Employment, family union and childbearing decisions in Great Britain

Transcription

1 J. R. Statist. Soc. A (2006) 169, Part 4, pp Employment, family union and childbearing decisions in Great Britain Arnstein Aassve, University of Essex, Colchester, UK Simon Burgess and Carol Propper University of Bristol, UK and Matt Dickson University of Warwick, Coventry, UK [Received January Final revision December 2005] Summary. The paper investigates the life-cycle relationship of work and family life in Britain based on the British Household Panel Survey. Using hazard regression techniques we estimate a five-equation model, which includes birth events, union formation, union dissolution, employment and non-employment events. We find that transitions in and out of employment for men are relatively independent of other transitions. In contrast, there are strong links between employment of females, having children and union formation. By undertaking a detailed microsimulations analysis, we show that different levels of labour force participation by females do not necessarily lead to large changes in fertility events. Changes in union formation and fertility events, in contrast, have larger effects on employment. Keywords: hazards British Household Panel Survey; Employment; Fertility; Marriage; Simultaneous 1. Introduction In the last few decades the UK has seen large changes in patterns of employment and family life. Employment rates for females, which stood at 72% in 2002, were only around 50% in The total fertility rate in the UK fell from a peak of 2.95 in 1964 to 1.73 in Marriage rates have also declined and only been partly offset by higher rates of cohabitation (Berrington and Diamond, 2000). There has been a sharp rise in divorce rates and overall both men and women tend to stay single (i.e. are not in a cohabiting union or married) for considerably longer than in the past (Kiernan, 1998). This change in demographic structures, which is commonly referred to as the second demographic transition (Lesthaeghe and Van de Kaa, 1986; Van de Kaa, 1987), is not unique to Britain, though there is variation across European countries (Billari and Wilson, 2001; Kohler et al., 2002). It is not therefore surprising that research into such transitions has been extensive. An important issue is that these demographic and employment processes cannot be considered in isolation of each other. Quite early on, the demography literature raised the issue of reciprocal relationships between processes of family life and working careers. Cramer (1980) discussed, for instance, Address for correspondence: Arnstein Aassve, Institute for Social and Economic Research, University of Essex, Wivenhoe Park, Colchester, CO4 3SQ, UK. aaassve@essex.ac.uk 2006 Royal Statistical Society /06/169781

2 782 A. Aassve, S. Burgess, C. Propper and M. Dickson the problems of identifying causal effects in studying fertility and employment of females. At the same time economic modelling began to address such interrelationships (McElroy (1985), Willis (1999) and later, for example, Francesconi (2002)). The recognition that processes are very much interrelated and that failing to take this into account explicitly can produce misleading results has over recent years sparked many new developments in empirical research. In this paper we build on the literature on estimation of simultaneous processes (which is reviewed below) to examine the determinants of transitions between demographic and employment states. We specify a simultaneous hazard regression framework, similarly to Lillard (1993), of five processes: birth events, union formation, union dissolution, employment and nonemployment events. In addition we allow for unobserved heterogeneity that is correlated across all five processes. Employment, childbearing and family formation transitions are all specified in separate equations but are estimated in a joint maximum likelihood procedure. This allows us to analyse the interaction of the transitions explicitly, controlling for the potential endogeneity of each transition with respect to all the others. We estimate this model by using data for Great Britain. We use the British Household Panel Survey (BHPS), merging within-panel information (for ) with retrospective information that was provided in 1992 to construct complete histories for employment, childbearing and union formation for all respondents in the panel. We thus create histories stretching back to the 1940s, which enables us to provide estimates for several cohorts of the UK population, and also avoids the problem of left censoring, which is often a problem when using only the panel component. This is the first time that the union formation process has been modelled jointly with fertility and employment for the UK and it allows a unique possibility to investigate the links between all three processes. We use simulations to provide further insight into the dynamics of these processes, which is not necessarily easy to infer from the parameter estimates alone. Joint modelling of several processes involves the use of a relatively large amount of covariates with detailed feed-back between the processes. This makes it particularly difficult to assess the magnitude of how the processes affect each other. Simulation analysis allows us to explore these interrelationships. The organization of the paper is as follows. The next section reviews the related literature. Section 3 presents the data and Section 4 the econometric model. Sections 5 and 6 present the results, and a final section offers some conclusions. 2. Related literature There is a huge literature on each of these demographic and employment processes. Here we focus on literature which examines the interrelationships between the processes. Much of the relevant theoretical considerations stems from the pioneering work by Becker (1960) and Oppenheimer (1988, 1994). Becker analysed childbearing in the framework of consumption goods, where demand for children depends on parents quantity and quality considerations. In the original framework, the quantity of children demanded depends positively on the income of the household, whereas the additional consideration of child quality (i.e. investment in children such as schooling and education) generally has an offsetting effect on the quantity component. The predictions of the original framework depend crucially on the notion of specialization within the household where mothers of dependent children tend to specialize in household activities and husbands in market activities. This stylistic framework is perhaps less applicable in the current setting in the UK where work participation among women with children is currently around 70%. An important extension of the Beckerian framework was provided by Ermisch (1989), in which the demand and timing of having children depend directly on the woman s opportunity costs, which in turn depend on the cost and availability of formal

3 Employment, Family Union and Childbearing 783 and informal child care facilities. An important insight from Ermisch s study is that, in so far as child care is readily available, high earning women (i.e. those with a high opportunity cost) may not demand fewer children, since they can more easily afford child care. The frameworks of Becker and Ermisch consider childbearing decisions only within marriage. Given that most people still perceive a stable union as a precursor for childbearing it is clear that the timing of unions also will have an influence on individuals childbearing careers. Oppenheimer (1988, 1994) argued that an important factor in union formation is men and women s employment careers. For instance, stable employment, and therefore a stable income stream, is one of the fundamental requirements for starting a new household. In this setting it is not necessarily obvious that there should be a certain specialization between men and women: both working men and women might want to contribute to the costly affair of establishing a new household. But the role of employment has other important consequences. Employment structures the individual s life and imposes specific ways of adult socialization according to working hours, place of work, mobility and business relations, and will therefore influence the partnership process itself (Oppenheimer, 1988). This is formalized in economic search models of marriage (Keeley, 1977; Aassve et al., 2002). In this framework, the probability of receiving an offer of marriage, and therefore the likelihood of forming a union, depends on the quality of the individuals in the marriage market. An important characteristic of quality is the potential partner s income, and generally it is observed that those with high income tend to receive more offers of marriage. Highlighting the inherently joint decision-making processes, Willis (1999) suggested that, in an environment of low income male candidates, single women may rationally decide to bring up children outside marriage or a union. However, this line of reasoning is also relevant for dissolution of unions. It has been argued (Becker et al., 1977; Oppenheimer, 1988; Weiss and Willis, 1997) that economic independence among women will reduce the gain from being in a partnership. Thus, individuals with a stable working career and income, who realize that their expected gain from a union is reduced, may actually dissolve the partnership, considering this option more beneficial than continuing it. The theoretical literature suggests that the interrelationships between family life and working careers are complex. Whereas the empirical literature on family life and working careers is vast, the literature which specifically focuses on causal interrelationships is much more modest. One strand of this literature takes a macroperspective (e.g. Mira and Ahn (2002) and Engelhardt et al. (2003)). Using aggregate measures from Organisation for Economic Co-operation and Development countries on females work participation and fertility rates, these studies show that the correlation between the two was clearly negative until the mid-1980s. But from this point onwards the results are more mixed. Mira and Ahn (2001) and Engelhardt et al. (2003) found that the correlation became positive during the 1990s, whereas Kögel (2004) argued that the sign of the correlation did not actually change though it weakened considerably compared with the pre-1980s level. Another strand takes a microperspective to analyse these interrelationships. Studies of fertility, marriage formation (and dissolution) and employment have a long tradition within empirical analysis in social sciences, including economics, demography and sociology. However, joint modelling of these processes, where the interrelationships are modelled explicitly, are more recent. Examples include Lillard and Waite (1993), modelling childbearing and marital dissolution, Brien (1997), investigating the role of the marriage market for different race groups in the USA, Brien et al. (1999) and Baizán et al. (2004), considering the issue of cohabitation, marriage and non-marital childbearing, Upchurch et al. (2002), analysing the interrelationship between union formation, childbearing and educational choice, van der Klauuw (1996), analysing the relationship between work and marriage, Francesconi (2002) on the relationship between full- and part-time work and fertility, Bloemen and Kalwij (2001), on fertility and

4 784 A. Aassve, S. Burgess, C. Propper and M. Dickson employment using Dutch data, and Steele et al. (2004), on partnership and fertility transitions, using the National Child Development Study. Our paper adds to this literature in that employment behaviour is modelled explicitly alongside childbearing and union careers, providing the opportunity to analyse how employment decisions interact with demographic behaviour. 3. Data The data sets that we use are from the BHPS for The first wave of the BHPS was designed as a nationally representative sample of the population of Great Britain living in private households in the autumn of Approximately 5500 households, containing about people, were interviewed. These original sample members were reinterviewed each successive year and, if they split off from their original households to form new households, all adult members of these new households were also interviewed. Similarly, children in the original sample households were interviewed when they reach 16 years of age. In addition to providing information on respondents within the panel survey period (1991 onwards) the BHPS asked respondents to provide detailed retrospective work, family and fertility histories in These retrospective data are matched to the within-panel data to construct detailed marriage, fertility and work histories from age 13 years for all adult respondents. Thus individual-specific behaviour is modelled from this age and avoids the initial condition problem that is normally encountered when estimating duration models that are based on the panel component only. The choice of age 13 years was based on the fertility and employment patterns in the data, on the fact that the school leaving age was 14 years or more for all observations in the sample and that marriage is legal from age 16 years. We have created five detailed event histories for each individual. These are forming and dissolving a partnership, having a(nother) child and entering and leaving employment. The censoring age for childbearing is for women set at 45 years and 55 years for men. Censoring ages for the other processes are determined by respondents recorded age at wave 8 (i.e. in 1999). Since we include individuals who were born in 1940 (or younger) the highest possible censoring age is 59 years. As cohabitation is an increasing form of union in the UK (either as a precursor to legal marriage or as a substitute), we define a partnership (referred to as marriage in the rest of the paper) as living in union with a person of the opposite gender, regardless of legal marital status. (The number of recorded same sex couples in Britain is only around 0.3% of all couples.) For the within-panel data we use self-reported marital status, which takes the following categories: married, living as a couple, separated, divorced, widowed and never married. We classify married and living as a couple as de facto married, with the remaining categories being de facto not married. Individuals are defined as being employed if they are in any paid employment (full or part time) and paid self-employment. Individuals who are on long-term leave due to sickness are classified as not employed. Individuals who are on maternity leave are also classified as not being employed. This definition was imposed by the data, as it was not clear in the data whether some individuals were on maternity leave or not. The number of women who were recorded within the panel element of the BHPS on maternity leave is 0.56% of all women (i.e. around 1 in 200) and the number who were eligible for maternity leave among the older cohorts who are in our sample in the retrospective data at ages of childbearing will be smaller because of a change in maternity leave legislation. An individual is classified as changing employment status only if she or he moves into or out of paid employment. Thus, cases where individuals change employer but remain continuously employed are not classified as a change in employment status. Similarly, individuals changing from full time to part time are recorded as having no

5 Employment, Family Union and Childbearing 785 change in employment status. Individuals moving from full-time education to job seeking are also recorded as having no change of (non-employment) status. We assume that all individuals are in full-time education and therefore non-employed at age 13 years. Births during the panel years are constructed from the household record of the respondent. In the majority of cases, there is only one birth event in the household in a given wave. The retrospective history that was collected in 1992 records the dates of birth of all the respondents children to that date. These data are recoded into a monthly panel of data covering the birth events in each individual s life up to the time of their interview in wave 2 of the BHPS. These data are then merged with the within-panel data to create one event history file, which records the date of the conception of children, where conceptions are assumed to have taken place 9 months before the date of birth (in what follows this is referred to as the date of birth). These data are also used to create stocks of each process as well as durations. For the stock of children, we make the assumption that children leave home at 21 years of age (the median age for leaving home in Britain is at present years; Billari et al. (2001)), and so decrease any positive stock by 1 at the date at which the oldest child will be 21 years old. (This is not an empty assumption as the stock of children does affect the hazards, but the assumption will only take effect for older respondents with children.) In general the data that are contained in the BHPS are of high quality (Lynn, 2003; Dex and McCulloch, 1997). However, it is generally acknowledged that misreporting among men can be a problem, and this may affect both the reported job histories (Elias, 1997) as well as the reported fertility histories (Rendall et al., 1999). Misreporting is more likely to happen for events that have taken place a long time ago; thus the effects of misreporting will be more pronounced among the older cohorts. By imposing certain assumptions about the nature of measurement error, corrections to the statistical analysis can be made. However, given the already complex nature of the model estimates here, this was not attempted. (See Chesher (1991) and Chesher et al. (2002) for further details on the effect of measurement error in duration models.) 3.1. Descriptive statistics Table 1 presents statistics for the transitions and is a description of the cross-sectional data, not a life-table. These descriptive statistics include censored individuals, which means that the statistics are both for individuals who have completed events and those who have not by the time of the last interview. Table 1 gives the mean duration until birth, union formation, union dissolution, employment and non-employment events, all measured in years. In addition, Table 1 indicates the proportions experiencing the events. The data that are presented are the mean times until the nth occurrence of an event, conditional on the (n 1)th occurrence of the event. The figures for the proportions experiencing each event are also for the relevant at-risk sample so, for example, the proportion experiencing a second birth is calculated for all those with one or more births. The relevant figures are given in the second and fourth columns (i.e. the columns shown as sample ). From Table 1 we see that the mean duration until first childbearing is years for women, and years for men, so the mean ages at first birth are and years respectively. The third column shows the mean time until the second birth, starting from the time of the first birth. Thus the mean times until the second childbearing events are 4.36 and 4.62 years for men and women, which correspond to mean ages of 30.9 years for women and 34 years for men. The data indicate that, once the first birth has occurred, both men and women accelerate the timing of the second birth, whereas third or fourth births generally tend to take place at a much later age. About 68% of all women and 58% of all men are recorded as having at least one child.

6 786 A. Aassve, S. Burgess, C. Propper and M. Dickson Table 1. Descriptive statistics for sample and simulated data Results for women Results for men Sample Simulated Sample Simulated Mean time until 1st birth (years) Mean time until 2nd birth (years) Mean time until 3rd birth (years) Mean time until 4th birth (years) Proportion having at least 1 child Proportion having at least 2 children Proportion having at least 3 children Proportion having at least 4 children Mean time until 1st union (years) Mean time until 2nd union (years) Proportion forming at least 1 union Proportion forming 2nd union after dissolution Mean time until 1st union dissolution (years) Mean time until 2nd union dissolution (years) Proportion dissolving at least 1 union Proportion dissolving 2nd union Mean time until 1st employment (years) Mean time until 2nd employment (years) Mean time until 3rd employment (years) Mean time until 4th employment (years) Mean time until 5th employment (years) Mean time until 6th employment (years) Proportion finding employment at least once Proportion finding employment after 1st non-employment Proportion finding employment after 2nd non-employment Proportion finding employment after 3rd non-employment Proportion finding employment after 4th non-employment Proportion finding employment after 5th non-employment Mean time until 1st non-employment spell (years) Mean time until 2nd non-employment spell (years) Mean time until 3rd non-employment spell (years) Mean time until 4th non-employment spell (years) Mean time until 5th non-employment spell (years) Proportion becoming non-employed after 1st employment Proportion becoming non-employed after 2nd employment Proportion becoming non-employed after 3rd employment Proportion becoming non-employed after 4th employment Proportion becoming non-employed after 5th employment Mean durations include censored observations.

7 Employment, Family Union and Childbearing 787 Of those who have at least one child, 76% have a second child. Given two children, relatively few go on to have a third birth. The fact that the data contained censored individuals, together with possible under-reporting among men, will explain some of the gender difference in the rate of parenthood. For union formation, the mean age of first-union formation is years for women and around 26 years for men. The mean time until first-union dissolution is quite long, indicating that a large proportion of those entering the first union tend to remain in this union and, in the majority of cases, until the censoring date (1998). Of those who experience a firstunion dissolution (35% of women and 28% of men) the mean time to the second union is 5.26 years for women and 4.10 years for men. Employment and non-employment transitions, in contrast, are much more frequent events. Both men and women find employment on average at the same age. Almost all women in this sample have at least one spell of employment. The difference between men and women becomes apparent when we look at any subsequent entries into employment, as well as the exit from employment transitions. The mean time for the first employment exit transition is 7.47 years for women and years for men. The mean age of women who exit employment is close to the mean age of having the first child. Furthermore, 80% of women who were employed at least once become non-employed at some stage in their career. The equivalent figure for men is considerably lower at 53%. Once having made an exit from employment, men tend to find new employment considerably quicker than women. For instance, the time until entering second employment for women is on average 4.89 years, whereas for men the mean time is only 1.62 years, and similar discrepancies exist between men and women for subsequent employment transitions. 4. Empirical modelling of the demographic and labour market events Following Lillard (1993) we specify a model of related dynamic discrete choices, where these are defined over childbearing, union formation, union dissolution, employment and non-employment. The model considers the dynamics of these processes jointly and allows the realizations of any of the related processes to enter as time-varying variables in the remaining processes. Each of the processes is specified as a hazard function, which is conditional both on exogenous and endogenous covariates, as well as potentially correlated unobserved heterogeneity components. The set of hazards is ln{h B.t/} = b 1 f {M.t/, D.t/} + b 2 f {E.t/, U.t/} + b 3 T B.t/ + b 4 A B.t/ + b 7 x B + b 8 P B + " B,.1/ ln{h M.t/} = 6 m i B i.t/ + m 7 f {E.t/, U.t/} + m 8 T M.t/ + m 9 A M.t/ + m 10 P M + m 11 x M + " M, i=1.2/ ln{h D.t/} = 6 d i B i.t/ + d 7 f {E.t/, U.t/} + d 8 T D.t/ + d 9 A D.t/ + d 10 P D + d 11 x D + " D,.3/ i=1 ln{h E.t/} = 6 e i B i.t/ + e 7 f {M.t/, D.t/} + e 8 T E.t/ + e 9 A E.t/ + e 10 P E + e 11 x E + " E,.4/ i=1 ln{h U.t/} = 6 u i B i.t/ + u 7 f {M.t/, D.t/} + u 8 T U.t/ + u 9 A U.t/ + u 10 P U + u 11 x U + " U.5/ i=1 where h j, j = B, M, D, E, U, are the hazards of a birth (measured at the time of conception), union formation, union dissolution, employment and non-employment respectively. Individuals

8 788 A. Aassve, S. Burgess, C. Propper and M. Dickson are assumed to be at risk of having the first conception from age 13 years and are consequently starting the childbearing process (i.e. ln{h B.t/}/ at this age. Once the first child has been born, individuals become at risk of having the second conception, once the second child has been born they become at risk of having the third conception, and so on. Thus conceptions are specified within one hazard function. The processes of union formation, union dissolution, employment and non-employment are similar in structure, except that being in a union and single are mutually exclusive, as are employment and non-employment. At age 13 years, which is the start of the union formation and employment processes, individuals are single and not working. As soon as employment has been obtained individuals are at risk of entering the state of non-employment, and as soon as they enter a union they become at risk of dissolution of the union. These events, however, may be repeated several times. For all processes we include controls for the stock of each event (parity) P j, which are implemented as dummy variables, and detailed controls for age effects, which are denoted as A j.t/, and defined as a piecewise linear spline function. More formally A j.t/ is a vector of N j A + 1 spline variables whose coefficients are allowed to differ between intervals separated by N j A nodes. Denoting the nodes as w k, we define the spline variable for the kth interval as A j.t/ = max{0, min.t w k 1, w k w k 1 /}, j.b, M, D, E, U/: By specifying several node points which are not necessarily the same for each of the processes the formulation allows for a variety of patterns of duration dependence. The base-line hazard function T j.t/ is defined in a similar way. The endogenous variables are denoted as B i.t/, a binary indicator taking value 1 if the individual has i children and is 0 otherwise (this stock is decremented by 1 when a child reaches age 21 years), f {M.t/, D.t/} is a binary indicator for marital status and f {E.t/, U.t/} a binary indicator for employment status, all of which are time varying. We also condition on a set of assumed exogenous variables x j. Note that, although the BHPS data contain a wealth of background information for both individuals and households, a very limited set only is available for the period that is covered by the retrospective histories. This limits the number of exogenous covariates that we can include in our estimation. We include completed education (five levels), cohort of birth (in four groups born in the 1940s, 1950s, 1960s and 1970s), parental socio-economic status, ethnic origin and a binary variable indicating whether the respondent lived with both their parents at age 14 years. In conditioning on final education levels, what is being treated as exogenous is the eventual level of education rather than the time-varying attainment to date. For events that take place at an early age, this means that we are conditioning on the future, but for events that take place after completed education (between the ages of 14 and 25 years or so) this is not so. Given that we are estimating the effects of education on life-cycle transitions, this may not have too large an effect on the estimates of marriage and fertility, though it may have more of an effect on the estimated effect of education on employment transitions, as those completing education are less likely to be in employment. Our specification of childbearing and union formation or dissolution deviates somewhat from the norm in the demography literature. It might, for instance, be more intuitive (and is more common) to formulate specific processes according to the order of birth and the order of the union. The reason for this is not only that the base-line hazard is likely to differ by the order of the event but also the explanatory variables may have quite different effects. Our focus on life-cycle relationships based on five related processes comes therefore at a cost. The estimated parameters are not specific to each order of birth and order of unions and employment, so that (for example) the effect of education is the same for the first and all subsequent transitions

9 Employment, Family Union and Childbearing 789 into employment. Though modelling employment transitions jointly with fertility and union formation extends previous work in this literature, it does involve important compromises. One concerns the fact that part-time and full-time work are collapsed into one employment category. Clearly, part-time work might have different influences on fertility events from full-time work, the latter having a lower effect on fertility (see Francesconi (2002)). Another issue concerns the issue of cohabitation and marriage, which again may have different influences on fertility and employment events (see Steele et al. (2005) for an application where cohabitation and marriage are treated as separate events). However, given the already complex nature of the statistical model, we did not distinguish between these states. This is important for the interpretation of the results (Section 5) and we speculate below on the extent to which our results are affected by these restrictions. For each of the five related processes we specify a random heterogeneity component. These will capture unobserved heterogeneity affecting (each of) the processes that is not picked up by the observed covariates. Given that the processes are related, it is likely that there will be correlation between the unobserved heterogeneity terms across the five processes. In essence this means that the time-varying explanatory variables are potentially correlated with the error terms in the other processes. For instance, if union formation and childbearing are related then f {M.t/, D.t/}, the outcome(s) of the functions ln{h M.t/} and ln{h D.t/}, will be correlated with the unobserved heterogeneity component " B in the function ln{h B.t/}. To allow for these various sources of correlation we specify the unobserved heterogeneity components to have a joint normal distribution: " B 0 σ 2 " M B ρ MB ρ DB ρ EB ρ UB 0 ρ BM σm 2 ρ DM ρ EM ρ UM " D " E " U N 0, ρ BD ρ MD σd 2 ρ ED ρ UD :.6/ 0 ρ BE ρ ME ρ DE σe 2 ρ UE 0 ρ BU ρ MU ρ DU ρ EU σu 2 By integrating out over the unobserved heterogeneity components, the observed completed durations and outcomes are independent and can therefore be estimated by maximum likelihood techniques. Identification is ensured by the fact that all events are repeated, whereas the unobserved heterogeneity components are assumed fixed over individuals lifetimes. It is possible that the sources of unobserved heterogeneity will vary over time (e.g. the results of changes in preferences, or of characteristics of the partner or the children). Our model assumes that this is not so. The potentially endogenous variables always enter the other processes as lagged explanatory variables, which ensures identification of their parameters (Maddala, 1983). With these restrictions in place further exclusion restrictions are not formally required (see originally Lillard (1993) and recently Steele et al. (2004) for a similar identification strategy). Finally, in our model we have not allowed separate processes for different orders of births, unions or employment. But this is primarily because of the already high complexity that is invoked by estimating five separate processes, and not because of identification restrictions. In principle we could estimate separate processes, thereby providing separate parameter estimates, for first, second and higher order births, as long as the error component remains the same for all birth processes. A similar argument applies to the union and employment processes. 5. Estimation results This section presents the parameter estimates of the econometric model (Tables 2 6).

10 790 A. Aassve, S. Burgess, C. Propper and M. Dickson Table 2. Parameter estimates for fertility transitions Results for the following models: Model 1, women, no Model 2, women, with Model 3, men, no Model 4, men, with heterogeneity heterogeneity heterogeneity heterogeneity Exogenous variables Cohort (0.0386) (0.0636) (0.0421) (0.0689) Cohort (0.0432) (0.0686) (0.0550) (0.0832) Cohort 1970 onwards (0.0977) (0.1315) (0.1475) (0.1753) Did not live with both parents at 14 years (0.0413) (0.0701) (0.0521) (0.0832) Father s professional occupation (0.0445) (0.0700) (0.0477) (0.0718) Mother s professional occupation (0.0556) (0.0927) (0.0643) (0.0997) Ethnic origin (0.0648) (0.1170) (0.0778) (0.1346) Below O-level qualification (0.0550) (0.0917) (0.0709) (0.1108) O-levels or equivalent (0.0491) (0.0813) (0.0566) (0.0885) A-levels or equivalent (0.0469) (0.0804) (0.0505) (0.0799) Higher qualification (0.0573) (0.0998) (0.0667) (0.1071) Endogenous time-varying variables Married or cohabiting (0.0500) (0.0598) (0.0600) (0.0782) Working (including part time) (0.0434) (0.0505) (0.0751) (0.0893) Event order 2nd birth (0.0902) (0.1221) (0.1243) (0.1672) 3rd birth (0.0995) (0.1444) (0.1352) (0.2044) 4th, 5th or 6th births (0.1219) (0.1888) (0.1596) (0.2559) Significant at the 1% level. Significant at the 5% level. Significant at the 10% level Background variables and education Non-time-varying background variables The non-time-varying background variables show that family background is an important determinant for an individual s family formation events, whereas it has less influence on the employment transitions. Both men and women coming from disrupted families have more children and form and dissolve unions quicker. The effects are weaker for exits from employment, although we find that women from disrupted families are more likely to leave employment, which is consistent with the fact that they have more children. Men and women from non-white ethnic

11 Employment, Family Union and Childbearing 791 Table 3. Parameter estimates for entering a union Results for the following models: Model 1, women, no Model 2, women, with Model 3, men, no Model 4, men, with heterogeneity heterogeneity heterogeneity heterogeneity Exogenous variables Cohort (0.0507) (0.0772) (0.0549) (0.0777) Cohort (0.0545) (0.0778) (0.0593) (0.0827) Cohort 1970 onwards (0.0825) (0.1079) (0.0998) (0.1226) Did not live with both parents at 14 years (0.0534) (0.0777) (0.0588) (0.0856) Father s professional occupation (0.0510) (0.0734) (0.0525) (0.0736) Mother s professional occupation (0.0658) (0.0956) (0.0795) (0.1068) Ethnic origin (0.0914) (0.1217) (0.1107) (0.1474) Below O-level qualification (0.0760) (0.1113) (0.0931) (0.1242) O-levels or equivalent (0.0605) (0.0877) (0.0773) (0.1049) A-levels or equivalent (0.0600) (0.0873) (0.0638) (0.0884) Higher qualification (0.0709) (0.1063) (0.0787) (0.1073) Endogenous time-varying variables 1st birth (0.0513) (0.0661) (0.0555) (0.0849) 2nd birth (0.0894) (0.1019) (0.1061) (0.1276) 3rd birth (0.1388) (0.1493) (0.2085) (0.2388) 4th birth (0.2461) (0.2712) (0.3646) (0.4548) 5th and 6th births (0.3530) (0.3801) (0.6545) (0.6903) Working (including part time) (0.0544) (0.0633) (0.0753) (0.0880) Event order 2nd marriage (0.1670) (0.2238) (0.2321) (0.2981) 3rd or 4th marriage (0.2097) (0.2945) (0.2486) (0.3662) Significant at the 5% level. Significant at the 1% level. Significant at the 10% level.

12 792 A. Aassve, S. Burgess, C. Propper and M. Dickson Table 4. Parameter estimates for union exits (union dissolutions) Results for the following models: Model 1, women, no Model 2, women, with Model 3, men, no Model 4, men, with heterogeneity heterogeneity heterogeneity heterogeneity Exogenous variables Cohort (0.0974) (0.1256) (0.1173) (0.1401) Cohort (0.1142) (0.1478) (0.1378) (0.1815) Cohort 1970 onwards (0.1645) (0.2204) (0.2217) (0.2724) Did not live with both parents at 14 years (0.0878) (0.1189) (0.1076) (0.1322) Father s professional occupation (0.0865) (0.1048) (0.1010) (0.1193) Mother s professional occupation (0.1189) (0.1401) (0.1354) (0.1564) Ethnic origin (0.1532) (0.1848) (0.2449) (0.2763) Below O-level qualification (0.1327) (0.1614) (0.1967) (0.2244) O-levels or equivalent (0.1089) (0.1367) (0.1511) (0.1775) A-levels or equivalent (0.1042) (0.1308) (0.1337) (0.1541) Higher qualification (0.1197) (0.1526) (0.1516) (0.1752) Endogenous time-varying variables 1st birth (0.1020) (0.1114) (0.1116) (0.1219) 2nd birth (0.1031) (0.1190) (0.1252) (0.1451) 3rd birth (0.1184) (0.1324) (0.1958) (0.2157) 4th, 5th and 6th births (0.1747) (0.2044) (0.2593) (0.2954) Working (including part time) (0.0775) (0.0889) (0.1151) (0.1352) Event order 2nd dissolution (0.1090) (0.3061) (0.1160) (0.2895) 3rd or 4th dissolution (0.2293) (0.5252) (0.1956) (0.5049) Significant at the 1% level. Significant at the 5% level. Significant at the 10% level. origins show quite different behaviour from that of the rest of the population. They generally tend to delay union formation but have a considerably higher fertility rate once in a union. They also have a lower rate of entering employment, whereas there is no difference in terms of leaving employment. Parental occupational status has only a limited effect. It is strongest in the fertility transitions, where a high parental socio-economic status for males is associated with

13 Employment, Family Union and Childbearing 793 Table 5. Parameter estimates for entering employment Results for the following models: Model 1, women, no Model 2, women, with Model 3, men, no Model 4, men, with heterogeneity heterogeneity heterogeneity heterogeneity Exogenous variables Cohort (0.0358) (0.0390) (0.0396) (0.0508) Cohort (0.0409) (0.0445) (0.0476) (0.0608) Cohort 1970 onwards (0.0599) (0.0635) (0.0549) (0.0694) Did not live with both parents at 14 years (0.0371) (0.0410) (0.0457) (0.0582) Father s professional occupation (0.0349) (0.0379) (0.0425) (0.0532) Mother s professional occupation (0.0465) (0.0516) (0.0599) (0.0744) Ethnic origin (0.0637) (0.0702) (0.0799) (0.0967) Below O-level qualification (0.0493) (0.0550) (0.0607) (0.0796) O-levels or equivalent (0.0434) (0.0491) (0.0478) (0.0646) A-levels or equivalent (0.0430) (0.0472) (0.0431) (0.0564) Higher qualification (0.0489) (0.0533) (0.0689) (0.0862) Endogenous time-varying variables 1st birth (0.0513) (0.0529) (0.0715) (0.0783) 2nd birth (0.0542) (0.0554) (0.0770) (0.0835) 3rd birth (0.0626) (0.0652) (0.0958) (0.1041) 4th birth (0.1063) (0.1119) (0.1497) (0.1547) 5th and 6th birth (0.2451) (0.2462) (0.4283) (0.4064) Married or cohabiting (0.0399) (0.0466) (0.0526) (0.0659) Event order 2nd employment (0.0629) (0.0761) (0.1082) (0.1240) 3rd employment (0.0772) (0.0993) (0.1188) (0.1454) 4th or higher employment (0.0866) (0.1190) (0.1235) (0.1544) Significant at the 1% level. Significant at the 5% level. Significant at the 10% level.

14 794 A. Aassve, S. Burgess, C. Propper and M. Dickson Table 6. Parameter estimates for employment exits (entering non-employment) Results for the following models: Model 1, women, no Model 2, women, with Model 3, men, no Model 4, men, with heterogeneity heterogeneity heterogeneity heterogeneity Exogenous variables Cohort (0.0396) (0.0808) (0.0684) (0.0957) Cohort (0.0451) (0.0831) (0.0779) (0.1151) Cohort 1970 onwards (0.0622) (0.1057) (0.0940) (0.1440) Did not live with both parents at 14 years (0.0425) (0.0782) (0.0619) (0.0874) Father s professional occupation (0.0371) (0.0719) (0.0613) (0.0835) Mother s professional occupation (0.0451) (0.0939) (0.0832) (0.1112) Ethnic origin (0.0665) (0.1271) (0.1197) (0.1536) Below O-level qualification (0.0539) (0.1093) (0.0937) (0.1376) O-levels or equivalent (0.0505) (0.0976) (0.0730) (0.1093) A-levels or equivalent (0.0458) (0.0911) (0.0612) (0.0970) Higher qualification (0.0508) (0.1012) (0.0895) (0.1339) Endogenous time-varying variables 1st birth (0.0386) (0.0512) (0.0759) (0.0826) 2nd birth (0.0492) (0.0564) (0.0833) (0.0944) 3rd birth (0.0702) (0.0810) (0.1018) (0.1183) 4th birth (0.1027) (0.1268) (0.1713) (0.2012) 5th and 6th birth (0.2146) (0.2553) (0.3277) (0.4355) Married or cohabiting (0.0415) (0.0510) (0.0624) (0.0747) Event order 2nd non-employment (0.0473) (0.0773) (0.0602) (0.1211) 3rd non-employment (0.0617) (0.1106) (0.0830) (0.1854) 4th or higher non employment (0.0714) (0.1442) (0.0962) (0.2419) Significant at the 1% level. Significant at the 5% level.

15 Employment, Family Union and Childbearing 795 a significant reduction in the fertility rate. Mother s occupational status has very little or no effect. The effect of father s occupational status on entering employment is negative for both men and women. This is somewhat surprising given that we also control for educational attainment. A possible explanation is that the father s occupational status is positively correlated with family income. The negative effect suggests that young individuals from wealthy family backgrounds tend to delay entry into employment, which might be due to longer time spent in education Education Note that we treat education as exogenous, though properly it should be modelled as part of the set of choices that individuals make about their post-compulsory school-leaving behaviour. Most of the estimates are consistent with previous findings, although we find some unexpected non-linear effects. For fertility, we find that highly educated women have a lower rate of childbearing. There is a monotonic gradient for the five categories of educational attainment. The influence of education on having children is weaker for men and only men with the highest educational attainment have a fertility rate that is significantly lower than that of the other education groups. For union formation, the effect is very similar to those of fertility, although the magnitude of the effects is generally smaller. In contrast, for union dissolution, education has no significant influence. In terms of entering employment, the results are more mixed. Women with medium levels of qualification have a higher rate of entry into employment compared with those with lower education. Women with very high educational attainment have a considerably lower transition rate into employment than all other groups. For men the effect is slightly different. Low and medium levels of education do not seem to have any differential effect on entry into employment. However, as for women, those with the highest educational attainment have a lower rate of entry into work. Higher education among men is negatively associated with leaving employment, suggesting that, although highly educated men are less likely to leave employment, once they have left, they spend longer searching for the next job. For women the educational effect on exit from employment is also negative, but there is no monotonic gradient. The lower rate of entry into employment for the best educated is probably driven by the later entry into initial employment of the more highly educated and the assumption of our model that exogenous variables have the same effect on all transitions within processes Differences between cohorts The differences between cohorts are as expected. Those in the more recent cohorts have lower fertility rates and a lower rate of entering unions than those in the earliest one. Once in a union, those in the recent cohorts are considerably more likely to dissolve the union, compared with those in the oldest cohort. Overall those in the recent cohorts tend to spend longer being single. The results are more mixed for the entry into and exit from employment. Among women we see that the women from the second earliest cohort (those born in the 1950s) enter employment at a higher rate than those in the earliest cohort, whereas there is no significant difference between the two most recent cohorts (those who were born in the 1960s and 1970s). For men, compared with the earliest cohort, those in the more recent cohorts all have lower transition rates into employment. For both genders there are clear positive gradients for exits from employment, indicating that those in the recent cohorts have higher transition rates out of work than those in the earliest cohort. As the model does not include the effects of macrovariables, such as the trade cycle, we cannot distinguish this cohort effect from the effects of the economic cycle. The base-line hazards and the effect of age were discussed in Aassve et al. (2004).

16 796 A. Aassve, S. Burgess, C. Propper and M. Dickson 5.2. Time-varying variables The effect of marital status and employment status on childbearing Marital status defined here includes cohabitation, and employment includes full-time and parttime work. The estimates in Table 2 indicate that being in a union has a large positive effect on fertility events (particularly for men), and that the effect remains strong when controlling for unobserved heterogeneity. As we would expect, being in employment has a negative effect on childbearing but, although the parameter estimate is highly significant, it is not extremely large, implying that working is not a particularly strong deterrent to having children. The relatively weak effect most likely reflects the fact that part-time and full-time work are incorporated in the same category. It is possible, for instance, that women in full-time work have a much lower fertility rate than women working part time. The positive effect of employment for men on having children fits with previous findings. The parameter is highly significant, but again the magnitude is somewhat small. Note again that the parameter estimate here averages over all orders of birth so the effect might have been stronger for the timing of the first birth, and even weaker for subsequent births Childbearing on union formation We can see that the effect very much depends on the order of birth. The birth events represent the stock of children. For instance, experiencing a first birth has a strong positive effect on forming a union, and this is so for both genders. However, having a second birth outside a union actually lowers the rate of union formation. The positive effect of the first birth event is consistent with economic theory, in that individuals consider a cohabiting union or a marriage to be more beneficial once they have acquired marital-specific capital. However, there might also be normative forces at play, in the sense that individuals might feel a pressure to legitimize the child. The negative sign of the second birth event indicates that those who do not form a union after the first birth are at a disadvantage in the marriage market when they have the second child. There might also be a selection effect here: individuals who are willing to have one child outside a union might be more willing to repeat the experience. The subsequent birth events have no significant effect on union formation. Turning to the union dissolution hazard, we find parameter estimates that are consistent with our expectations. The negative effect of the first and second births on dissolution indicates the role of children as marital-specific capital. Our specification does not include duration splines for the birth events, so we do not examine the effect of the age of the children on the rate of dissolution (see, for instance, Lillard and Waite (1993) who showed how dissolution depends on the age of the children). The influence of children becomes stronger once we control for unobserved heterogeneity. In fact, the negative effect of the second birth event is only significant for women when unobserved heterogeneity is included. The third birth event does not have any statistically significant effect on dissolution, whereas higher birth orders generally have a positive effect, but these variables are not particularly well defined owing to the small sample sizes Work status on union formation, union dissolution and fertility Work status has a positive and highly significant effect on union formation for both men and women, a finding which is consistent with most previous research (see Oppenheimer (1994) for a review). The effect of work status on divorce is not particularly strong, especially for men, independently of whether unobserved heterogeneity is controlled for. For women, in contrast, work has a positive effect only when we control for unobserved heterogeneity. The rate of entering employment is negatively associated with a first birth event. Although the effect is negative for both genders, it is considerably weaker for men. This negative effect for men is somewhat surprising, as the financial costs that are associated with childbearing, and the traditional divi-