Effects on preschool attendance
The increased availability of preschool seats may increase maternal employment if the reform generated an increase in preschool attendance. Following a strategy similar to Berlinski et al. (2009), we estimate the effects of preschool supply on attendance at the region level (see Table 2, Panel A).Footnote 15 We regress the change in the share of 3- to 5-year-olds in preschools on the change in the ratio of the preschool places available to 3- to 5-year-olds while including region and year fixed-effects (model 1).
We further use the DD specification presented by equation (1) to estimate the effect of preschool availability on attendance rates in models 2 and 3 (the DD model 3 includes region-specific time trends). The estimates show that for every 10 additional available places, about 5 additional children attend preschool. The effects are largest for 5-year-olds, which is in line with the statutory right given to 5-year-olds to attend preschool. However, the effects are also significant and sizable for mothers of 3- to 4-year-olds.
In Panel B of Table 2, we test the link between the number of 6-year-olds leaving pre-school and the number 3- to 5-year-olds entering preschool. On the left-hand-side, we have the change in the number of children aged 3–5 in pre-school while the treatment variable is defined as the change in the number of 6-year-olds in preschool. The final column with the DD specification that includes region-specific time trends shows that for each 6-year-old less in preschool (but in primary school instead), there is one additional 3- to 5-year-old in preschool. This suggests that the switch from preschool to school among 6-year-olds led to more spaces being available for children aged 3 to 5. The effect is sizable, suggesting that rationing of preschool places plays an important role in Poland (consistent with the discussion in Sect. 2.1). Given that the reform led to a significant increase in the use of preschool services, the reform may also have substantially facilitated mothers of young children to return to work.
Effects on maternal employment
The main estimation results of the DD and DDD models are presented in Table 3. The upper panel shows the effects on employment and the lower on working hours. We define several alternative model specifications (as in e.g. Nollenberger and Rodríguez-Planas (2015). The DD effect (model 1) is estimated by the interaction between post-reform dummy and the ratio of preschool places to 3- to 5-year-olds [Eq. (1)]. The DDD model estimates the effect through the triple interaction between the post-reform dummy, the ratio of preschool places to 3- to 5-year-olds and treatment group status (Eq. (2)). The DDD model 2 includes region-specific time trends. DDD model 3 is our main DDD specification; this model does not include region-specific time trends but does introduce interactions between treatment group status and region, year fixed effects and personal characteristics. In model 4, region-specific time-trends are re-introduced along with a linear trend for the treatment group mothers (models 5–7 are based on a different comparison group, see below).
The results indicate that the effects of increasing preschool availability are statistically insignificant and negative in the DD model, but positive and statistically significant in the DDD models. This is similar to previous studies such Nollenberger and Rodríguez-Planas (2015) for Spain and Cascio and Schanzenbach (2013) for US, reporting statistically significant positive effects in the DDD model but not in the DD model; below we discuss more extensively the reasons for the difference between the DD and DDD results. The effect size becomes larger once we control for interactions between region- and year-specific fixed effects, personal characteristics and treatment group status in model 3.Footnote 16 Similar to the estimates of Nollenberger and Rodríguez-Planas (2015), including region-specific time trends (model 4) does not seem to change the coefficient estimates. The coefficient estimate suggests that a 10% point increase in the ratio increases employment probability by 4.2% points.Footnote 17 The effect on working hours reflects the substantial effect on employment. Average working hours for working women prior to the reform was 38.5 and the average for all women was 24. A 10% increase in the ratio appears to increase working hours for all women by 1.2 h, which corresponds to a 5% increase in female labour supply. The effect found on working hours appears to be driven by the decision to enter the labour market rather than an increase in working hours of employed mothers. Table 7 in Appendix estimates the working hours effects for only working hours and the reported results show no statistically significant effects for mothers or fathers. This result is probably due to the low incidence of part-time work which is reflected in the average working hours of employed women being close to the full-time equivalent of 38.5 h.
While we use mothers with the youngest child aged 7 to 8 as the control group in models 2 to 4, we examined whether the results are sensitive to the choice of control group. We test whether the results change based on the control group by using mothers with a youngest child aged 7 to 9 in model 5, 9 to 10 in model 6 and 10 to 12 in model 7. The results show that the coefficient estimates change only marginally when mothers with slightly older children are used as the control group.Footnote 18
We also tested whether the effects differ depending on our choice of treatment and control years. Throughout our main specifications, we used the years 2005 to 2008 as our control years and 2009 to 2011 as our treatment years. 2009 is a special case since treatment begins in September. In Appendix Table 9, we extend the sample to all years between 2003 and 2013 in column 1. In column 2, we limit the control years to 2007 and 2008. In column 3, we use all available control years starting in 2003. In the final column, we allow treatment years to continue up to 2013. The results remain relatively stable throughout. The main difference is seen when including years 2012 and 2013, where the coefficient becomes smaller. This seems in line with the roll-back of the reform that began in 2012.
Finally, we also examined whether the results are sensitive to a change in our treatment variable by constructing the treatment variable as the ratio of all preschool places (i.e., without subtracting the seats occupied by 6-year-olds) to the number of 3- to 5-year-olds in the region. The model using this alternative treatment variable leads to qualitatively similar employment effect estimates (see Appendix Table 10). However, as expected, the effect sizes become smaller since we do not exploit all relevant variations in changes in preschool availability.
Our DDD results point out that the reform significantly increased maternal employment. One may expect that the effect of increased employment among mothers may lead to a negative effect on the employment of fathers through an income effect. For example, Bettendorf et al. (2015) found that a childcare subsidy reform in the Netherlands increased female employment but decreased male working hours. We test whether fathers are affected in the Polish setting by estimating the same DD and DDD models for fathers: the results are presented in Table 11. None of the models report significant effects for male employment. Since working hours flexibility is low in Poland (Plomien 2009), it seems unlikely that they would be able to adjust working hours in response to increased female employment even when faced with increased household consumption.
While the DDD results thus far suggest that maternal employment increased due to the increased availability of preschool services, the results hinge on the common trend assumption. If employment rates of mothers with older children who comprise our control group underwent a shock at the same time, our results will be biased. This potential threat to validity is especially relevant in our study since the years of the reform coincide with the financial and later sovereign debt crises in Europe which are likely to have had negative effects on employment. In Table 4 (panel A), we present the results of several placebo tests of our main DDD model. In the first four columns, we present results of models using mothers of children aged 6, 7, 8 as well as 7–8 combined as the treatment group. We adjust the control group according to mothers with youngest children up to 2 years older. In addition, on the rightmost column of Table 4, we show the results of a placebo reform that is assumed to have gone into effect in September 2007 (actual treatment years 2009–2011 are excluded). None of the placebo estimates appear to be statistically significant, confirming that our main DDD estimates capture an increase in employment that is unique to the mothers in the actual treatment group.
There is a large difference between the DD and the DDD results, indicating that there are significant cross-regional variations in the development of female employment during the relevant years. This is in line with Fig. 2 which suggests that the strong upward trend in female employment in Poland ended with the onset of the financial crisis in 2008–2009. To test whether this explains the difference between the DD and DDD results, we performed a series of placebo tests, similar to the ones discussed above for the DDD models (Table 4, Panel B). The results of these placebo tests point out the limitations of the DD models in our context: our treatment measure is significantly related to the employment of mothers with a youngest child aged 7–8. This indicates that regions that experienced a larger increase in preschool availability were hit harder by the economic crisis. We further investigated whether the crisis is causing the difference between DD and DDD results by estimating the DD and DDD models using a single year before and after treatment (i.e., excluding the crisis years). More specifically, we use 2007 or 2008 as the control year and 2011 as the treatment year. The results are presented in Appendix Table 12. The DDD estimates are still larger but the effects on employment are significantly positive in the DD estimates as well, which suggests that the regional variation in the impact of the crisis explains the differences in our baseline estimates between DD and DDD models.
While our results indicate that the reform had a positive effect on mothers’ employment, we can look further into sub-samples to test whether the effect is driven by a particular group. Table 5 shows the effects on the employment probability and working hours of various sub-samples of mothers. We use the specification in model 3 of Table 3 in all estimations.
According to the results in Table 5, the treatment effect is stronger for higher educated women. This result seems to be inconsistent with most other studies. A potential explanation is that the availability of other forms of care is low in Poland even for highly educated mothers. Since they are likely to earn the highest wages in the labor market, the impact of the reform is larger for this group. Simultaneously, cultural barriers for lower educated women might explain their lack of response to preschool availability. We also find stronger effects for married women, but the number of single women in the sample is relatively low.
When we estimate the effect for mothers with children at different ages in the last panel of the table, we find a stronger effect for mothers of 3-year-olds. A potential reason is that the expansion of preschool places benefited 3-year-olds’ mothers most. Preschool attendance was already relatively high for 5-year-olds in the pre-treatment period and there may be more mothers willing to work but unable to find preschool services among mothers of 3-year-olds. It should be noted that children can enroll in preschool from age 3 onward (conditional on sufficient availability) and that parental leave during the relevant period was around 3.5 years: age 3 therefore represent a critical transition phase, both for children and mothers. Hence, a larger share of mothers of 3-year-olds may be able to return to the labour market when more preschool seats become available.
To further examine age-specific effects of the reform we replaced the ratio of preschool places to the number of children with the actual attendance rate as the treatment variable. We consider the ratio of preschool places as a better treatment indicator since the policy reform had a direct effect on available places and not coverage. Moreover, the attendance rates may be considered endogenous as attendance rates could be demand driven. However, the results of the DD and DDD regressions based on the region-specific attendance rates show little qualitative differences from the main results: see Appendix Table 13. Nevertheless, one advantage of using the attendance rate is that we can calculate age-specific attendance rates instead of relying on seats available for the total group of 3- to 5-year-old children; these age-specific attendance rates can be used to estimate age-specific effects. Consistent with the results discussed above, Appendix Table 14 shows that the treatment effects are larger for mothers of 3-year-olds and that mothers of 4-year-olds are not affected by the reform. In contrast to the results based on the number of preschool places, these results show that the employment of mothers of 5-year-olds significantly increased. This may be expected given that a statutory right to preschool was provided to children aged 5.
Lastly, using attendance rather than coverage as the treatment variable allows us to test whether the switch from preschool to school affected the employment of the mothers of 6-year-olds. As 6-year-olds are either in preschool or school, we do not expect a substantial effect of the reform on the employment rate of mothers of this age group. The results show that, while there is a significantly positive effect in the DD specification, the preferred DDD models largely indicate that leaving preschool for school had no statistically significant effects on the employment of mothers of 6-year olds (Appendix Table 14).
Effects of ECEC reforms in Europe
Table 6 summarizes our results alongside the results from the previous quasi-experimental studies on the maternal employment effects of ECEC. The largest effects appear to be found in Poland and Germany, where per percentage point increase in ECEC availability or use (in case of Germany), mothers’ employment rises by around 0.4% points. The dead-weight loss was greatest in Norway, where a 57% point increase in childcare use led to a 4% point increase in employment (Havnes and Mogstad 2011). There was also no employment effect in Sweden from a decrease in childcare prices according to Lundin et al. (2008), but that seems hardly surprising given the limited increase in childcare use. The results from the Netherlands and Spain are in between the results from Poland and Germany on the one hand and the US and the Nordic countries on the other hand. For example, in the Netherlands a 10% increase in childcare use corresponded to an employment increase of around 1.5% points.
There may be a combination of factors explaining the heterogeneity in empirical findings. First, some studies suggest that effect sizes decline with the pre-reform maternal employment rate (Lundin et al. 2008; Bauernschuster and Schlotter 2015). However, other studies do not confirm this pattern. Second, the responsiveness of maternal employment to ECEC reforms may hinge critically on the availability of alternative forms of childcare services: several studies (Blanden et al. 2016; Cascio and Schanzenbach 2013; Havnes and Mogstad 2011) attribute the lack of strong effects on maternal employment from public childcare and preschool expansions to the high availability of other forms of childcare prior to the expansion. Basically, the expansion of public ECEC services may substantially crowd out existing (formal or informal) childcare arrangements. For instance, studies from the US where private preschools are already available generally find weak employment effects from preschool expansions targeted at older children (Cascio 2009; Fitzpatrick 2012). Similarly, effects may be limited due to crowding out of informal childcare arrangements (Havnes and Mogstad 2011). Although the use of informal care is relatively common in Poland, a very high share of Polish children below the primary school age are cared for only by their parents (as discussed in the last paragraph of Sect. 2.2): for these families, there will be no crowding out of alternative care arrangements. This may (partially) explain the rather large effect sizes reported in our study.
It is striking that the estimated effects of both the German and Polish ECEC reform are considerably larger than other estimates. The two cases share important similarities. First, both reforms concern an expansion of services for children from age three until primary school entry.Footnote 19 Second, during the relevant evaluation periods, the childcare infrastructure for children below the age of three was highly underdeveloped in Poland and West Germany (where the employment effects are found). Third, the total duration of job-protected leave is lengthy in both cases (three and three and a half years in Germany and Poland, respectively). We conjecture that women in Poland and West Germany preferring to return to their job (near) the end of parental leave were more likely to do so due to the increased availability of preschool places for their 3-year-olds. Interestingly, the results from Norway (Havnes and Mogstad 2011) pointing out that the expansion of preschool services for children aged 3 to 6 hardly affected maternal employment and mostly crowded out informal care arrangements are consistent with the childcare-parental leave interaction explanation. As the period of leave was rather short (12–18 weeks) in Norway, mothers who wished to return to the labor market at the end of leave had to rely on informal care arrangements. It is then also plausible that an expansion of preschool for 3- to 6-year-olds resulted in a crowding out of these informal care arrangements. Recent evidence indicates that increasing the duration of parental leave has generally limited effects on maternal employment in the medium and long run, unless mothers do not return to work at the end of the job protection period (Lalive et al. 2013; Schönberg and Ludsteck 2014; Mullerova 2017). The latter scenario is of course more likely when limited childcare services are available. This once more suggests that interactions between childcare policies and parental leave schemes are important. Overall, the evidence indicates that the maternal employment effect of a childcare expansion depends crucially on whether this allows mothers to use parental leave more effectively as a bridge to return to work.