Nonparametric and graphical results
Figure 2 describes the development of maternal labor force participation after birth, before, and after the reform. It shows smoothed hazards and survivor functions.Footnote 15 Before the reform, exit rates of prior recipients (see gray areas in Panels 1 and 2) peaked after 2, 12, 24, and 36 months. These peaks are likely related to the end of maternity leave (8 weeks), the earliest entry age to formal child care (typically 1 year), the end of child-rearing benefits and eased child care access (2 years), and the end of job protection under the parental leave program plus the guaranteed access to child care (3 years). After the reform, exit rates fall in the first few months after birth and increase significantly around month 12, relative to the pre-reform situation. Subsequent exit rates fall and peak again at month 36.
The survivor functions describe the probability of staying out of the labor force after birth. For prior recipients (see Panel 3), this probability increased in year 1 after childbirth; however, at the end of the new benefit payment period, it falls below prior levels for about 1 year. After the child reaches age 2, the survival probability is similar to the pre-reform level.
Panels 2 and 4 show the behavior of new benefit recipients. The pre-reform peaks in exit rates at months 12 and 24 are much smaller than for prior recipients, most likely because there are no expiring child-rearing benefits for this group. The survivor function in Panel 4 shows that after the reform, the probability of staying out of the labor force increases during year 1, then drops well below the pre-reform level in year 2, and subsequently converges towards the pre-reform level. The overall net-effect of the reform on long-term employment appears to be zero, and the impact of the reform thus appears to be intensive rather than extensive.
Estimation results: before-after comparisons
Next, we apply the semi-parametric before-after model with covariates in order to estimate the effect of the reform. Due to dynamic selection, the non-parametric descriptive hazard rate model cannot be interpreted in a causal way. We use a condensed specification of period-specific hazards. This allows us to estimate the reform effect separately for those who would and would not have been eligible for the pre-reform child-rearing benefits. We allow for different baseline hazards for the two groups. We present our estimation results in terms of hazard ratios and show the hazard ratios for the post-reform effect of exiting non-employment by the age of the child separately for prior and new recipients. The reference group consists of mothers of the given recipient status with a child of the same age in the pre-reform period.
Table 2 presents the estimation results for the three outcomes.Footnote 16 We do not find statistically significant reform effects for the exit rates in the first 11 months for either groupFootnote 17; however, generally exit hazards fall for new benefit recipients after the reform, as expected. The estimations yield mainly significant reform effects around month 12 after birth for both groups.Footnote 18 Mothers who would have been eligible for the pre-reform benefit show an increased exit rate when the new benefit expires. New recipients show mostly significant increases in the exit rates in months 12–14. This increase in exit rates is particularly large for overall labor force participation and substantial employment. For months 15–21, we find increased exit rates to the labor force for both groups after the reform. At later periods, the exit hazards are generally reduced. However, the latter patterns are not precisely estimated.Footnote 19
Table 2 Hazard models—basic specification In order to visualize these reform effects, we simulated the pre- and post-reform survivor functions for prior and new recipients using average characteristics of both groups. Figure 3 describes the predicted survivor functions, separately for prior and new recipients. The reform yields increased exit rates to the labor force starting around month 12 for both groups and for all three outcomes. The survivor rate has dropped by 14 (15) percentage points for prior (new) recipients at month 15 (see Panels 1 and 2). The predicted time for prior recipients to return to the labor force fell at the median by 10 months, from 29 to 19 months after the reform (see Panel 1). This duration fell by 8 months, from 37 to 29 months at the median after the reform for new recipients (see Panel 2). Due to the generally low employment rates of German mothers, we cannot determine the median change for average prior and new benefit recipients: Panels 3–6 show that over the entire period, the survivor curves do not cross the median line. The figures show, however, increased full-time employment probabilities after the reform, particularly for prior benefit recipients starting at month 12.
Based on the predicted survivor function, we can sign the cumulative change in the number of hours worked at months 24 or 36. If we assume a constant employment intensity among mothers before and after the reform and apply a “back-of-the-envelope” calculation, the overall number of hours worked increased both for substantial and full-time employment after the reform. This confirms a strengthened labor market attachment.
We can also calculate in a “back-of-the-envelope” fashion the elasticities of the probability of remaining out of the labor force after 6, (12), [24], and {36} months with respect to income lost if not working during the 24 months after birth.Footnote 20 For prior recipients, these elasticities amount to − 0.008, (− 1.429), [− 1.759], and {− 0.765} and for new recipients to − 0.174, (0.679), [1.389], and {1.604}. Prior recipients react as expected; on average, they permanently reduce the probability of staying out of the labor force after a 1% increase in lost income, i.e., they are more likely to return to work. New recipients react differently. They reduce the probability of staying out of the labor force after a reduction in income lost starting with year 1 after child birth.
Overall, we do not observe the expected significant drops in maternal labor force participation during benefit receipt (see Section 2.2), and we find increased labor force participation for all mothers after month 12. The strong increase in the propensity of newly eligible mothers to return to the labor market after month 12 does not agree with the prediction of no behavioral change or even falling labor supply discussed before. In the next section, we explore alternative explanations of this effect by considering specific mechanisms and subgroups.
Heterogeneity in before-after effects: hypotheses and results
A number of mechanisms may determine the post-reform labor market choices at the point when benefits run out for mothers who newly receive parental leave benefits. In this section, we discuss and evaluate the plausibility of five mechanisms: (i) speed premium, (ii) paternal involvement, (iii) child care availability, (iv) maternal preferences for own income and economic independence, and (v) social norms. We evaluate these mechanisms by comparing the behaviors of those who are and those who are not affected by any given mechanism.Footnote 21
(i) A first rationale for new recipients’ increased labor force attachment after month 12 is that employment after childbirth may now affect future parental leave benefits. This generates a work incentive for mothers who expect to have additional children. To evaluate the plausibility of this explanation, we tested whether mothers of first children respond more strongly to the reform (see Table 3). We do not find significantly higher exit rates after month 12 among first time mothers; thus, there seems to be no support for this mechanism.
(ii) A second mechanism that might explain increased maternal labor force attachment after month 12 may be related to the new regulation for fathers, who can now take two additional months of benefits: as couples often use paternal after maternal leave, the household employment situation changes after month 12. This may facilitate maternal return to work compared to a situation with static household labor supply. To test the plausibility of this mechanism, we evaluated the correlation of maternal exit to the labor force with paternal leave taking by adding interaction terms of indicators of paternal involvement with the reform to the specification (see Table 4). However, we find no evidence to support the hypothesis.
(iii) Next, we investigate whether changes in child care availability over time might be related to maternal labor force attachment.Footnote 22 As a first test, we control for child care coverage for children below age 3 in the maternal county of residence. We can incorporate region-specific and calendar-time varying information for all mothers. The results in Table 5 show small positive effects of child care availability on maternal return to the labor market which is statistically significant only for return to substantial employment. However, our main result, i.e., that new recipients increase their labor supply after 12 months after the reform is even stronger after controlling for child care availability. In additional estimations, we used more flexible specifications and interacted regional child care availability with the age of the child because availability may affect mothers differently depending on the age of her child. The results confirm this expectation (see Bergemann and Riphahn 2021) and show significantly positive effects of child care availability on labor force return. However, we continue to find strong and significant reform-induced increases in labor force return after year 1. We also allowed the child age-specific child care availability effects to change after the reform and to differ in urban (high demand) and rural (lower demand) areas. This did not affect our main estimates of the reform effects.Footnote 23
(iv) Another potential mechanism relates to mothers’ preferences with respect to economic independence and an own income (i.e., reference-dependent preferences, see DellaVigna et al. (2017)): before the reform, mothers without child-rearing benefits who left the labor force and cared for a child lost their benefit income at the end of maternity leave, i.e., 8 weeks after birth. After the reform, the loss of an own income typically occurs only after month 12. At that time, mothers may judge the option of returning to work and seeking external care for their child differently than after week 8. Particularly, for mothers who were used to relatively high own earnings prior to birth (see bottom panels of Table 1), the loss of an own income after month 12 can provide an impetus to return to work. This might increase labor force participation rates beyond pre-reform levels. A similar response can result from a consumption habit where behavior responds to a taste for certain consumption levels. Alternatively, it may be influenced by the mothers’ interest in maintaining her economic independence and bargaining position in the partnership.
Table 3 Hazard models—test whether first-time mothers respond more strongly to the reform Table 4 Hazard models—test for a response to paternal leave taking Table 5 Hazard models—test by controlling for local child care supply To test whether the high rate of return to the labor force at month 12 is associated with mothers’ preferences for an own income and economic independence, we apply two measures. First, we test whether women who strongly value being able “to afford something” react stronger to the reform.Footnote 24 These women might be particularly attracted by the new option of maintaining their financial independence. Indeed, we find an (weakly significant) increase in exits to the labor force around month 12 for this particular group (Table 6); a limitation of this result is that due to data restrictions, we have to use information that was gathered after birth and may be endogenous. In addition, we consider information on how couples handle their finances. We assume that women who manage their accounts separately or partly separately value their financial independence either because of a preference for independence or because they do not have access to their spouses’ account (see Bergemann and Riphahn 2021).Footnote 25 We find that those mothers who handled their finances independently before the birth generally have a higher hazard of returning to the labor force. Also, they respond stronger to the reform: they are significantly less likely to return to the labor force in months 1–11, and they are substantially (yet mostly insignificantly) more likely to return after the benefit runs out.
Table 6 Hazard models—differential effects by “Valuing to be able to afford something” Finally, we evaluate mothers’ labor market response by maternal share in household income and by level of education. Both measures also may not only be indicative of preferences regarding economic independence and an own income but also address potential pressure to earn household income. The results (see Bergemann and Riphahn 2021) yield that the propensity to return to the labor force is significantly higher for mothers who contribute a large share to household income. Also, these mothers—similar to those with high education—respond to the reform (insignificantly) stronger than others. Overall, the evidence appears to agree with our expectations.
(v) Alternatively, one might argue that the new benefit expiration after month 12 generates a social norm and a signal for young mothers: now it is socially acceptable (or even expected) to return to work and to use child care once the child has reached the age of 1 year (see Olivetti and Petrongolo 2017). Following the model of Akerlof and Kranton (2000), such social norms can influence economic outcomes as they affect a person’s identity that in turn influences the utility function. Similarly, young mothers might respond to (perceived) expectations of their employers (e.g., Bernheim 1994).Footnote 26 Such social norm effects are a common explanation of observed retirement behavior (e.g., Hanel and Riphahn 2012a).Footnote 27 If prior to the reform, the focal, expected, or normal point for young mothers to return to work was after 36 months at the end of employment protection (see Fig. 2); this may have shifted after the reform to month 12, the end of transfer receipt. Thus, increased maternal labor force participation after month 12 could result from a change in social norms.Footnote 28
We use various approaches to test the plausibility of this hypothesis. (a) As a change in social norms takes time, we expect a potential reform effect to increase over time. Thus, we consider an interaction term of the reform effect which indicates whether a child was born in 2008 rather than in 2007. The estimation results in Table 7 show that the increase in exit rates in months 12–14 was significantly higher for births that occurred in 2008 rather than in 2007. In addition, the decline in months 1–11 is (insignificantly) stronger for later births.Footnote 29 While they cannot offer final proof, these results support the social norm hypothesis. (b) Next, we test whether women who value success at work react stronger to the new policy.Footnote 30 Because the traditional social norm of staying at home after childbirth was particularly binding for this group, they might adjust stronger to the change in circumstances than others; following the model of Akerlof and Kranton (2000), for these women, gender identity was particularly binding due to the old social norm. While they do not offer formal proof, the results support this reasoning (see Bergemann and Riphahn 2021). (c) Third, personalities respond differently to changes in social norms. One might expect that women with a more external locus of control respond stronger to changes in social norms. We test whether mothers who agree with the statement that “others make the crucial decisions in my life” respond stronger to the reform; we add an interaction term of this characteristic with the reform effect to the empirical specification. The insignificant results agree with this presumption (see Bergemann and Riphahn 2021). (d) Finally, we compare the reform response between East- and West-German mothers. Given the socialist heritage of East Germany social norms, there are more in favor of maternal employment and early return to work (see, e.g., Campa and Serafinelli 2019 or Hanel and Riphahn 2012b). If a shift in social norms occurs after the reform, it should be visible particularly in West Germany. The estimation results show that the reform effects around month 12 are economically but not statistically significantly larger in the West (see Bergemann and Riphahn 2021). This confirms the plausibility of a shift in social norms after the reform which may drive increased labor force return in months 12–14.Footnote 31
Table 7 Hazard models—differential effects by time since reform Finally, as we consider a large number of heterogeneity tests to evaluate the plausibility of five separate mechanisms, our results may be subject to the effects of multiple hypotheses testing. In order to test the robustness of our findings, we estimated a model which considers all hypotheses simultaneously, i.e., interactions for a first birth, paternal involvement, year of birth, and valuing economic independence. In addition, the model accounts for child care availability and the relevant main effects. This joint testing reduces the problem of multiple hypotheses testing and estimates partial effects of the different hypotheses. We present the results Appendix Table 11. They confirm that women who value “to be able to afford something” and with a later born child return to the labor market faster around month 12. Overall, we interpret this as suggestive evidence, in support of the hypothesis that the increased labor force participation after month 12 relates to changes in social norms and to a preference for financial independence.
Robustness tests
Difference-in-differences (DID)
We apply a DID estimation approach to account for potential effects of the business cycle and secular shifts. We reestimated our model using mothers of 3-year-olds as a control group. We allow for different baseline hazards for the treatment and control groups because the form of their exit hazards may differ. Table 8 shows the estimation results when the period effect (αP) is constant across child age groups. In other specifications, we considered time trend controls, used duration-varying effects, and controlled for quarterly calendar effects (see Bergemann and Riphahn 2021). Our key results are robust: we find an intensified return to the labor force after year 1 in the post-reform regime for prior and new recipients. Our DID estimates generate a lower bound of the causal effect if the control group similarly responds to an overall shift in social norms. Given that we consider binary measures of labor force participation, potentially heterogeneous business cycle effects on, e.g., the number of hours worked in the treatment and control groups, do not affect our results.
Table 8 Hazard models—DiD specification without time trend Before-after observation window
So far, we considered maternal employment outcomes for births that occurred 2 years before and after the reform. We also set the time horizon to 6 months before. With this sample, it appears that after the reform prior, benefit recipients returned to employment faster already in months 1–11 rather than around month 12. However, the estimates confirm the large post-reform increase of exit rates into the labor force and substantial employment around month 12 for new recipients (see Table 9).Footnote 32 When setting the observation period to 1 year before and after the reform (see the Bergemann and Riphahn 2021), the reform effect for the new recipients around month 12 is significant for two of the three exit states and even larger than in Table 2. Again, we do not find an increase in the exit rate to substantial employment for prior recipients around month 12.
Table 9 Basic Specification with 6 Months Window without Time Trend Omitting December 2006 and January 2007 births
Tamm (2013) showed manipulations of the timing of births around the reform date. In response to this, we reestimated our model in Table 2 after dropping the births of December 2006 and January 2007 (N = 24). This does not affect the results (see Bergemann and Riphahn 2021).
Employment before birth
We do not control for the employment status before birth due to its potential endogeneity in our main specification. When controlling for pre-birth employment status in sensitivity analyses, the results remain very stable (see Bergemann and Riphahn 2021).Footnote 33