Abstract
We provide comprehensive evidence on Germany’s home care subsidy for one‐ and two‐year‐old children. In West Germany, take‐up was 60 percent, and the subsidy reduced mothers’ probability to work within three years after childbirth by 1.4 percentage points and increased exclusive parental care by 6.5 percentage points. The subsidy improved children’s development at age six, with the exception of children who do not speak German at home. In East Germany, 30 percent of families used the subsidy, neither affecting maternal employment nor exclusive parental care. As an income transfer, the subsidy did not benefit families with the least economic resources most.
I. Introduction
Over the last four decades, maternal employment rates have increased substantially across all OECD countries. Mirroring this trend, access to universal childcare programs for children below age three has also expanded, and current initiatives push to increase access further (for example, OECD 2017a). In most countries with universal programs, families using these programs pay only a fraction of the costs, and subsidies cover the remainder (OECD 2016). These substantial subsidies gave rise to discussions whether families not using subsidized childcare should also receive public support in the form of home care subsidies, also known as cash‐for‐care programs. Such home care subsidies were implemented, for instance, in Finland, Norway, Sweden, and in Germany at the national level and—in different forms—in the states of Thuringia and Bavaria. In each case, the home care subsidies caused heated debates. Apart from fairness concerns, proponents argued that home care subsidies give families the freedom to choose the optimal childcare arrangement for them, whereas opponents raised concerns that such subsidies reduce maternal employment and hinder children’s development by reducing enrollment in subsidized childcare, with particularly strong negative effects for the most disadvantaged children (Duvander and Ellingsæter 2016).
We examine how the introduction of a home care subsidy affected maternal employment, childcare choices, and children’s development, and which groups benefited most from the subsidy. We examine the case of Germany, where a conservative‐liberal government introduced a national home care subsidy in 2013. Parents with children under age three became eligible for the subsidy once both parents could no longer claim paid parental leave benefits and if they did not use any form of subsidized childcare. Parents could receive the subsidy for up to 22 months, typically from month 15 to month 36, which coincides with the last month of unpaid parental leave. The subsidy initially amounted to 100 euros per month, which corresponds to roughly 50 percent of average parental fees for subsidized care, and it was increased to 150 euros in August 2014. Our difference‐in‐differences (DiD) identification strategy exploits that only parents of children born from 1 August 2012 onwards were potentially eligible for the subsidy; to cancel out seasonal effects, we compare children born either before or after August in 2012 with children born in the same months but other years, who do not differ in their potential eligibility for the subsidy. Throughout the paper, we refer to families as potentially eligible if the child was born after 1 August 2012 and as eligible if they additionally did not use subsidized childcare.
As home care subsidies increase the costs of using subsidized childcare, we expect them to decrease the use of public childcare and furthermore to reduce maternal employment.1 Studies for Norway (Naz 2004; Schøne 2004; Drange 2015; Thoresen and Vattø 2019), Sweden (Giuliani and Duvander 2017), Finland (Kosonen 2014), and Thuringia (Gathmann and Sass 2018) document negative effects on maternal employment and on the use of formal childcare (for example, see Gathmann and Sass 2018; Thoresen and Vattø 2019), although both effects are typically small. These studies from Nordic countries, however, do not examine children’s outcomes and do not analyze who benefits financially from the subsidies.
The theoretical predictions on the effect of home care subsidies on children’s skill development are ambiguous because home care subsidies affect children’s development through three different mechanisms. First, the subsidy changes maternal employment—and parents’ time use more broadly—which influences the quality of parent–child interactions (for example, Hsin and Felfe 2014). Second, the subsidy affects the form of care that children receive. This change in childcare modes may either improve or impair children’s development, depending on the difference in the quality of childcare provided in the different care environments (see, for example, Cascio 2015; Cornelissen et al. 2018; Fort, Ichino, and Zanella 2020). Third, the subsidy increases family income, also for inframarginal recipients who neither adjust their labor market behavior nor their childcare choices. Previous studies suggest that changes in parental resources early in a child’s life may affect children’s development in the short and long run (for example, Currie and Almond 2011; Løken, Mogstad, and Wiswall 2012). Empirical evidence on the effects of home care subsidies on children’s development is limited to survey data from Thuringia reported by Gathmann and Sass (2018), who find positive effects on children’s development on average, though their evidence is restricted to the short run. We thus lack evidence on the effects of home care subsidies on children’s development, in particular beyond the short run.
Against this backdrop, we make three contributions to the literature. First, using high‐quality administrative social security records on employment alongside detailed household survey data on take‐up of the subsidy and on childcare choices, we examine how the introduction of a home care subsidy in Germany affected maternal employment and childcare arrangements. The results differ markedly between West and East Germany. In West Germany, around 60 percent of potentially eligible families received the subsidy, and the average receipt duration amounts to 20 months (conditional on using the subsidy). The subsidy still reduced maternal employment only minimally within three years after childbirth at the extensive (−1.4 percentage points, baseline 82.8 percent) and intensive margin (−13.5 days in employment, baseline 452 days). The subsidy also increased the use of exclusive parental care (+6.5 percentage points, baseline 30 percent) when children are one and two years old. Our estimates furthermore indicate reductions in the use of subsidized care and informal care, though these estimates are statistically insignificant. In East Germany, less than 30 percent of families received the subsidy, and it neither reduced maternal employment nor the use of subsidized childcare. We also find no evidence that the subsidy increased exclusive parental care. The different effects in East and West Germany point to the importance of social norms as a moderator for the effects of family policies.
Second, we examine how the home care subsidy affected children’s skill development in the medium run using administrative data. Our data stem from compulsory school entrance examinations at age six and cover the full population of children from one federal state in West Germany. The data include an overall assessment by a pediatrician on whether children need additional support to attend school. Similar measures of school readiness correlate strongly with later educational achievement (for example, Duncan et al. 2007; Grissmer et al. 2010), health outcomes, and labor market performance (for example, Cunha et al. 2006). Given that critics worried about the development of disadvantaged children, examining additional support needs seems particularly relevant. In addition, we use an overall index of children’s development, which we construct from several cognitive and noncognitive skill measures that are collected in a standardized way during the screening for a subsample of children. Our results show that the home care subsidy improved children’s development on average (−3.6 percentage points in support needs, baseline 22.5 percent), though children from nonnative families did not benefit from the subsidy, most likely due to a reduced exposure to the local language (that is, German).
Third, we document in detail which families received the home care subsidy. The vast majority of recipients of the subsidy did not change their behavior regarding employment and childcare choices—that is, they were inframarginal. Even in West Germany, 98 percent of the recipients were inframarginal regarding employment outcomes and around 90 percent regarding childcare choices. The subsidy hence accrued almost completely to inframarginal recipients, and describing the recipients is important for at least two reasons. First, the change in household income potentially affects children’s development, and analyzing inframarginal recipients thus contributes to our understanding of its effects on children’s development. Second, paying the home care subsidy to inframarginal recipients transfers income to specific families. This transfer is arguably more desirable if families with the least economic resources benefit most. The subsidy, in contrast, benefited single mothers less than two‐parent families. Further, the subsidy transferred income particularly towards families with medium earnings potential and towards couples in which the husband earned the major share of household income already before the child was born.
Our work extends the literature on the effects of home care subsidies that focuses on the effects on maternal employment and childcare use and is dominated by studies from the Nordic countries (Naz 2004; Schøne 2004; Kosonen 2014; Drange 2015; Giuliani and Duvander 2017; Thoresen and Vattø 2019). Overall, these quasi‐experimental evaluations show that home care subsidies have limited effects on maternal employment and on childcare choices, despite substantial take‐up. For further details on these studies, see our summary in Online Appendix Table A.1.2
Gathmann and Sass (2018) provide the most comprehensive evaluation of a home care subsidy. They study the effects of the home care subsidy in Thuringia, a federal state in East Germany that introduced a home care subsidy prior to the national subsidy. The Thuringian subsidy was offered to parents of two‐year‐old children, irrespective of parents’ employment status. Unlike other home care subsidies, the subsidy increases with children’s birth order from 150 euros for first‐born children to 300 euros for fourth‐ or higher‐order children. Using survey data from the German Microcensus and the German Socio‐Economic Panel (GSOEP) within a DiD design, they find that the subsidy decreased the use of formal childcare (−8 percentage points) and informal childcare (−18 percentage points), while at the same time increasing exclusive home care (+18 percentage points). Maternal employment rates decrease in the first two years after eligibility (−4 percentage points), whereas their husband’s employment probability (+7.9 percentage points) and weekly hours worked (+1.8) increase. To assess children’s development, the study uses the GSOEP, which contains parent‐reported information on children’s development for children aged two to three. They find that the reform improved children’s development on average, but was detrimental for children of low‐skilled parents.
Our study differs in five important aspects from Gathmann and Sass (2018) that may also explain differences in effect sizes. First, the Thuringian subsidy was much more generous than the national home care subsidy. Second, the home care subsidy we study is paid only to families not using public childcare at all and thus operates at the extensive margin, whereas Thuringia’s childcare subsidy was gradually withdrawn when families used some hours of subsidized childcare (that is, intensive margin). Third, we use large administrative data sets on employment outcomes and on children’s development, whereas Gathmann and Sass (2018) base their analysis on survey data. Their analysis of children’s outcomes is based on a sample size of 434 observations, which are spread out over several pre‐ and post‐treatment birth cohorts (2003–2010). In addition to providing much larger samples, administrative data have the additional advantage of providing outcome measures that are not self‐reported. This may be particularly important with respect to child development. Fourth, our social security data on employment excludes civil servants and self‐employed persons, who may have a more flexible labor supply. Fifth, the setting in Thuringia created differences in eligibility to the home care subsidy for future births. Our identifying variation, in contrast, creates differences in the eligibility for the subsidy for one specific child, but does does not affect eligibility for future births.
Our paper also adds to two related strands of literature. First, we connect to a large literature on the relationship between childcare costs and female labor supply (for example, Blau and Currie 2006; Black et al. 2014). A large quasi‐experimental literature examines how universal childcare program affect maternal employment (for example, Baker, Gruber, and Milligan 2019; Carta and Rizzica 2018). Overall, this literature arrives at no conclusive generalizations. Second, we also add to the literature on determinants of children’s skill development, in particular with respect to maternal employment (for example, Del Bono et al. 2016; Hsin and Felfe 2014), different types of childcare arrangements (for example, Cornelissen et al. 2018; Drange and Havnes 2018; Fort, Ichino, and Zanella 2020), and family income (for example, Løken, Mogstad, and Wiswall 2012; Dahl and Lochner 2012; Lundstrom 2017).
The paper proceeds as follows. Section II gives the institutional details, and Section III describes the three different data sets. Section IV discusses the empirical strategy and is followed by Section V, which presents the results for West Germany. Section VI provides several robustness checks, and Section VII presents the results for East Germany and compares them to West Germany. Section VIII concludes the paper, discusses the policy implications, and reflects on some empirical challenges in disentangling the different mechanisms on child development.
II. Institutional Background
This section describes the general institutional setting that families face in Germany, the home care subsidy introduced in 2013, and how the institutional setting allows us to identify the subsidy’s causal effects.
German family policy consists of three main pillars to support families in the first years of parenthood: maternity leave, job protection and parental leave, and paid parental leave benefits. Starting six weeks prior to the expected date of delivery, pregnant women are entitled to 14 weeks of paid maternity leave with full wage replacement. Parents can take up to three years of job‐protected parental leave from their employers and have the right to return to their pre‐birth employer afterwards. In particular, they are entitled to return to their previous (or an equivalent) job, that is, with the same hourly wage and working hours as before giving birth. Parents are also jointly entitled to 14 months of paid parental leave benefits (Elterngeld), a tax‐financed payment that typically replaces 66 percent of the parents’ pre‐birth net wages, where one parent (typically the mother) can take at most 12 months.
From the mid‐2000s onwards, Germany increased the supply of subsidized childcare for children below age three, and the enrollment rate of children below age three rose from 13.6 percent in 2006 to 32.7 percent in 2016 (BMFSFJ 2016). In August 2013, Germany introduced a legal entitlement to a slot in subsidized childcare for all children from age one onwards.3 The federal states are responsible for planning and funding formal childcare centers. Despite large regional differences in childcare coverage rates, childcare quality is highly regulated and thus fairly homogeneous (Cornelissen et al. 2018). A marked difference from the United States and Anglo‐Saxon countries is that no noteworthy private childcare market ever emerged in Germany, mainly because of strict regulations, high market entry, barriers, and dominance by publicly funded providers (Kreyenfeld and Hank 2000).
Germany introduced its national home care subsidy in August 2013. Families with children born from August 2012 onwards were eligible if they did not use subsidized childcare (including both childcare centers and child‐minders, Tagesmuetter). In West Germany, around 60 percent of potentially eligible families received the home care subsidy at some point, corresponding to around 90 percent of eligible families (see Section V and Section VII for East Germany). Parents could receive the subsidy for up to 22 months, starting after the end of paid parental leave benefits (when children are typically 12–14 months old) and ending at the latest at a child’s third birthday. According to administrative data, the average granted duration of receipt is 19.8 months conditional on receiving the subsidy (Alt et al. 2015, p. 59).
The subsidy initially amounted to 100 euros per month and was raised to 150 euros per month in August 2014. For comparison, the median net income of working mothers not eligible for the home care subsidy (that is, with children born between January and July 2012) amounted to 1,200 euros when the children were aged one and two (based on the KiBS data, see Section III). Thus, the home care subsidy replaced roughly 10 percent of what mothers would have typically earned in the labor market, which is comparable to similar subsidies in Nordic countries (see Table 4 in Eydal and Rostgaard 2011). Furthermore, parents using subsidized childcare for children under age three paid on average around 200 euros per slot per month (own calculation based on the GSOEP 2011 and 2013, for West Germany); the monthly fees for a full‐day slot in major cities ranged from 200 euros (in Frankfurt) to 540 euros (in Gelsenkirchen, FAZ 2012). The subsidy hence imposed substantial fixed costs on using subsidized childcare. The home care subsidy was fully accounted as income for means‐tested benefits (ALG II) and thus unavailable for families receiving other social benefits, but it was neither subject to income taxation, nor did it affect the marginal tax rate.
Although the home care subsidy and the entitlement to a childcare slot were both introduced in August 2013, we can isolate the effect of the home care subsidy because the eligibility criteria differ. Whereas the slot entitlement did not depend on children’s birth date, only families with children born from August 2012 onwards could receive the home care subsidy. These different entitlement rules give rise to a piece‐wise linear relationship between children’s date of birth and the age at which children become entitled to a slot—the age at slot entitlement decreases continuously from 36 months for children born before August 2010 to 12 months for children born from August 2012 onwards. In contrast, the eligibility criteria for the home care subsidy create a sharp discontinuity in the potential eligibility for the subsidy for children born before or after August 2012. This difference allows us to isolate cleanly the effect of the home care subsidy. Section VI presents additional robustness checks concerning the slot entitlement, which scrutinize our estimation strategy and confirm our main results.
Furthermore, our estimation strategy requires that parents did not time their births systematically around the eligibility cutoff. This assumption is particularly credible in our context due to the political process surrounding the implementation of the subsidy. First, the original draft proposal from June 2012 (Bundestag Drucksache 17/9917) listed January 1, 2012 as the eligibility cutoff date; thus, parent’s potential eligibility status had already been determined six months prior to the draft proposal. Second, without receiving any public attention (FAZ 2013), the eventual cutoff date (August 1, 2012) was introduced between the draft and the final law, which was passed in November 2012 (Bundestag Drucksache 17/11404). This unexpected and publicly unnoticed change effectively ruled out any strategic timing of births around the cutoff, such that parents with children born before August 2012 were surprised that they were not eligible for the subsidy once it was introduced.4 In the empirical analysis, we provide several tests and robustness checks that support the absence of strategic birth timing.
In July 2015, the German constitutional court abolished the home care subsidy on the grounds of interference with the autonomy of federal states concerning family policy. As previously approved claims were not revoked, the subsidy was not abolished at a specific date or from some birth date onwards, but rather it faded out. We thus cannot apply the same research design to estimate and compare the effects of the subsidy’s abolition.
III. Data and Descriptive Statistics
We use three different data sets to investigate the effects of the home care subsidy. In particular, we employ administrative social security records on employment, a representative childcare survey for childcare arrangements, and administrative data from mandatory school entrance examinations from Schleswig‐Holstein, a federal state in West Germany. Online Appendix Table A.2 provides a concise overview of the different data sets. As the state of Baden‐Wuerttemberg augmented its supplementary state‐level parental benefit for low‐income parents (Landeserziehungsgeld) shortly before the introduction of the national home care subsidy, we exclude all families residing in this state. Although our main analysis focuses on West Germany due to data limitations for East Germany, we supplement this analysis with a comparison of the effects in both parts of Germany in Section VII.5
A. Data on Employment Outcomes
To examine employment, we use the Integrated Employment Biographies (IEB V13.00.00 in combination with BEH V10.02.01) provided by the Institute for Employment Research (IAB). The data contain daily information on workers’ employment, education,6 and daily wages until December 31, 2016.7 The data include all workers covered by the social security system, but do not include civil servants and the self‐employed, thereby covering about 80 percent of all people employed in Germany. For a detailed data description, see Jacobebbinghaus and Seth (2007). As the employer‐reported gross wages and employment durations are used to calculate contributions to and benefits from the social security system, they are extremely reliable.
We examine the following employment outcomes three years after childbirth, that is, when eligibility for the home care subsidy and unpaid parental leave expire: (i) the probability of having returned to work, (ii) cumulated days in employment since childbirth, and (iii) cumulated labor income since childbirth. Unfortunately, the data do not include information on exact hours worked; we therefore examine cumulated outcomes to capture additional changes at the intensive margin.
One shortcoming of the social security records is that they do not provide direct information on childbirth. To overcome this issue, we use the procedure by Müller and Strauch (2017) to identify births and the expected date of delivery.8 Employers need to notify social insurances about the start date of the maternity leave period, that is, six weeks before the expected date of delivery. As the same notification code is also used in some cases of longer illnesses, the method by Müller and Strauch (2017) employs several checks to ensure that the notification reflects a childbirth and, if so, identifies the expected date of birth as six weeks after the beginning of maternity leave. Thus, the method identifies the expected date of delivery, which may differ slightly from the actual date of childbirth. To avoid misassigning mothers between the treatment and the control group, we exclude births with an expected date of delivery in a window of two weeks around the eligibility cutoff date (August 1) in each year. Higher‐order births are typically unobservable when the mother does not return to work between consecutive births. Our analysis thus focuses on the employment effects for first‐time mothers.
Our employment data include all women whom we identify as giving birth for the first time between 2008 and 2013 and who worked subject to social insurance directly before giving birth. We exclude mothers who receive social assistance (ALG II) two years prior or within three years after giving birth, as the home care subsidy affected these mothers differently.9
The first six columns in Table 1 provide the descriptive statistics for mothers of children born in 2012, focusing on West Germany. Overall, our estimation sample consists of 506,919 births.
Descriptive Statistics—Employment Data
B. Data on Childcare Choices and Take‐Up
To examine childcare choices and take‐up of the home care subsidy, we use the KiBS (Kinderbetreuungsstudie) childcare panel survey, which has been conducted annually by the German Youth Institute (DJI) since 2012. The survey follows a two‐stage sampling scheme. It first randomly samples municipalities by federal state and municipality size and then samples children from these municipalities. To account for this sampling scheme, we use the provided sample weights throughout our analysis. The survey asks parents about the childcare arrangements of the sampled child and provides detailed information on sociodemographic characteristics of the parents and the household. Importantly, the study includes children’s year and month of birth, which determine potential eligibility for the home care subsidy. Even though the KiBS is a panel study, the questionnaires differ between waves, which prevents us from including some variables in our DiD estimations and which limits the range of potential robustness checks. For a more detailed description of the survey, see Alt et al. (2018).
As outcome variables, we examine whether parents (i) use subsidized childcare (that is, childcare centers and child‐minders), (ii) use informal care arrangements (that is, care by grandparents or any other person), and (iii) are the sole providers of childcare, which we define as the absence of formal and informal care arrangements. For children born in the years 2011 and 2012, this information is collected between 2012 and 2014 (that is, waves 2 and 3). Parents can use several forms of subsidized and informal childcare at the same time. In our sample, we observe the following usage rates: 25.9 percent subsidized childcare centers, 7.9 percent subsidized child‐minders, 50.3 percent grandparental care, and 7.7 percent any other person (for example, au pair). Around 29 percent of parents do not use any of these forms of childcare. The data do not provide information on the intensity of use for these childcare forms.
The survey includes information on the take‐up (current and ever) of the home care subsidy in wave 4 of the survey (collected in 2014 and 2015). We use information on whether families ever receive the subsidy to calculate the take‐up rates.
Table 2 provides descriptive statistics for mothers of children born in 2012, focusing on West Germany. Our final estimation sample consists of 6,932 observations.
Descriptive Statistics—Childcare Data
C. Data on Children’s Development
To examine children’s development, we use administrative data from school entrance examinations that cover the full population of children from one West German federal state with three million inhabitants, Schleswig‐Holstein. Schleswig‐Holstein closely resembles all other West German states in demographic and socio‐economic characteristics.10 The examinations are mandatory in Schleswig‐Holstein and are carried out before children enter primary school at age six. During the medical examination, a public health pediatrician examines children’s development in numerous dimensions and ultimately assesses whether children need any additional support.11
Our data contain information for all school entrance cohorts between 2014 and 2019. A school entrance cohort includes children born between July of the previous year and June of the year of school entrance. We therefore have information for all children who where born between July 2007 and June 2013. To rule out that differences between examination cohorts, for example, changes in examinations or assessment criteria, confound our analysis, we only compare children born in July and August of a given calendar year, as they are part of the same school entrance cohort.12
The data contain several variables that capture children’s development. For all children, the data include an overall assessment by the pediatrician on whether a child needs additional support to attend school. This measure identifies children in the lower part of the skill distribution and provides parents and teachers with important information concerning additional support measures. Around 20 percent of children in our data need additional support. As previously stated, similar measures of school readiness predict later educational achievement (for example, Duncan et al. 2007; Grissmer et al. 2010), health outcomes, and labor market performance (for example, Cunha et al. 2006). We will therefore use this indicator as our main measure of children’s development.
For a subset of children, the data furthermore include information about children’s cognitive and noncognitive development from a screening tool that was developed specifically for school entry examinations (Sozialpädiatrisches Entwicklungsscreening für Schuleingangsuntersuchungen, SOPESS; see Oldenhage, Daseking, and Petermann 2009; Petermann 2009). Even though some districts do not use the screening tool, the SOPESS is available for 77 percent of all children in our sample. It collects information in a standardized way for the following dimensions: language, cognition, counting and numeracy, hand–eye coordination and motor skills, and attention. To avoid multiple hypothesis testing, we construct an overall development index based on the individual test scores from the SOPESS.13 We standardize this average SOPESS score and refer to it throughout the rest of this paper as the children’s development index.
The binary support needs indicator and the development index capture different aspects of children’s development. Specifically, the development index complements the support needs indicator by capturing changes over the whole distribution and yields effect sizes that are easier to compare across studies. In our sample, children with additional needs score 0.75 standard deviations (SD) lower on the development index compared to children without additional needs.
The data further contain children’s year and month of birth and the month of the examination, as well as some information about family characteristics, such as parental schooling, the main language spoken at home, and parents’ country of birth. Information on family characteristics is reported voluntarily by the accompanying parent (typically the mother). The data do not contain information about parental income. Table 3 provides descriptive statistics for mothers of children born in July and August of 2012. Our main estimation sample consists of 19,191 observations.
Descriptive Statistics—School Entrance Examination Data
IV. Empirical Strategy
The cutoff date for the home care subsidy (August 1, 2012) partially coincides with the start of the childcare year, which begins in August or September, depending on the federal state. The start of the childcare year leads to systematic variation in childcare attendance and, consequently, also mothers’ labor supply between children born before and after August. We therefore use a difference‐in‐differences (DiD) strategy that corrects for the difference in outcomes between children born in the same calendar months, but in years not affected by the subsidy.14 Specifically, we distinguish between mothers giving birth between March and July (control group) and those giving birth between August and December (treatment group) of the same calendar year. The identifying comparison then comes down to asking whether any differences between these two groups differ between children born in 2012 and those born in other years.
To estimate the effect of the home care subsidy on different outcomes, we specify the following baseline DiD model:
1
where yi represents the outcomes of mothers or children, byeari denotes a vector of indicator variables for children’s year of birth, autumni indicates whether a child was born between August and December, and εi is an error term. The coefficient δ yields the average effect of being potentially eligible for the subsidy. For our baseline specification examining mothers’ employment outcomes and childcare choices, we include births within five months of the eligibility cutoff, that is, between March and December; we include only children born in July and August when investigating children’s development (see Section III.C). To increase precision and to check the robustness of our estimations, we add predetermined maternal observable characteristics and region fixed effects as available in the different data sets. For childcare choices and children’s development, we also include indicator variables for children’s age in months.
To provide unbiased estimates for the causal effect of the home care subsidy, our identification strategy requires three main assumptions. The first assumption requires common trends in potential outcomes between the treated and control groups. The event‐study graphs and placebo tests that we discuss in Section V support the common trends assumption.15
Second, our identification strategy requires that the composition of the treatment and control groups does not change differentially over time. To investigate this assumption, we perform covariate balancing tests within the DiD framework and replace the dependent variable in Equation 1 with each of these explanatory variables. The final two columns in Tables 1–3 present the results. Overall, they support the absence of statistically or economically meaningful compositional changes. The differences in children’s age at the time of the KiBS survey are caused by changes to the timing of the survey and do not threaten our estimation strategy.16 We also observe no shifts in the distribution of births (see Online Appendix Figure A.6). Hence, not only does the institutional setting minimize the scope for selection into the treatment, but the absence of such selection is also supported by the data.
Third, our strategy requires that no other policy changes differently affected childcare choices or employment outcomes for the treatment and the control group. As explained in detail in Section II, we can cleanly disentangle the effect of the home care subsidy from the entitlement to a childcare slot because the eligibility criteria differ; we conduct several robustness checks, which support our main conclusions. We are not aware of any other policy changes.
We can thus exploit the difference in potential eligibility to the home care subsidy by birth month to explore the subsidy’s effects on maternal employment outcomes, childcare choices, and children’s development.
V. Effects of the Home Care Subsidy in West Germany
A. Take‐up of the Subsidy
As a starting point for our analysis, Panel A of Figure 1 depicts whether families ever claimed the home care subsidy by children’s month of birth for children born between March 2012 and December 2012. The figure clearly shows that take‐up jumps discontinuously by around 60 percentage points between July and August. Although families with children born before July 2012 were not eligible for the subsidy, a few of these families reported that they received the subsidy. These answers most likely reflect reporting errors. From August onwards, the share of families receiving the subsidy remains fairly constant. Thus, around 60 percent of potentially eligible families used the subsidy. Combining this share with the average granted duration of 19.8 months conditional on receiving the subsidy (Alt et al. 2015, p. 59) yields an average duration of receipt for potentially eligible families of around 12 months.
Usage of Home Care Subsidy and Subsidized Childcare
Source: Own calculations based on KiBS data from Wave 4 (interviews in 2014–2015).
Notes: Panel A shows whether families ever receive the home care subsidy for the relevant child, by birth month for children born in 2012 using cross‐sectional survey weights (N = 2,204). Panel B shows take‐up rates at the time of the interview, by birth month for children born in 2012 using cross‐sectional survey weights (N = 2,196).
Panel B of Figure 1 depicts take‐up rates at the time of the survey. To determine the share of eligible families using the subsidy, we additionally show the proportion of children who (i) are in subsidized care, (ii) are neither in subsidized care nor in receipt of the subsidy, and (iii) report claiming both (which was legally not possible, so likely reporting errors). The figure shows the expected discontinuity in take‐up of the home care subsidy between July and August. Once children become potentially eligible for the subsidy, we see an almost seamless transition between families not using either type of support and families receiving the subsidy. The figure also reveals that almost 90 percent of the families who were eligible claimed the subsidy at the time of the interview.
To illuminate which types of families received the subsidy most often and thus benefited financially from it, Table 4 reports the take‐up rates among potentially eligible families overall and for the subgroups that we will consider in our analysis. The share of recipients decreases in maternal education, but does not differ by families’ migration background, either measured by the main language spoken at home or by parents’ place of birth.
Take‐up of the Home Care Subsidy
To give a more detailed view on the relationship between families’ resources and the receipt of the home care subsidy, Figure 2 reports take‐up of the subsidy by household’s pre‐birth income and documents that take‐up of the subsidy is hump‐shaped. Whereas 60 percent of potentially eligible households in the lowest decile receive the subsidy, the share of recipients increases up to 77 percent in the fourth decile and drops steadily as income increases further. In the highest decile, 43 percent of families receive the subsidy. Further, the data also indicate that 64 percent of the two‐parent families ever receive the subsidy compared to only 39 percent of the single parents. Given the financial incentives, one may reasonably argue that the subsidy was not designed for single parents because it was insufficient to help single mothers cover (informal) childcare and time for a job.
Take‐up (Ever) of the Home Care Subsidy by Household Income Prior to Birth
Source: Own calculations based on KiBS Wave 4 (2014–2015) for children born between August 2012 and December 2013, West Germany only, children aged 12–36 months, using cross‐sectional survey weights.
For two‐parent families, Table 5 additionally splits the sample by mothers’ and fathers’ pre‐birth earnings, where we use both partners’ pre‐birth earnings as a measure of their earnings potential after childbirth. Conditional on fathers’ earnings, take‐up of the home care subsidy is similar for families with mothers in the first and second earnings tertile and about 20 percentage points lower among families with mothers in the highest pre‐birth earnings tertile. Conditional on mothers’ earnings, take‐up is somewhat higher when the father earns a medium income, though the differences by fathers’ income are rather modest. We further examine take‐up by the relative income distribution within couples before the first child arrived to explore the potential role of gender norms. We find that the take‐up of the subsidy decreases across the share of the household labor income that is earned by the mother, from nearly 80 percent in families where the mother’s income share is below 10 percent to just above 20 percent when the mother’s income share is above 90 percent of the overall pre‐birth household income (see also Online Appendix Figure A.7). These patterns show that families where mothers have lower opportunity costs of providing childcare used the subsidy more often and, further, are in line with more families using the subsidy where gender norms are more traditional.
Take‐up of the Home Care Subsidy by Parents’ Wage Tertile Prior to Birth
Overall, this section shows that the subsidy did not accrue to the families with the least economic resources, but rather to families with lower‐medium resources and to more traditional families where the mother has lower earnings (potential) than the father.
B. Employment Effects
Figure 3 shows mothers’ employment outcomes. Panel A shows the outcomes in levels separately for mothers of children born in spring and autumn from 2008 to 2013, and Panel B plots the corresponding event‐study coefficients (with 2011 as the reference year). Prior to the introduction of the home care subsidy, women who give birth in autumn have better employment outcomes three years post‐birth than mothers who give birth in spring. These level differences, which are likely caused by the start of the childcare year, are stable across the different birth cohorts until 2011. When the subsidy is introduced, this pattern reverses. Mothers who were potentially eligible for the subsidy (with children born in autumn of 2012) exhibit a lower likelihood to return to work within three years, were employed fewer days, and had slightly lower earnings compared to mothers who gave birth in spring of 2012. For 2013, when all mothers were potentially eligible for the home care subsidy, the difference between mothers giving birth in spring and autumn reverses to the pattern observed until 2011.
Employment Outcomes
Source: Own calculations based on BEH data for for children born between March 2008 and December 2013, West Germany only.
Notes: Estimations for the event‐study graph are conditional on the control variables included in the main analysis, as presented in Table 6.
Table 6 presents the regression results for the three employment outcomes. The first row reports our baseline DiD estimates according to Equation 1 with controls for maternal characteristics (education, age at birth, and pre‐birth wages) and district‐level fixed effects. Being potentially eligible for the subsidy reduced the probability to have ever worked within 36 months after childbirth by 1.4 percentage points (−1.7 percent). The smaller probability to return to work translates into lower work experience and cumulated earnings in these 36 months. The reform reduced the cumulated time in employment by 13.5 days (−3 percent) and the cumulated labor income by 907 euros (−3.2 percent). Overall, employment does not respond strongly to the increased childcare costs.17
Employment Outcomes—Main Results
Panels B and C of Table 6 present the results by educational attainment and citizenship. Prior to the introduction of the reform, critics worried that the subsidies’ employment disincentives would have larger effects on employment (and childcare choices) among nonnatives or low‐educated mothers due to their lower wages. Panel B shows that the effects on the probability to return to work and on cumulated days in employment are smaller for mothers with tertiary degrees than for mothers without tertiary degrees. Due to the higher earnings of mothers with tertiary degrees, the employment effects translate into a loss of cumulated labor income of similar relative size in both groups. Splitting the sample by citizenship (Panel C) yields no difference in the effect sizes, though the effect remains statistically insignificant for nonnative mothers due to the smaller sample size. Finally, Panel D of Table 6 shows the results by childcare availability (above and below the sample median), calculated as the share of childcare slots for children younger than three divided by the number of children younger than three in a given district. The estimates indicate that the effect with regards to all three outcomes is larger in magnitude in districts with higher childcare availability. These results suggest that women may react more flexibly to the reform if actually given a choice, that is, when the supply of childcare is larger. The differences between the groups are statistically insignificant throughout Panels B–D.
Comparing the share of potentially eligible families who received the subsidy (62.4 percent) with the estimated effect on the likelihood of returning to the labor market (−1.4 percentage points) implies a share of inframarginal recipients as large as 98 percent. We arrive at a similar share of inframarginal receipt when we take into account whether mothers changed the timing of their return to employment after childbirth. For this calculation, we compare the average effect on days in employment to the average days of subsidy receipt. As potentially eligible mothers were employed for 13.5 days less and received the subsidy for on average 360 days, the vast majority of days in receipt of the subsidy (96 percent) were inframarginal with respect to employment.
C. Effects on Childcare Choices
Despite the small effects on maternal employment outcomes, the home care subsidy could still have affected childcare choices. As the home care subsidy increased the costs of subsidized care, we expect parents to shift away from using subsidized childcare and instead provide exclusive parental care or rely on informal care. Note that taking up the subsidy does not mechanically reduce the use of subsidized childcare because many families would not have used subsidized childcare regardless of the subsidy. Childcare usage rates for children below age three are quite low in West Germany, amounting to 30 percent in our sample. Moreover, parents could also have increased informal care instead of parental care. Thus, the overall effect is unclear and not a mechanical consequence of the reform.
Table 7 presents the estimated treatment effects on childcare choices—that is, on using exclusively parental care, on using subsidized care, and on using informal care—controlling for maternal characteristics, state fixed effects, survey year fixed effects, and children’s age. Panel A shows that the home care subsidy on average increased the share of children exclusively cared for by their parents by 6.5 percentage points. This effect is statistically significant at the 10 percent level. The subsidy furthermore reduced the probability of attending subsidized childcare by 4.2 percentage points, which is marginally not statistically significant. Nevertheless, the magnitude of the effect is economically meaningful. The reduction in subsidized childcare attendance (prior to age three) is three times as large as the effect on returning to work within three years. Furthermore, our results indicate a reduction in the use of informal childcare by 2.6 percentage points, though the estimate is statistically insignificant. Overall, the results document a shift from both subsidized care and informal care towards exclusive parental childcare.
Childcare Choices—Main Results
As critics worried that the reform would crowd out and displace formal care with lower‐quality informal or parental care—particularly in families with low education and migration background—we again split the sample by these characteristics. The estimates are, however, relatively noisy in some subsamples. For mothers without a tertiary degree, we find an increase in exclusive parental care by 9.1 percentage points and reductions in the use of subsidized and informal care by 3.9 percentage points and 4.6 percentage points, respectively. For mothers with a tertiary degree, we do find only a small increase in exclusive parental care and a reduction in using subsidized care of 5.8 percentage points, but no change in informal care.
Next, we investigate the reform’s effects by migration background using two different definitions of migration background. We first split the sample by the main language spoken at home. For native families, the results presented in Panel C again reveal a statistically significant increase in the use of exclusive parental care alongside a (statistically insignificant) reduction in the use of subsidized and informal care; we observe a similar pattern for foreign families, though these estimates are less precise. Panel D examines the effects by parental country of origin. Again we find similar increases in exclusive parental care for both groups, though the point estimates for foreign foreign families now suggest a larger decrease in using subsidized care and no change for informal care.18
Comparing the take‐up of the subsidy (62.4 percent) with the estimated effects on the probability to use subsidized childcare (−4.2 percentage points) implies that around 93 percent of the home care subsidy accrued to inframarginal families who would not have used subsidized childcare anyway. The share of inframarginal families is slightly lower for exclusive parental care (90 percent) and higher for informal care (96 percent).
Overall, families responded to the home care subsidy by providing more exclusive parental care. Although families used less subsidized care, they did not substitute subsidized care with informal care. Rather the results show that families also used less informal care. We find no effects of the subsidy on childcare choices at ages five and six, that is, around the time the school entrance examinations take place (see Online Appendix Table A.6).
D. Effects on Children’s Development
We next examine the effects of the home care subsidy on children’s development. We present children’s skill development in levels in Panel A of Figure 4 and the corresponding event‐study coefficients (relative to 2011) in Panel B. The first column shows the results on support needs (higher values indicate worse development) for the full sample, the second column depicts support needs for children for whom the standardized development index is available, and the third column displays the results for the development index (higher values indicate better development).
Children’s Development
Source: Own calculations based on school entrance examinations data for children born in July and August between 2007 and 2012, Schleswig‐Holstein only.
Notes: Panels in the first column include all children, and panels in the second and third columns include children for whom the development index is available. Estimations for the event‐study graph are conditional on the control variables included in the main analysis, as presented in Table 8.
Figure 4 reveals no pre‐reform differences in the evolution of the share of children with support needs between August‐ and July‐born children, which lends credibility to the common trends assumption. We then observe a decline in the share of children with support needs for children potentially eligible for the home care subsidy, that is, born in August 2012. The decline indicates that these children benefited from being potentially eligible for the home care subsidy. The patterns for the development index are less clear as the index fluctuates more strongly than the needs indicator. Before the reform, the index increased slightly for children born in August relative to children born in July, though Panel B shows that the changes are well within the range of statistical uncertainty. In particular, in comparison with the earlier changes, the event‐study estimates suggest that the home care subsidy did not substantially increase the development index for potentially eligible children.
Table 8 presents the DiD regression results according to Equation 1. We report the estimates for the complete sample in the first column and for the sample for which the development index is available in Columns 2 and 3. The estimates for the full sample show that being potentially eligible for the home care subsidy reduces the share of children with support needs by 3.6 percentage points, a substantial effect in comparison to the baseline of 22.5 percent. We estimate similar improvements (−3.3 percentage points) for the subsample of children for whom the development index is available. The development index, in contrast, improved only slightly, if anything. The point estimate suggests a small improvement by 0.04 SD, though this effect is not statistically significant, and the graphical analysis calls for additional robustness checks, due to the patterns described in the previous paragraph.
Children’s Development—Main Results
We examine effect heterogeneities in Panels B–E. We find positive effects irrespective of maternal education and parental country of origin. The effect is more pronounced for children living in districts with a higher supply of childcare, that is, lower excess demand. Although the difference by childcare supply is not statistically significant, the pattern is consistent with our finding that female employment reacts more strongly when the supply of childcare is larger. The only group for which we do we not find any positive effects are children who do not speak German at home. The point estimates in Panel C even suggest an increase in support needs by 1.3 percentage points, though the estimation is fairly imprecise due to the sample size.
To probe the effects on children who do not speak German at home, we examine the effects of the home care subsidy on specific skills and on childcare choices in more detail. Regarding specific skills, we additionally generate summary measures for cognitive and noncognitive development based on the subcomponents from the development index as shown in Online Appendix Table A.7 (we adjust the p‐values for multiple testing as proposed by Romano and Wolf 2005). Our results provide suggestive evidence that children who do not speak German at home perform worse regarding cognitive skills overall (−0.11 SD) and, in particular, language skills (0.37 SD). Rather, they performed better in noncognitive skills overall (+0.09 SD), most likely because these domains, such as motor skills and attention, rely less on speaking German.19 As noncognitive skills are less important than cognitive skills in determining children’s support needs (see Online Appendix Table A.9), these effects offer an explanation for the increased support needs of children who do not speak German at home.
Regarding the role of childcare choices for children’s development, Online Appendix Table A.10 provides suggestive evidence that the home care subsidy reduced center‐based care for nonnative children but not the use of child‐minders. As care by child‐minders is typically considered to be of lower quality than center‐based care (for example, Datta Gupta and Simonsen 2010), these changes in childcare choices fit well with our findings on children’s development. Even though all of these differences are statistically insignificant, they consistently indicate that the home care subsidy did not foster the development of children who do not speak German at home and might even have impeded their cognitive development, presumably by reducing children’s exposure to the local language.
As we have shown, children who do not speak German at home did not benefit from the subsidy, even though they have the highest support needs irrespective of the home care subsidy (see Table 8). This finding suggests that the subsidy increased early life inequalities. To shed more light on this, in a final step, we also differentiate the treatment effect by children’s predicted support needs. Specifically, we estimate a probit model for the probability that a child has support needs depending on children’s and parents’ predetermined characteristics using data on children born before 2012, that is, the cohorts not affected by the home care subsidy. From the estimated coefficients, we predict the probability to have support needs for all children and obtain the treatment effect by tertiles of this predicted probability. Online Appendix Table A.11 shows that the beneficial effects are concentrated among children with a medium probability to have support needs. For this group, we find a reduction of 5.8 percentage points against a baseline of 15.3 percent. In contrast, the effects are smaller for children with a high probability to have support needs (3.1 percentage points against a baseline of 36.6 percent) and close to zero for children with a low probability.20 The point estimates for the development index, although imprecise, generally confirm this pattern of increased early life inequalities.
VI. Robustness Checks
To corroborate the validity of our main results, we perform a number of additional robustness checks. We start with a series of robustness checks concerning the effects on employment. Regarding our results on childcare choices and children’s development, we apply the robustness checks that apply to the respective data sets and are feasible given their limitations.
We begin with the employment outcomes in Table 9. Panel A replicates our baseline specification for comparison. In Panel B, we begin by estimating a standard DiD model without any covariates. To account for changes in employment trends that differently affect mothers giving birth in spring and autumn, we next add group‐specific linear and quadratic time trends to our main specification. If the treatment and control groups were on different time trends prior to the reform, including these time trends would change the estimation results. Overall, the results presented in Panel B show that these modelling choices do not affect our conclusions.
Employment Outcomes—Robustness Checks
In Panel C, we evaluate the sensitivity of our results with respect to changes in the observation window. To mitigate concerns about measurement error in children’s date of birth in the employment data (see Section III), we now exclude all children born in the months of July and August to avoid misclassification, and our point estimates hardly change. Next, we make the treated and control groups more similar by narrowing the observation window from ten to six and to four months, while still excluding mid‐July to mid‐August as in our baseline specification. Again, the results are almost identical to our baseline specifications.
In Panel D, we perform several checks to address the concern that the introduction of a legal entitlement to a slot in subsidized care might confound our results. First, we directly control for children’s months of slot entitlement before turning three, based on the date of birth. The estimates are almost identical to our baseline specification. Second, we additionally control for the availability of subsidized childcare at the district level to investigate whether differences in the provision of childcare affect our results. Again, the results are unaffected.
Third, we run a placebo test based on the following logic. All children born in 2011 became legally entitled to a slot in public childcare on August 1, 2013; this was after the second birthday for children born in spring 2011 and before the second birthday for children born in autumn 2011. This pattern is similar to the difference between children in our treatment and our control group. We therefore use 2011 as our placebo treatment year, pretending that children born in autumn 2011 became potentially eligible for the home care subsidy. Because none of these children were actually eligible for the home care subsidy, comparing these groups isolates differences that stem from becoming entitled to a slot before or after a birthday. The difference in the age at slot entitlement between the groups is, however, twice as large in the placebo treatment than in the actual treatment (five months vs. 2.5 months). The placebo coefficients still show no differences between these two groups; that is, the placebo effects are all small and not significantly different from zero.
Fourth, we estimate a difference‐in‐discontinuities model with separate linear time trends for autumn and spring births fully interacted with birth cohort dummies. The trends should pick up any linear trends, which may arise due to slot entitlements. The point estimates are slightly smaller but in line with the baseline. The effect on the likelihood to return is marginally insignificant in this specification because the standard errors more than double. Overall, we thus conclude that the introduction of a legal entitlement to a slot in subsidized care does not drive our results.
In Panel E, we check whether the top‐coding of income affects our estimations, we exclude all women from our sample whose income was top‐coded in any year. Again, this sample restriction barely affects our estimates.
Table 10 gives the robustness checks regarding childcare choices. For these outcomes, we can neither include group‐specific trends nor perform a placebo test, as our analysis sample contains only births in 2011 and 2012. Panel A reports the baseline results; Panel B shows several specification checks: the results of an estimation without control variables, a model with additional controls for having a partner, as well as partner’s education, including a dummy for missing values in either, in the estimation. Furthermore, we present the main coefficients from a difference‐in‐discontinuities model. Additionally, we present estimation results in which we include maternal age at birth and children’s age as squared and cubic polynomials. Panel C shows estimation results using only births from May to October and from June to September, respectively. Overall, the magnitude of the effects hardly changes, and the estimation results even gain statistical significance in some specification. This supports the robustness of our main results, despite the relatively small sample size.
Childcare Choices—Robustness
Table 11 reports the robustness checks concerning children’s development. For these outcomes, we cannot apply a difference‐in‐discontinuities model, as the data only include births from July and August. As before, Panel A repeats the baseline results. Panel B presents the results from robustness checks using different specifications. We report the results from the DiD specification without control variables, including linear groups‐specific trends and additionally including quadratic group‐specific trends. Including quadratic group‐specific trends doubles the standard errors rendering the effect on support needs statistically insignificant, but it does not reduce the point estimate. Hence, all specifications confirm that the home care subsidy reduced support needs but did not affect the development index. The placebo test in Panel C reinforces that the slot entitlement does not drive our findings.
Children’s Development—Robustness Checks
Finally, we investigate whether the subsidy’s effects shortened birth spacing and whether such an effect may drive our previous estimates. We use both the childcare and the school entrance examinations data to investigate the reform effect on the number of children in the household. Using the childcare data, we measure fertility when children are on average 30 months old. The effect is close to zero and statistically insignificant (the coefficient is −0.029 with a standard error of 0.068). We estimate a similar effect in the school entrance examinations data when children are aged 78 months (coefficient of −0.052 with a standard error of 0.044). Thus, we find no evidence that the subsidy shortened birth spacing or increased fertility. Although we cannot examine mothers’ higher‐order fertility in the employment data directly, we can investigate their labor market outcomes 48 months after childbirth, that is one year after expiry of the home care subsidy. In line with the absence of fertility effects, Panel F of Table 9 shows that the magnitude of the employment effects are similar to those after 36 month. Taken together, the results speak against the idea that mothers shortened birth spacing in response to the subsidy.
VII. Comparison of West and East Germany
To shed some light on the interaction between social norms and family policies, this section explores the effects of the subsidy in East Germany. Specifically, we examine whether families in East Germany responded to the home care subsidy differently from families in West Germany.21 Given the data limitations for East Germany mentioned in Section III, this comparison is noisy regarding childcare choices and not possible regarding children’s development.
As a starting point, Figure 5 depicts the take‐up of the subsidy among families in East Germany. Only 28 percent of families in East Germany ever received the subsidy, which is less than half the share of families in West Germany. The lower take‐up rate likely reflects different childcare patterns that lead to less inframarginal families receiving the subsidy. Whereas 70 percent of children aged two in West Germany did not attend subsidized childcare in 2016, this share amounted to 34 percent in East Germany (BMFSFJ 2016, p. 7). Take‐up of the subsidy is hence almost universal among families who do not use subsidized childcare, but this group is much smaller in East compared to West Germany.
Take‐up Rate of Home Care Subsidy in East Germany
Source: Own calculations based on KiBS data from Wave 4 (interviews in 2014–2015), East Germany without Thuringia.
Notes: Panel A shows whether families ever receive the home care subsidy for the relevant child, by birth month for children born in 2012 using cross‐sectional survey weights (N = 1,096). Panel B shows take‐up rates at the time of the interview, by birth month for children born in 2012 using cross‐sectional survey weights (N = 1,087).
Turning to the subsidy’s effects on maternal employment, Table 12 shows that the subsidy did not reduce maternal employment in East Germany. Furthermore, the 95 percent confidence intervals for the effects in East Germany do not overlap with those for West Germany for all three outcome variables. The effect also does not differ significantly from zero in either subgroup analysis, as shown in Panel B and C of Table 12.22
Employment Outcomes—East Germany
Finally, Table 13 presents the effect of the subsidy on childcare choices. We find a negative effect on the use of exclusive parental care and a positive effect on the use of informal care arrangements. Although we estimate the effect on exclusive parental care rather imprecisely due to the small sample size, our estimates clearly reject the 6.5 percentage points increase in exclusive parental care that was observed in West Germany. We furthermore find no reduction in using subsidized childcare. The home care subsidy had different effects on maternal employment and childcare choices in East and in West Germany. One potential explanation for this findings are differences in social norms with regards to gender roles: East Germans still hold more gender‐egalitarian beliefs compared to West Germans, even a long time after reunification (Bauernschuster and Rainer 2012). These differences in social norms could render East German mothers less responsive to the subsidy.
Childcare Choices—East Germany
The results are, however, also compatible with standard economic theory. Conditional on returning to work within three years, mothers in East Germany earn a higher labor income than mothers in West Germany. For instance, mothers who gave birth in spring 2012 and were employed three years after childbirth earned on average 2,130 euros per month in employment in East Germany, but only around 1,940 euros in West Germany, despite the generally higher wage levels in West Germany compared to East Germany. The higher earnings in East Germany suggest it is plausible that a smaller share of mothers are close to the margin of whether or not to return to work and that the home care subsidy therefore has a smaller effect in East Germany. Conditional on using childcare, children in East Germany on average also attend for longer hours than children in West Germany (35 hours per week vs. 28 hours for children born in the first half of 2012). The difference may also partially explain why fewer families responded to the lump‐sum subsidy in East Germany.
VIII. Discussion and Conclusions
This work investigates the effects of the introduction of a home care subsidy in Germany in 2013 on maternal employment, childcare choices, and children’s development. We make three contributions to the literature. First, we use high‐quality social security records on employment alongside detailed household survey data on take‐up and childcare choices to examine the effects of the subsidy on maternal employment and childcare arrangements. Second, we examine how the subsidy affects children’s skill development using administrative data from mandatory school entrance examinations conducted at age six. Third, we document in detail who received the subsidy. To estimate the effects of the subsidy, we exploit a date of birth cutoff in eligibility to the subsidy within a DiD design that removes seasonal patterns in the outcomes.
Our results demonstrate that the introduction of the subsidy slightly reduced maternal employment and increased exclusive parental care in West Germany. The results furthermore show that families did not substitute subsidized care with informal care, but rather indicate a reduction in both informal care and subsidized care. On average, children benefited from the subsidy in terms of reduced support needs when starting school. Whereas these results are rather homogeneous across subgroups, we find one important heterogeneity—children who do not speak German at home have the highest support needs irrespective of the subsidy and are the only group that seems not to benefit from the subsidy, likely because they had less contact with the local language. Additional analyses by children’s probability to have support needs also indicate that the subsidy increased inequality in children’s development. In terms of financial beneficiaries, we show that the subsidy accrued almost completely to families who were not on the margin of changing their employment or childcare choices and transferred income towards specific inframarginal families. As an income transfer, the subsidy did not benefit the families with the least economic resources most, but rather families with lower‐medium resources and more traditional families. Taken together, our results hence suggest that home care subsidies, particularly those that are not means‐tested, benefit the families and children with the strongest need for additional support less than other families, thereby increasing early life inequalities.
In East Germany, take‐up of the subsidy was much lower than in West Germany, and our results indicate that the subsidy neither reduced maternal employment nor the use of subsidized childcare. We also find no evidence that the subsidy increased exclusive parental care. The different effects in East and West Germany point to the importance of social norms for understanding the effects of family policies, though the differences can also be partially explained using traditional economic models.
From a purely fiscal perspective, one might argue that a home care subsidy is beneficial if the savings on subsidized childcare outweigh the expenditures for the home care subsidy. The average monthly total fiscal cost per slot for children aged three and younger in formal childcare amounted to 620 euros.23 Given the initial home care subsidy of 100 euros, the savings on childcare expenditures then outweigh the expenditures for the home care subsidy if at least one out of six families that receive the subsidy no longer uses childcare. This corresponds to a situation with up to 83 percent inframarginal recipients. As their actual share amounts to 93 percent (even in West Germany), the subsidy imposes a net fiscal burden. Moreover, this calculation provides a lower bound of the fiscal burden, as it ignores mothers’ payroll taxes, their social security contributions, the increase of the subsidy to 150 euros in August 2014, and the marginal costs of public funds.
Our findings on the effects of the home care subsidy on maternal employment and on childcare choices line up well with previous evidence on home care subsidies. Regarding children’s development, both Gathmann and Sass (2018) and our study document that such subsidies foster children’s development on average, though they may hinder some children’s development. Our study goes beyond Gathmann and Sass (2018) by using high‐quality administrative data on children’s outcomes and by providing evidence for a specific mechanism for such detrimental effects, that is, that some children are exposed less to the local language. This finding implies that home care subsidies should be accompanied by programs that provide additional support for specific groups, for example, language courses for children with migration background.
From a policy perspective, it is important to understand whether the effect of the home care subsidy on children’s development operates through changes in maternal employment, in childcare choices, or through additional income effects. Our results by childcare availability indicate that the effects on maternal employment and children’s development were more pronounced where more childcare was available, which suggests a role for maternal employment as a mechanism underlying the effects on children’s development. The 1.4 percentage point decrease in maternal employment, however, is too small to explain the 3.7 percentage point decrease in the share of children who need additional support when entering school, absent other mechanisms. The modest change in childcare choices would also require implausibly large effects on those who alter their childcare choices to be the sole cause of the improvement in children’s development. On the other hand, structural models of children’s development that regard children’s skills as a function of investments, which in turn depend on family income, also predict smaller (average) effects on children’s development when ignoring the change in the cost of different care modes that the home care subsidy induced.24 The effect on children’s development thus appears to stem from a combination of these mechanisms, likely in conjunction with heterogeneous effects of the mechanisms for different families.
Given our data, we cannot disentangle to what extent the positive average effect on children’s development stems from changes in maternal employment, parental care replacing subsidized care, parental care replacing informal care, and from the increase in household income for families who did not alter employment and childcare choices. In our setting, families that did not alter maternal employment due to the home care subsidy still changed their childcare choices, and families that changed neither experienced an income increase. These patterns caution against using such policies as an instrumental variable to estimate the effect of a single mechanism on child development. Rather, such analyses need to account explicitly for the other mechanisms, for example, for childcare choices and household income when examining the effects of maternal employment.
More broadly speaking, any analysis of the mechanisms that underlie child development would be facilitated by the availability of better and additional data, such as time‐use data and household expenditure data, as well as more information on the quantity (for example, hours in childcare) and quality of different care modes, ideally also covering parent–child interactions. Researchers could then assess how reforms affect the inputs into children’s skill production function, including consumption and investment goods, and educational or recreational time investments by the parents (see Del Bono et al. 2016; Carneiro and Ginja 2016; Del Boca, Flinn, and Wiswall 2016). For future research, disentangling these different mechanisms, the role of which likely differs across families, promises important insights into how public policies can foster children’s development.
Acknowledgments
The authors thank Luise Görges, Laszlo Goerke, Jonas Jessen, Guyonne Kalb, Regina T. Riphahn, Stefanie Schurer, Katharina Wrohlich, Miriam Wuest, Florian Zimmermann, two anonymous referees, and participants at various conferences and seminars for their helpful comments and suggestions. The authors have nothing to disclose. This paper uses confidential data sets from the IAB, the DJI, and SH. The IAB data can be accessed for replication purposes through the Research Data Centre of the Institute for Employment Research according to the IAB guidelines available at https://iab.de/en/daten/replikationen.aspx. The DJI data can be requested at https://surveys.dji.de/index.php. Access to the school entrance data from Schleswig Holstein can be requested through a confidentiality agreement from the Ministry of Social Affairs, Health, Science and Equality of Schleswig Holstein. The authors are willing to assist. The programs for replication are available from the authors upon request.
Footnotes
↵1. Home care subsidies in Norway, Sweden, and Thuringia are gradually withdrawn as families use more subsidized childcare, and thus the subsidies increase the hourly costs of childcare. Finland, Germany at the national level, and Bavaria paid lump‐sum subsidies that were withdrawn completely if families used any subsidized childcare, and the subsidies thus act as an entrance fee. Both types enlarge families’ choice set and, in particular, raise their consumption opportunities if they do not use subsidized care (and do not participate in the labor market). Hence, a static model of labor supply predicts that both types of home care subsidies reduce labor supply.
↵2. One additional study by Bettinger, Hægeland, and Rege (2014) also documents that the Norwegian home care subsidy positively affected the grade point average (GPA) of older siblings of eligible children in tenth grade.
↵3. Since August 1996, children from age three until school entry had already been legally entitled to a slot in subsidized childcare. Enrollment rates for children aged three to five in formal childcare amounted to 95 percent in 2016 (OECD 2017b).
↵4. Google Trends data on searches regarding a cutoff date for the home care subsidy confirm this surprise. Online Appendix Figure A.1 documents hardly any searches for a cutoff date until late 2012. Rather, searches for the cutoff spiked dramatically in August 2013, when many families eventually realized they were not eligible for the subsidy.
↵5. Two data problems arise for East Germany. Most importantly, we lack administrative data on children’s skill development; second, the sample on childcare choices is relatively small, making the analysis imprecise.
↵6. We use the procedure by Thomsen, Ludsteck, and Schmucker (2018) to impute missing information on education. We still have around 4 percent of missing values in our data after the imputation and account for this using a dummy in the estimation.
↵7. The information on wages is censored at the social contributions ceiling. However, 98.7 percent of mothers in our sample have pre‐birth wages below this threshold. For observations with wages above this ceiling, we recode their wages to the threshold value in the respective year. To check whether the top‐coding affects our results, we exclude all women whose wage was above social contributions ceiling at any time during our analysis period in a robustness check, see Section VI.
↵8. Using the same data source to examine the effect of changes in parental leave regulations on mothers’ careers, Schönberg and Ludsteck (2014) applied a similar method.
↵9. We also checked whether the reform affected the probability of receiving social assistance at any time within 36 months after childbirth. The results presented in Online Appendix Table A.3 show that the reform did not affect take‐up of social assistance.
↵10. Kuehnle and Oberfichtner (2020) use data from the German Microcensus to show that the demographic and socio‐economic characteristics of the population of Schleswig‐Holstein are almost indistinguishable from the population of other West German states, with the exception of urbanicity and migration background. To further alleviate concerns about comparability, we calculate take‐up rates in the KiBS data and find similar rates for Schleswig‐Holstein (57 percent) and West Germany (62 percent). Online Appendix Figure A.2 provides information on take‐up by birth month for Schleswig‐Holstein. Finally, Online Appendix Table A.4 documents that we find similar employment results for Schleswig‐Holstein as for entire West Germany.
↵11. This data source has been used used in other contexts. See Felfe and Zierow (2018); Felfe and Lalive (2018); Felfe, Rainer, and Saurer (2020); Huebener, Kuehnle, and Spiess (2019); and Kuehnle and Oberfichtner (2020).
↵12. We exclude observations from children who take part in the examinations but do not belong to the regular school entrance cohort in that year, which happens if children start school earlier or are deferred from starting school. As there is no person identifier in the data, some children would otherwise be included in the data more than once. The restriction creates a gender imbalance in our sample because girls are more likely to start school earlier. That said, Online Appendix Table A.5 shows that there is no relationship between the home care subsidy and share of children not belonging to the regular school entrance cohort, overall and by sex.
↵13. We construct the combined SOPESS score as an overall measure of children’s development as follows. We first standardize each score. Second, we average each child’s available scores. Third, we standardize these averaged scores. We standardize the scores using information on all children. As the children in our analysis sample are the oldest in their cohorts, they tend to perform better than average. See Online Appendix Figure A.3 for the distribution of the resulting development index, and Online Appendix B for more details on the construction of scores for the underlying subdomains. When reporting the results for the specific score underlying the average SOPESS score, we apply the Romano‐Wolf multiple‐hypothesis correction method (Romano and Wolf 2005) to control for the familywise error rate in outcomes.
↵14. Forty‐four percent of children who attend center‐based care start childcare in August and September, but only 15 percent start in June and July (Online Appendix Figure A.4). Depending on whether a mother gave birth in spring or in autumn, we measure the employment outcomes before or after the start of the new childcare year, which might give rise to differences in labor market outcomes. As shown in Online Appendix Figure A.5, mothers of children born in autumn indeed return more quickly to the labor market. Because such a difference close to the cutoff threatens the validity of a regression discontinuity approach, we use a DiD strategy. Section VI additionally present results from a difference‐in‐discontinuities approach. These estimates are much less precise, but still confirm our main results.
↵15. As the KiBS data do not contain information on childcare choices at the same age for for children born before 2011, we cannot investigate the pre‐treatment trends for these outcomes.
↵16. The timing of the interviews differs slightly across the groups. Specifically, families with children born in spring 2012 were interviewed slightly more often at the end of 2013, and families with children born in autumn 2012 were interviewed more often at the beginning of 2014. Because this does not coincide with the typically starting date of attending childcare (see Online Appendix Figure A.4) and because we control for children’s age at interview using monthly dummies, this difference does not affect our estimations.
↵17. Exploiting the information on employment at the daily level, Online Appendix Figure A.5 plots the Kaplan–Meier return‐to‐work patterns for the first three years after childbirth separately for spring and autumn births from 2011 and 2012. The figure shows that return‐to‐work patterns are similar between the four groups during the first year after childbirth, and that they start to diverge between autumn and spring births when children are aged 13–24 months.
↵18. We cannot examine differences in the effects on childcare choices by the regional availability of childcare because the KiBS does not provide district‐level information.
↵19. For some children, only some of the subcomponents are available. Online Appendix Table A.8 therefore replicates the estimations for the smaller sample children for whom all subcomponents are available. The described patterns are even more pronounced in this group.
↵20. Online Appendix Table A.12 shows the marginal effects from the probit estimation to obtain the predicted support needs.
↵21. We omit Thuringia from this analysis because it already had a home care subsidy in place when the federal subsidy was introduced; see Gathmann and Sass (2018) for details.
↵22. We omit the analysis by citizenship, as the sample of foreigners in East Germany is too small for a meaningful analysis.
↵23. We base our calculations on Haider and Schmiedel (2012). One‐third of children in subsidized childcare attended publicly run childcare centers, and the remaining two‐thirds attended childcare run by other providers, for example, churches. For both types of providers, the annual costs per child below age three amounted to roughly 9,900 euros. The government covered 81 percent of these costs for publicly run centers and 73 percent for centers run by other providers. The monthly average costs to the government for each slot and month hence amounted to 620 euros.
↵24. According to the KiBS data, the average net family income of families with a child born in spring 2012 and in which the mother did not work at the time of the survey amounts to 3,200 euros. The home care subsidy of initially 100 euros and later 150 euros hence corresponds to an increase in family income by on average 3.9 percent. Agostinelli and Wiswall (2020) estimate an elasticity of investments with respect to family income of 0.34 and an elasticity of children’s skills with respect to investment of 0.88. Based on their estimates, we would thus expect the home care subsidy to improve children’s development by 1.2 log points of their skill measure. As the standard deviation of their skill measure amounts to 220 log points, this increase is rather small.
- Received July 1, 2020.
- Accepted December 1, 2021.
This open access article is distributed under the terms of the CC-BY-NC-ND license (https://creativecommons.org/licenses/by-nc-nd/4.0) and is freely available online at: https://jhr.uwpress.org.