The influence of living arrangements, marital patterns and family configuration on employment rates among the 1945–1954 birth cohort: evidence from ten European countries

Abstract

As they approach retirement, Europeans in mid-life display a range of living arrangements and marital patterns. These configurations influence labour force participation for men and women in different ways and these differences are accentuated between countries. Using data from the first Wave (2004) of the Survey on Health, Ageing and Retirement in Europe (SHARE), the paper examines the relationship between living arrangements, marital patterns, family configurations and participation in the labour force for the birth cohort of 1945–1954. The data show that the probability of being in paid employment was higher for respondents living in a couple in northern Europe than in southern Europe. In all countries, men in a couple had significantly higher employment rates than women in a couple, but employment rates of women in a couple differed significantly between countries. Multivariate analysis with country effects confirmed the negative influence of age, poor health, lower levels of education and household income on the probability of being in paid employment, but the effect of variables concerning living arrangements, marital patterns and family configurations varied according to country. A multilevel analysis showed that the between country variance of being in paid employment could not be explained by individual characteristics alone, that a large part of the country variance could be explained by the country specific effect of women in a couple, and that the level of ‘modern’ life styles in each country (rates of cohabitation outside marriage, divorce or separation and recomposed families) had a significant effect on employment rates, especially for women in a couple.

Appendices

Appendix 1: Steps in the multilevel analysis

Multilevel models for discrete response data: notation for two-level models

• Explanatory variables: X

•  Fixed parameters of explanatory variables

  • Individual characteristics: Xij for level 1, individual i;

  • Macro variables: Xj for level 2, country j

• Random parameters of individual characteristics, level 2 (country j).

Model 5a, we observed that the probability of being in employment varies across the countries.

$$ \begin{aligned} & \hbox{Logit} (\Pi_{ij}) = Y_{ij} = \beta_{0{j}}\\ & \beta_{0{j}} = \beta_0+ u_{0{j}}\quad \sigma_{{u}0}^2 = \hbox{Var} ({u}_{0{j}}) \end{aligned} $$

→ between country variance: σ ²_u0 =  0.213.

β_0j the random intercept consists of two terms: β₀ a fixed component and u _0j the random effect (country specific component). u _0j follows a normal distribution with mean zero and variance σ ²_u0 .

Model 5b, we observed that explanatory variables have an effect on the probability of being in employment, but we conclude that these effects were the same for each country.

$$ \begin{aligned} \,& {Y}_{ij} = \beta_{0{j}} + \beta_1 {X}_{1{ij}} + \beta_2 {X}_{2{ij}} + \beta_3 {X}_{3ij {\rm (woman\, in\, a\, couple)}} + \cdots+\beta_{n} {X}_{nij}\\ \,& \beta_{0{j}} = \beta_0 + {u}_{0{j}} \quad\sigma_{{u}0}^2 = \hbox{Var} ({u}_{0{j}}) \end{aligned} $$

→ between country variance: σ ²_u0 =  0.268.

The fixed parameter (level 1) for individual characteristic ‘woman in a couple’ is: β₃ =  −1.150 (0.061).

Model 5c consisted in allowing a random effect for ‘woman in a couple’; in others words, the difference between the category ‘woman in a couple’ and the other three categories within a country varies across the countries.

$$ \begin{aligned} \,& {Y}_{ij} = \beta_{0{j}} + \beta_1 {X}_{1{ij}} + \beta_2 {X}_{2{ij}} + \beta_{3{j}} {X}_{3ij {\rm (woman\, in \,a \,couple)}} + \cdots + \beta_{n} {X}_{nij}\\ \,& \beta_{0{j}} = \beta_0 +{u}_{0{j}} \quad\sigma_{{u}0}^2 = \hbox{Var} ({u}_{0{j}})\\ \,& \beta_{3{j}} = \beta_3 +{u}_{3{j}} \quad\sigma_{{u}3}^2 = \hbox{Var} ({u}_{3{j}}) \end{aligned} $$

→ between country variance: σ ²_u0 =  0.150.

The fixed parameter (level 1) for individual characteristic ‘woman in a couple’ is: β₃ =  −1.148 (0.167) (there is a strong rise of the standard error from 0.061 to 0.167 which means there is a random effect (for country, level 2): β_3j  = β₃ + u _3j ).

Because of this random effect for ‘woman in a couple’, the parameter σ _u30 was introduced: it is the covariance between u _0j and u _3j .

The residual variance between countries is a function of explanatory variables that have random coefficients: Var (u _0j +  u _3j ) = Var (u _0j ) +  2Cov(u _0j +  u _3j ) +  Var (u _3j )² =  σ ²_u0 +  2 σ _u30 +  σ ²_u3 .

→ residual country variance for ‘woman in a couple’: σ ²_u3 =  0.526.

So there is a greater country level variation in the probability of being in employment for ‘woman in a couple’ (σ ²_u3  = 0.526) than in other situations (‘person not in a couple, man in a couple’) (σ ²_u0  = 0.150).

Model 5d, we added a country level explanatory variable (contextual variable) to see whether ‘modern life style’ explains some of the country level variation for ‘woman in a couple’ and others.

$$ \begin{aligned} \,& {Y}_{ij} = \beta_{0{j}} + \beta_1 {X}_{1ij} + \beta_2 {X}_{2ij} + \beta_{3j} {X}_{3ij {\rm (woman\, in\, a \,couple)}} +\cdots+ \beta_{n} {X}_{nij} + \beta_{{n}+1} {X}_{n+1 j {\rm (Modern\, style)}}\\ \,&\beta_{0{j}}=\beta_0+{u}_{0{j}} \quad\sigma_{{u0}}^2=\hbox{Var} ({u}_{0{j}})\\ \,&\beta_{3j}=\beta_3+{u}_{3j} \quad\sigma_{{u}3}^2=\hbox{Var} ({u}_{3j}) \end{aligned} $$

→ between country variance: σ ²_u0 =  0.112.

The fixed parameter (level 2, country) for the macro variable ‘modern life style’ is: β_n+1 =  0.1.646 (0.242).

The proportion of ‘Modern life style’ within countries has a positive and significant effect on the probability of being in employment.

→ residual country variance for ‘woman in a couple’: σ ²_u3 =  0.101.

So there is a large decrease in country level variation between the probability of being in employment for ‘woman in a couple’, from σ ²_u3  = 0.526 in model 3c to σ ²_u3  = 0.101 in model 3d.

The reduction of country variance is 80% for the category ‘woman in a couple’ and 25% for the others.

Model 5e, in the same way, we observed another contextual variable—‘the public expenditure on labour market’ to see whether it explains some of the country level variation for ‘woman in a couple’ and other categories.

$$ \begin{aligned} \,& {Y}_{ij} = \beta_{0{j}} + \beta_1 {X}_{1ij} + \beta_2 {X}_{2ij} + \beta_{3{j}} {X}_{3ij {\rm (femme\, en\, couple)}} +\cdots+ \beta_{n} {X}_{n ij} + \beta_{ n+2} {X}_{n+2 j {\rm (public\, expenditure\, on\, labour\, market)}}\\ \,& \beta_{0{j}}=\beta_0+{u}_{0{j}} \quad\sigma_{{u}0}^2= \hbox{Var} ({u}_{0{j}})\\ \,& \beta_{3j}=\beta_3+{u}_{3j} \quad \sigma_{{u}3}^2= \hbox{Var} ({u}_{3j}) \end{aligned} $$

→ between country variance: σ ²_u0 =  0.099.

The fixed parameter for the macro variable public expenditure on labour market is: β _n  + 2 =  0.889 (0.407).

→ residual country variance for ‘woman in a couple’: σ ²_u3 =  0.323.

In this last model, the reduction of country variance is 38% for the category ‘woman in a couple’ (from σ ²_u3  = 0.526 in model 4c to σ ²_u3  = 0.101 in model 4e) and 34% for the others (σ ²_u3  = 0.150 in model 4c fell to σ ²_u3  = 0.099 in model 4e).

Appendix 2

Table 6

Table 6 Binary logistic regression model and multilevel analysis, two-level models: probability of being in employment in 2004 in nine European countries, 1945–1954 birth cohort