3.1 Occupation and occupational sex segregation in the German labour market

Occupations can be described as “institutionalised structures that define a job in terms of its contents, differentiate and structure it” (Brater 2010, p. 805). They are specialisations that arise as part of the division of labour and take the form of bundles of skills learned in education or training (Sackmann and Rasztar 1998, p. 26). At the general societal level, occupations thus have the function of assigning workers to jobs (Abraham et al. 2011), and thereby contribute to social stratification (Weber 1972).

Especially in the German labour market, an individual’s occupation has a lasting impact on the entire life course and career trajectory. Through the dual vocational training system, occupations closely link the education system to the labour market (Solga and Konietzka 1999). Even tertiary education in Germany has a strong occupational focus (Leuze 2010). Accordingly, labour market entry is characterised by a high skills and status match between the apprenticeship a person completes and her first job (Konietzka 1999; Solga and Konietzka 1999). Therefore, changing occupation increases downward mobility if an individual does not hold the necessary credentials (Konietzka 1999). Overall, switching occupation entails the loss of occupation-specific human capital, resulting, for example, in wage penalties (Mertens 1997) and discontinuous earning trajectories (Hacket 2009). Accordingly, occupational mobility is low in Germany (Allmendinger and Hinz 1997).

In fact, to highlight their great importance for both the German labour market and individual action, occupations could be understood as institutions (Stuth et al. 2009; Abraham et al. 2011). This means that occupations represent institutionalised rules that “harmonise and coordinate qualifications and job requirements” (Abraham et al. 2011, p. 6). This reduces uncertainty between employer and employee in the labour market and thus facilitates the matching process (Abraham et al. 2011). In this context, the occupational title not only signals that the bearer has a specific skills profile; it also signals specific norms and values that are communicated during vocational education. The rules of the occupation limit an occupation member’s scope of action and influences his decisions (Stuth et al. 2009). Hence, occupations fulfil another central property of institutions: they define and limit the set of choices of individuals (North 1998, p. 4). Applying this to (family-related) employment interruptions, this means that not only the different standards and rules implicit in an occupation, but also the associated institutionally regulated requirements and conditions may promote or impede women’s return to the labour market (Stuth et al. 2009). Hence, occupations and occupational characteristics should influence the decision to return to the labour market in different ways.

Employees are not randomly distributed among different occupations; instead it emerges that “decisions to choose an occupation [are] structured socially, i. e. they are influenced by social categories such as gender and social background” (Abraham et al. 2011, p. 1). In particular, the unequal distribution of men and women across different occupations is a persistent and characteristic feature of the German labour market² that has changed only marginally over the past 35 years despite the increase in the female employment rate (Hausmann and Kleinert 2014).³ Occupational sex segregation is systematically linked to different opportunities and conditions in the labour market (Achatz et al. 2005; Anker 1997; Busch 2013; Falk 2005; Liebeskind 2004; Leuze and Strauß 2016). This is illustrated by the fact that employees in occupations with a high proportion of women are confronted with lower wages, fewer promotion opportunities and higher part-time rates (Binder 2007; Busch and Holst 2009; Liebeskind 2004). The theory of compensating differentials suggests a direct relationship between these factors: according to the theory, women systematically choose occupations with higher part-time rates so that they can better combine job and family, in turn accepting lower wages in these occupations (Becker 1985; Filer 1985). However, the assumption that more family-friendly working hours compensate for lower wages in women’s occupations is highly contested in literature (Budig and England 2001 u. a.; Glauber 2011, 2012; Jacobs and Steinberg 1990).

3.2 The impact of occupational sex segregation on the duration of family-related employment interruptions

What incentives and limits do these occupational characteristics set regarding the ability to reconcile job and family following the birth of a child? Numerous studies show that occupations with higher proportions of women also pay lower wages than male-dominated occupations (for Germany see e. g. Achatz et al. 2005; Busch and Holst 2013; Hinz and Gartner 2005). The wage level in an occupation reflects the general income prospects it offers. Based on the economic theory of labour supply (Franz 2003; Killingsworth and Heckman 1986; Blau et al. 2010), these income prospects should have a direct influence on mothers’ employment decisions. According to this theory, the decision to offer one’s labour is based on an individual cost-benefit consideration: individuals compare the value of their time invested in the labour market to the value of time spent at home, for example, free time, time for domestic work or child-raising. Consequently, an individual will choose to participate in the labour market if this generates a greater value (Blau et al. 2010). Accordingly, with regard to family-related career breaks, mothers should return to the labour market if this allows them to generate a greater benefit than continuing the employment interruption would.

The benefit of labour market participation is not only influenced by the wage received before the employment interruption⁴, but in particular by the wage a person can expect to earn by returning to the labour market. The higher this expected wage is, the sooner a mother is expected to return to work after childbirth (Ziefle and Gangl 2014). Especially in Germany, the expected wage depends heavily on the income prospects of an occupation, that is, on the average wage level in an occupation. The same holds true for the opportunity costs of an employment interruption, which refers to the loss of wages due to non-employment. The higher the potential income and thus the potential opportunity costs, the more likely it is that mothers should return to work. Accordingly, a lower wage level, as is typical in occupations dominated by women, is expected to reduce mothers’ propensity to return to work. After all, mothers in these occupations not only have less to lose from an interruption; they also have less to gain by returning to the labour market. These considerations lead to our first hypothesis:

In addition to wages, working hour arrangements are also related to the gender composition of an occupation. In occupations with a high proportion of men, working overtime, constant availability and business trips are standard requirements (Crompton and Lyonette 2011; Busch 2013; Leuze and Strauß 2016; Williams 2000), whereas working part-time is associated with insecure employment and is considered harmful to the career (Busch 2013). In contrast “many female-dominated occupations are conceived as part-time jobs” (Busch 2013, p. 280) allowing women to reconcile gainful employment and family obligations. Accordingly, occupations with a high proportion of women have higher rates of part-time employees (Binder 2007). The part-time rate in an occupation is an indicator of the ability to reconcile work and family or of whether an occupation is generally suitable for working part-time.⁵ The option to be employed on a part-time basis should provide for a quicker return to the labour market. Therefore, our second hypothesis says:

While the part-time rate of an occupation addresses concrete options to achieve work-family balance, the proportion of women itself can reflect the social acceptance of violating male-influenced, traditional employment histories and work standards (Williams 2000). Usually, it is still women who take a break from gainful employment after a child is born (Statistisches Bundesamt 2015). Accordingly, employees in occupations dominated by women, where a lot of people are working who anticipate employment interruptions themselves, should show solidarity towards mothers, encourage them to use such work-life policies and support them when they return to work (Blair-Loy and Wharton 2002).

The proportion of women in an occupation also increases the occurrence of work-family conflicts resulting from the dual socialisation (“doppelte Vergesellschaftung”, Becker-Schmidt 2008, p. 66) of women into the worlds of gainful employment and family. The standard of the “ideal worker” (Williams 2000) is challenged more in occupations with a high proportion of women, because absolute dedication to the job, non-stop employment, all-time availability and employee flexibility are based on access to family work, which men tend to have and women usually do not (Williams 2000). A lot of women therefore find it impossible to concentrate on employment alone (Cech and Blair-Loy 2014).

Against this background, a high proportion of women in an occupation should signal an occupation and work culture where the dedication of a (male) “ideal worker” is not taken for granted, and deviations from this standard are punished less severely regardless of the wage structure and the possibilities to reconcile work and family in this occupation. The pressure to return quickly to the workplace is hence lower, leading to the following hypothesis:

Besides the mechanisms discussed here, which lead occupational characteristics to have a direct impact on mothers’ decisions to return to work, self-selection effects that operate for specific groups of women in different occupations could also drive the different lengths of the motherhood breaks observed in different occupations. For example, family-oriented women might systematically choose occupations dominated by women, whereas career-oriented women may tend to opt for male-dominated occupations. If that was the case, different interruption behaviours by women in occupations dominated by men or women would not necessarily be a result of the different characteristics of these occupations, such as the wage level or part-time rate, but instead be the result of mothers’ personality traits and attitudes. However, previous research challenges whether job selection processes are actually influenced by family-related attitudes and expectations. For example, Hakim (2002) does not find any connection between women’s preferences for family or employment and the selection of female- or male-dominated occupations. Ochsenfeld (2016) among others finds no indication that university subject choice is determined by the anticipation of gendered family roles (see also Lörz and Schindler 2011). Additionally, research has shown that actual occupational changes in the context of starting a family are not necessarily based on the gender composition of the occupations (Trappe and Rosenfeld 2004) and that occupations dominated by women have a similar probability of being populated by women with and without children (Roos 1985). On the whole, the empirical results suggest that family-oriented and career-oriented women do not necessarily choose occupations with a different proportion of women. This supports the assumption that the occupational characteristics described above represent incentives or impediments to resuming gainful employment and that different interruption periods from different occupations are not preferentially based on self-selection.

4.1 Data description

The data used in this study consists of the first wave of starting cohort 6 (adults) of the National Educational Panel Study (NEPS).⁶ This survey contains retrospective information on the life course of 11,649 persons born between 1944 and 1986. The respondents stated their education and employment history on a monthly basis and also reported any periods of unemployment or parental leave (Blossfeld et al. 2011). To analyse the duration of family-related employment interruptions, the data set is initially restricted to women who have had at least one child. Moreover, we focus on women who are closely attached to the labour market to counteract distortions caused by heterogeneous employment inclinations and thus to absorb possible selection effects. Consequently, the sample consists of women who had been employed without any interruption for at least 15 h per week⁷ for a minimum of six months before their first child was born and whose last job ended within six months prior to giving birth. Furthermore, we restrict our analysis to mothers who had their first child between 1992 and 2010⁸ in West Germany⁹. During this time span, one aspect of parental leave regulation central to our study, namely job protection during leave, did not change and was consistently at 36 months (Drasch 2012).¹⁰

For our analysis, we supplemented the individual life course data from the NEPS with data on the occupational level. For this purpose, an occupational panel data set, named “OccPan”, is used, which was created on the basis of the Sample of Integrated Labour Market Biographies (SIAB) (vom Berge et al. 2013)¹¹. The SIAB data represent a 2% sample of the Integrated Employment Biographies (IEB), process-generated data of the Institute for Employment Research (IAB) of the Federal Employment Agency (BA). This data is based on the annual mandatory notifications by employers to the social security agencies and provides exact daily information on the employment status of all employees in Germany who are covered by social security.¹² This reliable individual data is used to aggregate annual information, such as the proportion of women or the average wage level at the occupational level. The final panel comprises occupational characteristics for 254 different occupations per year (for a detailed documentation of the generation of the occupation panel “OccPan”, see Hausmann et al. (2015)). Finally, this data basis was further supplemented by central, occupational level characteristics from the German Microcensus (1991 to 2010)¹³ to increase the informational value of the data. Thus, aggregated information on working hours in different occupations are added based on the Microcensus.

4.2 Methodological approach

To study the impact of occupational sex segregation and associated occupational characteristics on the duration of women’s employment interruptions after their first child is born, we employ event history analysis. The relevant event is the resuming of employment by the mother, which marks the end of the employment interruption. Event history analysis makes it possible to calculate the probability of occurrence of the event, which is, the return to the labour market, by considering the time that has already passed.

One advantage of this type of analysis is that it allows us to include (right) censored observations (Allison 1982). Accordingly, the analysis also includes mothers whose employment interruption has not yet ended by returning to the labour market. The method moreover enables the use of time-varying independent variables. Consequently, when external circumstances or living conditions at the start of the interruption episode change over the course of the employment break, e. g. the childcare enrolment rate, this can be taken into consideration.

The individual level data of the NEPS is recorded on a monthly basis, which justifies the use of event models for discrete time units (Singer and Willett 2003). In real life, the return to work after a family-related employment interruption can take place at any time during a month (BMFSFJ 2012).¹⁴ Accordingly, it is a process that was recorded discretely, but happens continuously. For this case, Singer and Willett (2003) recommend the usage of a complementary log-log model.¹⁵ In the logistic regression, the odds ratios represent the ratio of the odds of event occurrence in one group to the odds of event occurrence in another group (Singer and Willett 2003). In the complementary log-logistic regression, in contrast, the odds ratios express the ratio of the hazard rates, this means the ratio of the conditional probabilities (Lois and Kopp 2009). These models specify the probability that an event occurs at the time t on condition that the event has not yet occurred at the time t−1. The regression equation thus is: clog − log h(t_i) ≡ log[−log(1 − h(t_i))] = α_t + β′x_it + γ′z_i (following Allison 1982, p. 72).

Where β′ represents the regression coefficient of the time-varying independent variable x for individual i at time t. Analogously, γ′ stands for the regression coefficient of the time-constant variable z of individual i. The time dependence is modelled by dichotomous variables α_t, which represent time intervals that cover the time since child birth. The definition of these intervals is based on empirically observed limits and takes the parental leave regulation into account.¹⁶

As there are multiple observations available for each individual due to the person-month structure of the data, robust standard errors are calculated.¹⁷ The results show average marginal effects (AME). They indicate the average influence of the independent variable on the probability of event occurrence (Mood 2010; Wolf and Best 2010).

4.3 Operationalisation

The dependent variable in our study is a dichotomous variable that specifies for each individual and month of observation whether a woman has returned to the labour market (1) or not (0). We restrict our analysis to the birth of the first child, because this event is considered decisive for the further life and career trajectory (and any additional births). Therefore, our observation begins in the month the first child is born or in the first month of parental leave¹⁸. Taking up employment of at least 15 h a week is considered as return to the labour market. If a woman has not returned to the labour market after 72 months or if a second child is born, the episodes are right censored.¹⁹ The decision to right censor after 72 months is based on our research question, as our aim is to analyse the influence of occupational characteristics on the mothers’ individual decision to return.²⁰ The later a mother returns (or if the mother returns after the end of the statutory job-protection period), the more this decision also depends on the employer. Moreover, it can be assumed that the effects of the characteristics of the pre-parental-leave occupation decrease over time or that other factors are decisive for the decision to return after a period of more than six years.

Our central independent variables are the proportion of women, the part-time rate and the occupational wage level of the reference occupation, that is the occupation the woman worked in immediately before interrupting employment.²¹ Based on the SIAB data, the proportion of women states the percentage of women among all employees by occupation and by year. Accordingly, the variable may take on values between 0 and 1, where 1 means that 100% of the incumbents in this occupation are women, whereas a value of 0 means that not a single woman is working in this occupation. For the descriptive observations, we put this variable into three categories based on Trappe (2006): Female-dominated occupations are occupations with a proportion of women of at least 70%. The proportion of women in male-dominated occupations can be 30% at the most, and the percentage of women in integrated occupations can range between 30 and 70%. The proportion of part-time employees was calculated on the basis of the Microcensus and states the percentage of employees (women and men) who work up to 20 h a week²². This variable also ranges from 0 to 1. Based on the SIAB data, moreover the average daily wage ²³ of the full-time employees of an occupation was calculated.²⁴

The selection of our control variables is based on the results of previous research (see Sect. 2). Accordingly, we include marital status (married vs. unmarried) as an indicator of alternative sources of income. In previous studies, the level of education has additionally turned out to be a key influencing factor (Drasch 2013), which is why we take into account the highest educational qualification at the time the child is born. As the data we use here are retrospective, they do not include individual-level information on mothers’ attitudes and personality traits. Therefore, we use the level of educational achievement to capture possible selection effects with regard to the mothers’ career orientation. If specific personality traits actually cause self-selection in female-dominated occupations and longer employment interruptions, we would also have to assume that these women already based their educational decision systematically on their family orientation, which could be controlled for by taking the level of educational qualification into account. We differentiate the educational groups according to the CASMIN classification scheme and create three categories: mothers without vocational qualification, with vocational qualification and university/university of applied sciences degree.²⁵ Additionally, mothers’ previous employment experience is included.²⁶ The level of the educational qualification and mothers’ previous employment experience also serve as an indicator of the individual wage before the employment interruption. To take any differences in socialisation into consideration, we also control for whether the mother has a migration background (at least one parent or the person herself was born abroad). Moreover, a time-varying variable is used to factor in whether mothers become pregnant again during the interruption.²⁷

At the individual level, we also control for characteristics of the pre-leave job. For this, we consider whether the woman held a management position prior to child birth. This also allows us to address women’s career orientations. Moreover, we identify mothers who worked in the civil service before the interruption, as this is indicative of working conditions and should thus influence the speed of return. A final characteristic of the previous job that is included is whether the respondents had been employed for less than 20 h a week before the child was born. This piece of information maps the possibility of returning to the previous job in part-time without reducing the working hours and can additionally serve as an indicator of the woman’s employment orientation.

As described above, the parental leave regulations have a significant impact on the mothers’ speed of returning to the labour market. Although the duration of job protection remained constant across our analysis sample, there was a change in the parental leave regulation during the observation period: from the year 2007, couples who divided childcare duties were rewarded, i. e. couples where both partners take parental leave were given two extra months of parental leave. Furthermore, the period for receiving parental benefit was reduced to 12 or 14 months and it was changed into a wage-dependent benefit (BMFSFJ 2012). To map these changes, we control whether the first child was born before 2007. Finally, the childcare rate for children under the age of three by federal state is also included in the models as a time-varying control variable, since the availability of (public) external childcare is often necessary for the mother to return to the labour market.²⁸ Table 2 in the Appendix gives an overview of the distribution of the variables used.

5.1 Descriptive results on the duration of family-related employment interruptions

In a first step, we conduct a descriptive analysis of whether we can determine any impact of the proportion of women in the occupation held prior to giving birth on the length of the employment interruption. For this purpose, we analyse mothers’ return to the labour market using Kaplan-Meier estimates separately for female-dominated, male-dominated and integrated occupations. Fig. 1 shows the months of employment interruption on the X‑axis, while the Y‑axis depicts the proportion of women who return to the labour market in month X, that is, those who start to work at least 15 h a week. The calculation of the proportion is based on the number of all mothers who are still at risk at this point, meaning that they have neither returned to the labour market nor have they been right censored. The Kaplan-Meier failure function depicts the proportion of women that return to the labour market for every month after the first child was born. The graph thus shows the accumulated proportion of returners.

Fig. 1 shows the importance of statutory parental leave regulations, as most returns take place during the first three years of parental leave. In month 36 of the employment interruption we see a distinct step: Almost 20% of the mothers who have not returned by then re-enter employment in this month. The occupation only becomes relevant after the end of the statutory parental leave period. From this point, we observe significant differences in the return behaviour of women who had previously worked in a male-dominated occupation compared to women who had worked in an occupation dominated by women.²⁹ On the whole, women from male-dominated occupations return to the labour market more quickly and in a greater proportion than women from female-dominated occupations. Women who had worked in integrated occupations before the birth of the first child are in the middle.

This descriptive finding can be interpreted as a first hint regarding hypothesis 3. According to this hypothesis, mothers from female-dominated occupations are expected to interrupt for longer periods, because longer employment interruptions are better accepted due to the working culture in female-dominated occupations compared to male-dominated occupations, where there is a greater pressure to return quickly. However, the Kaplan-Meier failure function cannot answer the question whether the lower speed of return to the labour market of mothers from female-dominated occupations is in fact due to the high percentage of women in these occupations and the associated occupational culture. The observed pattern could also be caused by lower wage levels in female-dominated occupations and hence by lower opportunity costs of interrupting employment, as stated in hypothesis 1. The poorer income prospects in occupations dominated by women could also reduce women’s propensity to return. These descriptive findings do not initially seem to support the suggestion put forward in hypothesis 2, namely that high part-time rates in female-dominated occupations motivate women to return to the labour market more quickly. How the three different mechanisms, namely the proportion of women, the wage level and the part-time rate actually interact will be analysed in the following section using multivariate models that also take the control variables into account.

5.2 Multivariate results on the duration of family-related employment interruptions

Regarding the percentage of women in an occupation, model 1 shows a significant negative effect on the return to the labour market. This means that as the percentage of women in the pre-leave occupation increases, the probability of returning to the labour market decreases or the interruption period after the birth of a child increases. This finding is in line with the result of the Kaplan-Meier estimate and supports hypothesis 3. In the next step, we will investigate whether this connection remains when controlling for the factors associated with a high percentage of women, namely wage level and part-time rate.

Model 2 shows that accounting for the part-time rate in an occupation does not fundamentally affect the relationship observed. Accordingly, the proportion of women still has a negative impact on the mothers’ probability of returning, but the effect’s level of significance decreases. By contrast, the part-time rate alone does not have any significant impact on returning to the labour market. This contradicts our assumption made in hypothesis 2 that increasing part-time rates in occupations reduce the length of an interruption by offering better options for reconciling work and family. Obviously, high part-time rates in occupations are not sufficient to enable women to re-enter employment quickly following the birth of a child.

But when we include the average wage level of all full-time employees in the respective occupation (model 3) instead of the part-time rate, the significant impact of the proportion of women disappears. Thus, we can no longer determine any effect of the occupational gender composition on the individual probability to return. The wage level, however, has a significant positive impact on the probability of re-entering the labour market. This means the duration of family-related employment interruptions decreases as the average wage in the reference occupation increases. Women in better-paid occupations hence re-enter the labour market more quickly.

We see the same picture in our final model that includes all three occupational characteristics.³⁰ Here, too, only the wage level has a significant influence, whereas the percentage of women and the part-time rate make no significant contribution to explaining the probability to return.³¹ This points to our first hypothesis: women from occupations with a lower wage level interrupt their employment for longer periods and have a lower probability of returning after their first child is born than women from occupations with better income opportunities. This occupational effect is found regardless of the individual level of educational achievement, previous employment experience and the mother’s position before interrupting the employment.³² In contrast, the percentage of women as a sole indicator of the prevailing working culture in an occupation and the part-time rate as an indicator of the ability to reconcile work and family do not play any role in mothers’ decision to return.