaving school. Contrary to a priori expectations, a high level of parental education exerted a negative influence on the employment probability of the individual. This is

3

assumed to indicate interaction between parental expectations and child’s failure academically i.e. “while the children of educated parents may be more likely to succeed at school ceteris paribus, once they have ‘failed’ in a narrow academic sense this academic background may actually prove a disadvantage” (Main and Raffe, p.12). Extending on the previous analysis, Garner et al. (1988) found no effect of the overall unemployment rate on youth unemployment. Instead they find for Scottish cities that although unemployment varies across the areas of the cities, the variation is attributable to personal characteristics rather than the nature of the areas. They found father’s employment status and educational qualifications to matter the most. This implies that policy should be targeted towards the high-risk individuals with certain labour market characteristics e.g. those with no school leaving credentials. The importance of personal characteristics is further highlighted in a study by Andrews and Bradley (1997). In their study of transition from school, evidence of a segmented youth labour market appears. They use a multinomial logit analysis on cross-section data of all school leavers in Lancashire in 1991 and find that the chance of becoming unemployed is the greatest for the least able with no formal qualifications and for those who have been to large schools with lower academic achievements. They conclude that the unemployed are the “most marginal in the youth labour market, as typically they have lost out in the race for jobs--- [and] ---are reluctant to enter youth training” (Andrews and Bradley, p. 407). As a policy initiative they suggest subsidies to encourage young people to obtain intermediate skills and employers to value this kind of labour more highly. As low education level is a significant determinant of incidence of unemployment, it is important to understand what affects the school leaving decision.

Micklewright (1989) found that family

background, measured by parental education, socioeconomic class, and the number of siblings, all have a significant impact on the probability of leaving school at the minimum school leaving age while holding ability and type of school attended constant. Furthermore, Lindley (1996) investigates the school to work transition. His research suggests that the government schemes do not help the least able, especially those who are socially disadvantaged and he concludes that there is no evidence that post-compulsory vocational education and youth training programmes raise their employability, the probability of finding a job, or the pay once a job is found.

2

Unemployment is defined as the people ‘capable of and available for work’ who registered unemployed

4

Bradley (1990) stresses the importance of identifying the school-leavers at greatest risk of becoming unemployed before leaving school such that a more proactive approach can be taken on behalf of the teachers and careers officers. He finds expected qualifications, poor attendance at school and attendance at inner schools significant and states that these characteristics may be regarded as indicative of transitional difficulties. The inclusion of parental background variables in the estimated equations have given significant insights into the possibility of intergenerational transmission of inequality. It is suggested that the presence of household unemployment may be due to either a weaker work ethic or fewer job contacts. O’Neill and Sweetman (1998) examined the extent of intergenerational transmission of inequality in the form of unemployment. Using the National Child Development Study, they found that sons with unemployed fathers are almost twice as likely to experience unemployment compared to sons of fathers who have not been unemployed. They conclude that these results “highlight the importance of family background in explaining a child’s future labour market prospects” (O’Neill and Sweetman p.444). Similarly, O’Higgins (1994) examined the probabilities of several variables on the labour market outcome and found that especially paternal unemployment is correlated to individuals’ likelihood of finding work. This is believed to reflect constraints on fathers’ access to the labour market which may be transmitted to the offspring. A similar argument in favour of the intergenerational transfer of preferences is raised in Hart (1988). He applies the income-leisure trade-off theory to the behaviour of an atypical young person who is most likely to be unemployed. Supposing that this young person’s family does not accept the conventional work ethic i.e. supposing unemployed parents with a detailed knowledge of the rules of obtaining state benefits, the young person might want to continue “the parental tradition and decide not to participate unless there is a large excess of net wages over benefits…” (Hart, p.43). The argument is that the expectations of these young people are too ambitious. Extending the analysis from parental unemployment, Bradbury, Garde and Vipond (1986) found that youth unemployment is affected by family background and location as well as personal characteristics. They used a logistic regression analysis on data from the 1981 Australian Census of Population and found that youth in single-parent families with low level of education and low language ability are at

or claim to be looking for work. The definition is used by the General Household Survey.

5

highest risk. Additionally, youth coming from families that move frequently (as informal job seeking networks are disrupted) or live in rural areas were more likely to be unemployed. Also for Australia, Harris (1996), modelled youth unemployment to quantify the effects of personal characteristics on the probability of unemployment. He used a probit analysis on panel data and the results suggest that place of residence, marital status and accommodation type are important determinants of duration and/or incidence of unemployment.

By modelling the incidence of

unemployment, he attempts to ascertain whether by possessing certain characteristics, these youth are disadvantaged in the labour market. The identified explanatory variables can then be incorporated into duration models such that policy measures can be taken to ascertain that disadvantaged individuals will be less likely to be unemployed. A similar study by Miller (1998) distinguishes between three sets of factors influencing unemployment outcomes: personal characteristics such as age and educational attainment; family characteristics such as employment status of parents; and regional characteristics such as area of residence. Using the 1991 Australian Census of Population and Housing, he found that foreign born who have only recently arrived in the country, young people with unemployed parents or unemployed siblings and young people in low family income households are among those who experience exceptionally high unemployment rates. He concludes that family and regional circumstances are more important in determining labour market outcomes than personal characteristics. Hence policy should focus on regional employment initiatives rather than just job training aimed to improve the employability of an individual. Caspi et al. (1998) investigate the childhood and adolescent predictors of youth unemployment for the United States. They found that poor education and limited parental resources, growing up in a singleparent family, family conflicts, lack of attachment to school and antisocial behaviour significantly increase the risk of future unemployment.

They conclude these various personal and family

characteristics affect labour market outcomes years before youths enter the labour force, first by restricting the later accumulation of human capital e.g. education, and second by having direct effects on labour market behaviour e.g. job-search behaviour. Hence it is not simply the lack of skills that makes youth vulnerable to unemployment but rather “a constellation of psychosocial and family characteristics emerging early in the life course is implicated in a turbulent school-to-work transition”

6

(Caspi et al. p.445). This implies that policy should be directed towards family and educational support early in the life. This argument is supported by O’Higgins (1994) who found that in general YTS did not eliminate the influence of labour market disadvantages. Hence we can infer from this result that the preventative policy measures should be taken at an earlier stage and possibly even at the parental generation. A similar argument is presented by Armstrong (1997); his study shows that careers guidance does not have a significant effect on the probabilities of 16-18 year old youths in Northern Ireland experiencing unemployment. Instead, he finds that the number of GCSE passes, type of family (both parents present/single parent family) and local unemployment rates for males significantly influence the incidence of unemployment. Gardecki and Neumark (1998) find that for American women early labour market stability has positive effects on the adult labour market outcomes in the form of increased wages and more generous benefits. They also found that training in the early years has lasting benefits in the form of higher wages (approximately 7-10% increase in wages per one year of training) for both sexes thus supporting the human capital model. Similar findings are presented by Hashimoto and Miller (1999) who find that training and early labour market experiences affect wages positively for both sexes for the American youth.

On the other hand, both studies found that training did not reduce the likelihood of

unemployment thus indicating the importance of personal characteristics in the determination of labour market outcomes. The General Household Survey is a household-based file thus the records for members of a family living together can be linked together. This enables us to determine whether there is inter-generational transmission of inequality in the UK in the form of unemployment patterns in addition to using the more conventional explanatory variables.

III DATA AND ECONOMETRIC METHODOLOGY

The General Household Survey (GHS) provides an ideal data set for the analysis as it contains vast information on the respondents economic activity, family composition, and education. As the GHS is a household based survey, it is possible to link together the records for members of a family living

7

together. In the analysis, I have included a pooled sample of all the 16 -24 year olds over a three-year period 1993-6. The total sample size is 2910 observations after removing married youths and those not living at home from the dataset. Description of the variables of interest is included in the Appendix I. A further division is made between males and females as the labour market experience is gender biased. I will use models similar to Miller (1998), Andrews and Bradley (1997), and Bradbury, Garde and Vipond (1986).

These models distinguish three sets of factors that influence unemployment outcomes:

personal characteristics (such as low level of education or no formal qualifications, low language ability, and attendance to large schools with lower academic achievements), family characteristics (such as presence of unemployed parents or unemployed siblings, low family income and single-parent families), and regional characteristics (such as level of unemployment). The multinomial logit model allows us to consider the four possible unordered outcomes together. The unordered choice model is motivated by a random utility model. Supposing that the utility of choice j, with j = 0,…,J is Uij = βTxij + εij

(1)

, where i = 1,…,N denotes the individual and j = 0,…,J labour market outcomes/choices. So if the youth makes a particular choice j, we assume that Uij is the maximum among the J utilities i.e. ,∀ k ≠ j

Prob[Uij > Uik]

(2)

The J + 1 = 4 labour market choices/outcomes are defined as: - full-time education (y = 0) - unemployment (y = 1) - employment (y = 2) - other (y = 3) These four categories make up the dependent variables i.e. y = j, j = 0,…, J. The multinomial logit model for labour market outcome is:

Prob(Y = j)

=

e

β

T j

x

i

∑k = 0 e β k xi 3

8

T

(3)

The estimated equations provide a set of probabilities for the J + 1 choices with characteristics xi. Denoting Pij as Prob(Yi = j), the probability that the jth outcome is made for the ith young person with characteristics xi, then the multinomial logit regression model can be written

log

Pij Pi0

=

(

x Ti β j − β0

)

(4)

, where i indexes N individuals and j indexes 0,…,J choices/outcomes. Rather than reporting the maximum likelihood estimates of βj which are difficult to interpret, we follow Greene (1991, p.697) and report the ratio of relative risk for a one-unit change in xi i.e.

Pr ( y = 1 ) =e Pr ( y = 2 )



( 1) i

(5)

, where y = 2 (employment) is the base category. Thus, the exponentiated value of the coefficient is the relative risk ratio for a one unit change in the corresponding variable, it being understood that risk is being measured as the risk of the category relative to the base category.

IV EMPIRICAL FINDINGS

There are 1224 males and 990 females in the sample. Only individuals who live with their parents at the time of the survey are included in the sample. The omitted categories define the benchmark group as a white 16 year-old employee who has taken GCSE’s and lives with both parents neither one of which is unemployed. Table 1 reports the results from the multinomial logit regression in relative risk ratios (RRR)3. All of the variables included in the regression are kept constant i.e. they are controlled for against the benchmark group in the analysis. As an example of how to interpret these, males with no qualifications are 3.7 times as likely to be unemployed than their counterparts in the base category with O-levels; furthermore, this result is statistically significant at 1% level4.

3

All econometric analysis is conducted using Stata 5.0. Statistical significance is based on the ratio of the coefficient β to its standard error; in this exercise, the coefficient β is transformed into relative risk ratio as shown in equation (5) to ensure clearer interpretation of results, thus the t-statistic cannot be calculated by the above method. Appendices I and II include all of the results. 4

9

The differences between gender are remarkable which might reflect the fact that males form 2/3 of the unemployed group. One of the variables that is significant both for females and males, however, is the lack of qualifications. This has the effect of increasing the probability of unemployment, with the effect being almost twice as large for males than for females. This may be interpreted in such manner that there are differences in the types of entry-level jobs males and females acquire, with males being expected more e.g. technical or mathematical skills. This finding that the least qualified face highest risk of experiencing unemployment at leaving school is in accordance with Main and Raffe (1983), Lynch (1987), and Bradley (1990). The variable on whether one has taken A-levels is not significant for either sex. However, the sign is in accordance with both the human capital theory and the screening hypothesis meaning that acquiring education decreases the probability of unemployment. The differences in gender are further illustrated within the age variables as none are significant for females. For males, the variable increases both in significance and relative risk as age increases. This is likely to be the consequence of a ‘disillusioned worker’ effect as it can be argued that the longer an

Table 1 Multinomial logit estimates of unemployment probability, 16 - 21 year olds, GHS5 Variable Education level No qualifications A-levels Age 17 years 18 years 19 years 20 years 21 years Deprived area Married Non-white Mobile Renting Sole parent Big family Family income Father unemployed no qualifications manual worker professional/manager Mother 5

Males, n = 1224

Females, n = 990

3.729 0.711

(1.067) [0.000] *** (0.213) [0.254]

1.918 0.808

(0.697) [0.073] * (0.301) [0567]

1.889 2.295 2.326 4.865 1.844 1.537 1.479 4.852 1.309 1.199 3.781 1.428 0.9996

(0.740) [0.105] (0.876) [0.029] ** (0.931) [0.035] ** (1.882) [0.000] *** (0.757) [0.136] (0.327) [0.043] ** (1.576) [0.713] (2.078) [0.000] *** (0.365) [0.334] (0.294) [0.459] (1.025) [0.000] *** (0.591) [0.389] (0.0006)[0.507]

1.447 1.245 1.399 1.903 2.134 1.327 4.214 1.706 1.076 3.394 1.181 1.554 1.0002

(0.644) [0.407] (0.586) [0.642] (0.702) [0.504] (1.003) [0.222] (1.057) [0.126] (0.387) [0.333] (4.363) [0.165] (0.750) [0.225] (0.380) [0.836] (1.125) [0.000] *** (0.423) [0.643] (0.843) [0.417] (0.0006)[0.709]

1.861 0.919 1.147 0.891

(0.543) [0.033] ** (0.218) [0.721] (0.275) [0.566] (0.289) [0.722]

1.961 0.573 1.432 2.025

(0.723) [0.068] * (0.201) [0.113] (0.491) [0.296] (0.795) [0.072] *

Tables of results with all the outcome groups are presented in the Appendices II and III.

10

unemployed no qualifications Note:

1.249 (0.317) [0.381] 1.874 (0.421) [0.005] *** Pseudo R2 = 0.19

1.946 (0.614) [0.035] ** 1.027 (0.303) [0.928] Pseudo R2 = 0.16

Asymptotic t -statistics in ( ). Probability values in [ ]. Statistical significance at 10%, 5%, 1% levels denoted by *, **, ***, respectively.

individual is economically inactive, the harder it is to re-enter employment. Additionally, the human capital deteriorates with duration and number of spells of unemployment. The dummy variable ‘deprived area’ takes value 1 if the regional unemployment rate is above the national average unemployment rate. The coefficient for females not being significant could reflect the fact that the female unemployment rates are lower in general as females may be more likely to drop from the labour force thus not registering as unemployed. The sign of the coefficients is economically consistent as one is more likely to experience unemployment in a state with higher overall unemployment. According to this analysis, race is a significant variable in determining unemployment outcomes only for males. This highly significant variable suggests that being black or Asian increases the likelihood of becoming unemployed almost five-fold compared to our base model6. This may reflect cultural differences or it may be an indication of pure discrimination on the part of the employer. The variable ‘mobile’ takes the value 1 if the family has moved more than 3 times within the past five years. Miller (1998) suggests that mobility is an important determinant of labour market outcomes as moving can be disruptive to the job search process. The sign of the coefficient is in accordance with Miller’s findings but the variable is not significant for either sex. On the other hand, the individuals whose parents are renting their dwelling as opposed to owning or buying it are more likely to become unemployed. This variable is highly significant for females with a relative risk ratio of 3.4 possibly reflecting a wealth or a neighbourhood effect. The effect of family income is inconclusive due to the large amount of missing variables thus it may be advisable to use the renting variable as a proxy for low family income. The family background variable indicating a single parent family is highly significant for males. This indicates that being brought up in the absence of the other parent, in the majority of cases the father,

11

increases the probability of becoming unemployed by almost four times. In our sample, 40% of the unemployed youths come from single-parent families. The high unemployment within this group may be an indication of the lower family income, the lack of a ‘father figure’, or more simply, the high number of single parents that are economically inactive; the latter two interpretations assume some kind of mechanism of intergenerational transfer of inequality. Father’s unemployment is significant and of the expected sign for both sexes but mother’s unemployment is significant only for the females. These results may indicate the above mentioned transfer of preferences e.g. weaker work ethic as parents can be considered the most influential role models for their offspring. Miller (1998) argues that unemployment in the family unit affects the job search process as informal contacts are crucial for the youth and also because parental unemployment may affect the behavioural patterns. Quite unexpectedly, mother’s lack of qualifications has a highly significant negative effect on the son’s labour market outcome. The correlation matrix in Appendix IV shows mother’s lack of qualifications being positively correlated with father being a manual worker and negatively correlated with father being manager/professional. Hence this result may reflect a socialclass effect assuming that lower social classes significantly differ from the higher ones in their experiences and preferences concerning unemployment. Overall, the analysis reveals that differences exist between males and females in their labour market experiences. Much more traditional variables seem to affect the unemployment probabilities of young males compared to those of females. This may reflect the weaker labour market attachment of the females or it could be an indication of inherent differences and preferences. Appendix VI reports the results of a factor analysis which indicate that the dimensions of the problem can ultimately be reduced to three dimensions out of the eight variables examined7. The results lay heavily on single-parenthood, disposable family income, and on neighbourhood effects and race as explanations of youth unemployment. Thus it can be concluded that family background and regional effects are important determinants of the incidence of youth unemployment.

6

This analysis has not differentiated between the minorities as the pooled surveys did not give a large enough sample size for statistical significance. 7 The variables include no qualifications, mother’s unemployment, father’s unemployment, single parenthood, large family size, mobility, renting accommodation, and living in a deprived area with high overall unemployment.

12

V CONCLUSION

Youth unemployment has increased at a higher rate than overall unemployment since the 1950’s and conventional economic theories have not been able to explain this trend. The importance of reducing the incidence of unemployment among the young labour market entrants is reflected in the human capital and the dual labour market theories. They assume that the experience of unemployment in one’s youth has irreversible consequences on later labour market outcomes (Corcoran 1982; Ellwood 1982; Baker and Elias 1991; Gregory and Jukes 1997; Caspi et al. 1998; Lynch 1989). A notable difference is detected between characteristics of the unemployed females and males. Gender unemployment rate differential can arise from difference in the marketable characteristics of the two groups, also labour market structures differ as many entry-level jobs may be female dominated e.g. shop assistant. Quality of education measured by subject choice may restrict females from entering the more technical occupations. It must be remembered though that the parameter estimates may be biased due to omission of unobserved variables on subjective data such as IQ, attitude, and motivation. The differences in gender may be explained using such idiosyncratic characteristics as, for example, females may prefer to follow their mother’s example and stay out of the labour force. The results support the human capital model as the groups with inferior levels of human capital skills i.e. no qualifications are more likely to experience unemployment.

In addition to these personal

characteristics, especially family background characteristics prove to be important in determining the likelihood of unemployment. The unemployment status of family members may capture a multitude of unobserved dimensions of family background such as preferences. The transmission of preferences has only recently arisen as a potential research topic. I would especially like to extend the analysis and investigate first, whether there is intergenerational transmission of inequality in the UK in unemployment patterns and second, whether this is due to transfer of preferences and/or information from the parental generation to the offspring. If this is found to be the case, youth unemployment may be more of a symptom of social problems and policy should be oriented towards the regional and family effects.

13

APPENDIX I: Data appendix

The variables constructed for use in the analysis are defined as follows. Big family: Variable takes the value one if there are three or more children in the household. Deprived area: A dummy variable taking value 1 if the regional unemployment rate exceeds the national unemployment rate. Educational attainment: This is divided into three variables: (i) no qualifications, (ii) O-levels, and (iii) A-levels. Family type: Distinction is made between dual-parent families and sole-parent families. Father’s/mother’s educational attainment: This is divided into three variables: (i) no qualifications, (ii) O-levels, and (iii) A-levels. Father’s/mother’s employment status: The variable takes three values: (i) employed, (ii) unemployed, and (iii) other. Father’s/mother’s occupational status: A distinction is made between (i) manual workers, (ii) nonmanual workers, and (iii) professional/managerial. Mobile: A variable expressing whether the family has moved more than three times within the past five years. Nature of occupancy: The survey data on the type of occupancy can be grouped as follows: ( i ) owned, ( ii ) rented. Parental income: A continuous variable expressing the gross weekly earnings of the household. Population: The population is restricted to 16-24 year olds. Race: The race variable distinguishes between the reported white category and others.

14

APPENDIX II: Multinomial logistic regression results for males using GHS No. of obs chi2(66)=451.87 Prob>chi2= 0.0000 Log Likelihood = -84.88866 Pseudo R2 = 0.1866 =1224 Economic status Variable RRR Std.Err. z P>|z| [95% Conf. Interval] Full-time education

Unemployed

age17

.5963337

age18

.2563529

age19

.1199963

age20

.0648083

age21

.0331749

noqual

1.428864

alevels

1.759445

faunemp

1.460654

mounemp

1.516713

mnoqual

1.051336

fnoqual

.802366

fmanual

.6901651

fprofes

1.154216

rent

.809998

income

1.000486

nonwhite

13.22487

lonepare

1.246697

bigfamil

.5422475

mobile

1.302419

foreign

.8461203

deprarea

1.545089

married

.2576045

age17

1.889091

age18

2.295377

age19

2.325819

age20

4.864639

age21

1.844397

noqual

3.729319

alevels

.7106704

faunemp

1.860876

mounemp

1.248824

.127905 9 .064046 5 .038102 7 .026354 8 .015683 3 .344213 4 .428826 4 .367783 4 .306130 8 .192329 1 .166141 9 .140823 8 .242938 4 .177618 4 .000314 9 4.67330 2 .284018 5 .203399 .290642 3 .373824 3 .277321 4 .362238 3 .740376 3 .876066 4 .931254 8 1.88185 6 .757440 2 1.06677 3 .212904 4 .542993 2 .316843

15

2.410 5.448 6.677 6.729 7.205 1.481

0.016

.3916676

.9079482

0.000

.1571

0.000

.0643997

.2235896

0.000

.0292067

.1438066

0.000

.0131343

.0837939

0.138

.8911206

2.291107

2.318

0.020

1.091224

2.836858

1.505

0.132

.8917029

2.392624

2.064

0.039

1.02117

0.274

0.784

.7345552

1.504729

1.063 1.817 0.681

0.288

.5347103

1.204

0.069

.4626693

1.029521

0.496

.7640591

1.743602

0.961 1.543

0.337

.527024

0.123

.9998687

1.001103

7.307

0.000

6.616072

26.43519

0.968

0.333

.797706

1.632 1.184

0.103

.2599614

1.131062

0.236

.8410078

2.016978

0.378 2.424

0.705

.3559286

2.011413

0.015

1.086864

2.196504

0.965 1.623

0.335

.0163686

4.054114

0.105

.8762912

4.072466

2.177

0.029

1.086375

4.84985

2.108

0.035

1.061099

5.097953

4.089

0.000

2.279126

10.38324

1.491

0.136

.8246902

4.124945

4.601

0.000

2.12884

1.140 2.128

0.254

.3950601

1.278419

0.033

1.050368

3.296805

0.876

0.381

.7595209

2.053347

.4183119

2.252726

1.244909

1.948404

6.533052

Other

mnoqual

1.874331

fnoqual

.9185265

fmanual

1.147313

fprofes

.8909156

rent

1.199037

income

.9995819

nonwhite

4.852423

lonepare bigfamil

3.780547 1.428466

mobile foreign

1.30886 .8001042

deprarea married

1.537389 1.479431

age17

1.369404

age18

1.391331

age19

1.751761

age20

.4245214

age21

.9092108

noqual

1.816922

alevels

.7983656

faunemp

.9922998

mounemp

2.052869

mnoqual

.8802873

fnoqual

2.151611

fmanual

.5763709

fprofes

.8872173

rent

1.689374

income nonwhite

1.000673 4.370927

lonepare

1.484298

bigfamil

.6137713

mobile

.5204857

foreign

2.648767

deprarea

2.275732

married

1.62e-13

4 .421416 5 .218292 7 .274780 9 .288978 6 .293938 6 .000629 4 2.07768 2 1.02505 .591006 4 .36478 .464908 6 .327428 1.57602 9 .816075 5 .817862 4 1.06319 6 .371525 8 .611632 2 .942997 5 .419285 .531011 5 .876543 3 .344593 9 .874703 9 .248186 7 .445117 7 .718621 3 .000744 2.66364 9 .722066 9 .447826 3 .206202 6 1.66672 1 .832610 1 5.07e-07

16

2.794

0.005

1.206329

2.912237

0.358 0.574

0.721

.5764998

1.463471

0.566

.7174956

1.834612

0.356 0.740

0.722

.4717801

1.682416

0.459

.741591

1.938657

0.664 3.689

0.507

.998349

1.000816

0.000

2.096507

4.905 0.862

0.000 0.389

2.22209 .634889

0.966 0.384 2.019 0.368

0.334 0.701

.7579887 .2561812

2.260078 2.498882

0.043 0.713

1.012739 .1833644

2.333833 11.93643

0.528

0.598

.4258643

4.403441

0.562

0.574

.4396133

4.403419

0.924

0.356

.5331572

5.755654

0.979 0.141 1.151

0.328

.0763763

2.359611

0.887

.2432516

3.398393

0.250

.6569855

5.024778

0.429 0.014 1.684

0.668

.2852124

2.234782

0.988

.3476463

2.832358

0.092

.88902

0.326 1.885

0.745

.408711

1.895975

0.059

.969887

4.773166

1.280 0.239 1.233

0.201

.2478407

1.340391

0.811

.3318803

2.371802

0.218

.7339122

3.888728

0.905 2.420

0.365 0.016

.9992159 1.323885

1.002132 14.43102

0.812

0.417

.572054

0.669 1.648 1.548

0.503

.1468736

2.564893

0.099

.2394337

1.131442

0.122

.7716675

9.091956

2.248

0.025

1.110955

4.661715

0.000

1.000

0

11.23107 6.432025 3.21397

4.740358

3.851282

.

APPENDIX III: Multinomial logistic regression results for females using GHS No. of obs =990 chi2(66)=314.05 Prob>chi2= 0.0000 Log Likelihood = -820.8192 Pseudo R2 = 0.1606 Economic status Variable RRR Std.Err. z P>|z| [95% Conf. Interval] Full-time education

Unemployed

age17

.5366429

age18

.2740435

age19

.0953196

age20

.0580592

age21

.046559

noqual

1.233545

alevels

1.739162

faunemp

1.402771

mounemp mnoqual

2.409366 1.072378

fnoqual

1.221512

fmanual

.7059372

fprofes

1.009859

rent

.6857113

income

1.00064

nonwhite

3.658831

lonepare

1.72326

bigfamil

.81776

mobile

1.113186

foreign

3.446874

deprarea

1.370042

married age17

2.01e-14 1.446601

age18

1.244727

age19

1.398719

age20

1.903282

age21

2.134431

noqual

1.918194

alevels

.8082644

faunemp

1.961173

mounemp

1.94581

.123278 9 .071369 4 .035480 8 .025973 4 .024160 7 .299514 8 .467505 4 .376948 2 .513536 .205104 6 .268812 6 .155522 .220660 6 .156829 5 .000296 6 1.00604 2 .391808 7 .346738 1 .254673 7 1.59372 9 .256431 9 1.31e-07 .644470 2 .585652 7 .702054 2 1.00312 3 1.05731 4 .696519 7 .300910 4 .722533 5 .613671

17

2.709 4.970 6.315 6.362 5.910 0.864

0.007

.3420939

.8418319

0.000

.1644902

.456561

0.000

.0459557

.1977083

0.000

.0241589

.1395293

0.000

.0168381

.1287405

0.387

.7664372

1.985335

2.059

0.040

1.0269

1.260

0.208

.8284272

4.126 0.365

0.000 0.715

1.586633 3.658721 .737133 1.560091

0.909

0.363

.793556

1.581 0.045

0.114

.4583967

1.087153

0.964

.6580633

1.549721

1.650 2.158

0.099

.4379881

1.073545

0.031

1.000059

1.001221

4.718

0.000

2.134488

6.271782

2.394

0.017

1.103615

2.690816

0.474 0.469

0.635

.3562118

1.877342

0.639

.710936

1.74303

2.676

0.007

1.392695

8.530903

1.682

0.093

.9493228

1.977216

0.000 0.829

1.000 0.407

0 . .6041334

3.463894

0.465

0.642

.4949692

3.130188

0.669

0.504

.5229859

3.740857

1.221

0.222

.6774519

5.347219

1.531

0.126

.8084065

5.635524

1.794

0.073

.9414795

3.908177

0.572 1.828

0.567

.3896349

1.676675

0.068

.9526142

4.037521

2.111

0.035

1.048692

3.61038

2.945453 2.375302

1.880259

Other

mnoqual

1.027084

fnoqual

.5734871

fmanual

1.43188

fprofes

2.025288

rent

3.394233

income

1.000213

nonwhite

1.705712

lonepare

1.180837

bigfamil mobile

1.553606 1.076052

foreign

1.979866

deprarea

1.326923

married age17

4.213841 1.557344

age18

.9842877

age19

1.203996

age20

.9848325

age21

1.589786

noqual

1.631185

alevels

.7817079

faunemp

1.974143

mounemp

1.497911

mnoqual fnoqual

1.612221 1.047777

fmanual

.9586364

fprofes rent income

1.454398 1.776953 .9996601

nonwhite

1.250821

lonepare

2.83767

bigfamil

3.141433

mobile

1.310775

foreign deprarea

5.169483 1.3843

married

17.08528

.303229 3 .201138 3 .491430 2 .794693 7 1.12533 6 .000569 7 .750465 1 .423461 4 .842612 .380218 2 1.25582 5 .387424 2 4.36311 .824688 9 .591099 7 .737245 8 .679673 1.00571 9 .716800 7 .393469 7 .933300 2 .598159 2 .589678 .447695 2 .387865 9 .730016 .708031 .000882 8 .680593 8 1.20645 3 1.73651 2 .596541 7 3.32435 .489879 3 15.2690 9

18

0.091

0.928

.5758417

1.83193

1.585 1.046

0.113

.2883919

1.140419

0.296

.7307503

2.80572

1.799

0.072

.938615

3.686

0.000

1.772275

6.50058

0.374

0.709

.9990968

1.00133

1.214

0.225

.7201136

4.040269

0.464

0.643

.5847087

2.384736

0.812 0.207

0.417 0.836

.5366416 4.497774 .53835 2.150807

1.077

0.282

.5711107

6.863588

0.969

0.333

.7487189

2.35165

1.389 0.837

0.165 0.403

.5537632 .5516098

32.06508 4.396802

0.026 0.303

0.979

.3033509

3.193734

0.762

.3625821

3.998005

0.022 0.733

0.982

.2546323

3.809002

0.464

.4601061

5.493125

1.113

0.265

.6893745

3.859681

0.489 1.439

0.625

.2914711

2.096494

0.150

.781561

1.012

0.312

.6848206

1.306 0.109

0.192 0.913

.787215 3.301839 .4534889 2.420868

0.104 0.746 1.443 0.385 0.411

0.917

.4337668

2.118612

0.455 0.149 0.700

.5437931 .8137905 .9979315

3.889849 3.880066 1.001392

0.681

.430568

2.453

0.014

1.233299

6.529132

2.071

0.038

1.063171

9.282231

0.595

0.552

.537206

3.19827

2.555 0.919

0.011 0.358

1.465761 .6918388

18.23186 2.769844

3.176

0.001

2.964181

98.47807

4.370045

4.986485 3.276389

3.633699

APPENDIX VI Factor analysis results for unemployed youths using GHS (maximum likelihood factors; 3 factors retained)

Factor analysis is a statistical technique for data reduction which allows you to reduce the number of variables in an analysis. Factor analysis is concerned with finding a small number of common factors (say q of them) that linearly reconstruct the p original variables: y ij = zi1b 1j + zi2b 2j + … +ziqb qj + eij where yij is the value of the ith observation on the jth variable, zik is the ith observation on the kth common factor, bkj is the set of linear coefficients called factor loadings, and eij is similar to a residual but known as the jth variable’s unique factor. Factor 1 2 3 Test: Test:

Variance 1.14997 1.22657 0.70589

Difference -0.07659 0.52068 .

3 vs. no factors. 3 vs. more factors.

Variable noqual mounemp faunemp lonepare nonwhite bigfamil mobile rent deprarea

Chi2( Chi2(

1 0.16897 0.03239 -0.23516 1.00000 0.00536 0.00367 -0.02628 0.22966 0.10770

27) = 12) =

Factor Loadings 2 0.28976 0.48481 0.29525 0.00000 0.07001 0.29715 -0.22522 0.81664 0.09781

Proportion 0.3731 0.3979 0.2290

Cumulative 0.3731 0.7710 1.0000

224.15, Prob > chi2 = 12.70, Prob > chi2 =

3 0.03193 0.35421 0.21946 0.00000 0.42737 -0.04350 0.15703 -0.13583 0.55101

0.0000 0.3914

Uniqueness 0.88647 0.63844 0.80936 0.00000 0.81242 0.90980 0.92393 0.26184 0.67524

The Chi2 test results show us that the data can be explained by three factors i.e. it tells us how well the correlation matrix is fit. The first factor loads heavily on single-parenthood as an explanation of youth unemployment. The second factor explains the phenomenon mainly by disposable family income, and the third factor on neighbourhood effects and race. Uniqueness is the percentage of variance for the variable that is not explained by the factors i.e. if the uniqueness is high then the variable is not well explained by the factors. The uniqueness is initially assumed to be zero. Rotated Variable noqual mounemp faunemp lonepare nonwhite bigfamil mobile rent deprarea

1 0.14128 -0.01229 -0.26126 0.99565 -0.00041 -0.02412 -0.00489 0.15230 0.09911

Factor Loadings 2 0.29262 0.40924 0.22511 0.09183 -0.01260 0.29905 -0.25248 0.84513 0.00039

(varimax 3 0.08912 0.44037 0.26778 0.01601 0.43291 0.01384 0.11092 0.02556 0.56120

rotation) Uniqueness 0.88647 0.63844 0.80936 0.00000 0.81242 0.90980 0.92393 0.26184 0.67524

The maximum likelihood analysis possesses the problem that there may be more than one local maximum.

19

REFERENCES Andrews, Martyn and Steve Bradley (1997). ‘Modelling the Transition from School and the Demand for Training in the United Kingdom’, Economica, August 1997, pp.387-413. Armstrong, David (1997). ‘Careers Guidance, Psychometric Testing and Unemployment Amongst Young People: an Empirical Analysis for Northern Ireland’ EALE Conference Paper. Baker, Meredith and Peter Elias (1991). ‘Youth Unemployment and Work Histories’ in Life and Work History Analyses: Qualitative and Quantitative Developments, ed. by S. Dex. London, England: Routledge. Bradbury, Bruce, Pauline Garde and Joan Vipond (1986). ‘Youth Unemployment and Intergenerational Immobility’, Journal of Industrial Relations, Vol. 28, No. 2, pp. 191-210. Bradley, Steve (1990). ‘School-leaver Unemployment: a Proactive Approach’, Applied Economics, Vol. 22, pp. 439-461. Caspi, Avshalom et al. (1998). ‘Early Failure in the Labor Market: Childhood and Adolescent Predictors of Unemployment in the Transition to Adulthood’, American Sociological Review, Vol. 63, pp. 424-451. Corcoran, Mary (1982). ‘The Employment and Wage Consequences of Teenage Women’s Nonemployment’ in The Youth Labour Market Problem: Its Nature, Causes, and Consequences’ ed. by Richard B. Freeman and David A. Wise. Chicago, IL: University of Chicago Press. Dolton, P.J. and G.H. Makepeace (1997). ‘The Earnings and Employment Effects of Training in Britain’, Mimeo, 1997. Dolton, P.J., G.H. Makepeace and J.G. Treble (1994). ‘The Youth Training Scheme and the School-toWork Transition’, Oxford Economic Papers, Vol. 46, pp.629-657. Ellwood, David T. (1982). ‘Teenage Unemployment: Permanent Scars or Temporary Blemishes’ in the Youth Labour Market Problem: Its Nature, Causes, and Consequences’ ed. by Richard B. Freeman and David A. Wise. Chicago, IL: University of Chicago Press. Feldstein, Martin and David T. Ellwood (1982). ‘Teenage Unemployment: What is the Problem?’ in the Youth Labour Market Problem: Its Nature, Causes, and Consequences’ ed. by Richard B. Freeman and David A. Wise. Chicago, IL: University of Chicago Press. Franz, W et al. (1997). ‘Young and Out in Germany: on the Youths’ Chances of Labor Market Entrance in Germany’, NBER Working Paper, No.6212. Gardecki, Rosella and David Neumark (1998). ‘Order from Chaos? The Effects of Early Labor Market Experiences on Adult Labor Market Outcomes’, Industrial and Labor Relations Review, Vol.51, No. 2, pp.299-322. Garner, Catherine, Brian G. M. Main and David Raffe (1988). ‘The Distribution of School-Leaver Unemployment Within Scottish Cities’, Urban Studies, Vol.25, pp. 133-144. Greene, William H. (1991). Econometric Analysis. New York: MacMillan. Gregory, Mary and Robert Jukes (1997). ‘The Effects of Unemployment on Subsequent Earnings: a Study of British Men 1984-94’, Mimeo, October 1997, St. Hilda’s College, Oxford University.

20

Harris, Mark N. (1996). ‘Modelling the Probability of Youth Unemployment in Australia’, The Economic Record, Vol.72, No. 217, pp.118-129. Hart, P.E. (1988). Youth Unemployment in Great Britain. Cambridge: Press Syndicate. Hashimoto, Masanori and Ross A. Miller (1999). ‘How Do Training and Early Labour Market Experience Affect the Economic Well-being of Youths?’, mimeo, Ohio State University. Lindley, Robert M. (1996). ‘The School-to-Work Transition in the United Kingdom’, International Labour Review, Vol.135, No. 2, pp. 159-180. Lynch, Lisa M. (1987). ‘Individual Differences in the Youth Labour Market: a Cross-section Analysis of London Youths’, in Junankar, P(ed.) (1987). From School to Unemployment? The Labour Market for Young People. London: MacMillan. Main, Brian G. M. and David Raffe (1983). ‘Determinants of Employment and Unemployment Among School Leavers. Evidence from the 1979 Survey of Scottish School Leavers’, Scottish Journal of Political Economy, Vol.30, No. 1, pp. 1-17. Micklewright, John (1989). ‘Choice at Sixteen’, Economica, Vol.56, pp.25-39. Miller, Paul W.(1998). ‘Youth Unemployment : Does the Family Matter?’, The Journal of Industrial Relations, Vol. 40, No. 2, pp. 247-276. O’Higgins, Niall (1994). ‘YTS, Employment, and Sample Selection Bias’, Oxford Economic Papers, Vol.46, pp.605-628. O’Neill, Donal and Olive Sweetman (1998). ‘Intergenerational Mobility in Britain: Evidence from Unemployment Patterns’, Oxford Bulletin of Economics and Statistics, Vol. 60, No. 4, pp.431447.

21