How Does Early Child Care affect Child Development?
Christina Felfe, University of St. Gallen and CESifo∗ Rafael Lalive, University of Lausanne, CEPR, CESifo and IZA †
First version: October 2009 This version: November 2011
Abstract
This paper analyzes the effects of early child care on child development in a setting with rationed high quality center-based care – Germany. The empirical analysis contrasts the average effect of child care with the effect of child care on the marginal children – those enrolling because supply constraints are lifted. Data on children aged 2-3 years from the German Socio-Economic Panel (GSOEP) and data on children aged 5-10 from the German Child Panel (GCP) provide comprehensive information on the role of attending child care before age 3 on child-development in later stages of life. Results suggest that center-based care initially promotes child development for the average child in child care but these initial gains disappear. In contrast, the effects on the marginal children are positive, even several years after early child care. This difference in gains can be partly explained by differences in children’s socio-economic background; marginal children are from a low socio-economic background with strong gains from child care whereas the average children are from a high socio-economic background with low gains from child care. Center based care improves child development because it substitutes for mothers’ low quality activities.
JEL Classification: J13, I21, I38
Keywords: child care, child development, non-cognitive skills
∗We appreciate comments on previous drafts of the paper by Josh Angrist, Jo Blanden, Amy Challen, Bernd Fitzenberger, Maria Fitzpatrick, Marco Francesconi, Tarjei Havnes, Michael Lechner, Alan Manning, Bj¨orn Ockert, Hessel Oosterbeek, Steve Pischke, Kjell Salvanes, Anal´ıa Schlosser, Emma Tominey, Bruce Weinberg¨ and seminar participants at Albert-Ludwig University, Norwegian School of Economics and Business Adminis- tration, CEP London, CESIfo Munich, DIW Berlin, University Pompeu Fabra, University of Lausanne, Uni- versity of St. Gallen and SOLE, 2011. Address: Christina Felfe, Varnb¨uelstrasse 14 CH-9000 St. Gallen, Christina.Felfe@unisg.ch.
†Address: Rafael Lalive, Department of Economics, University of Lausanne, CH-1015 Lausanne-Dorigny, Rafael.Lalive@unil.ch.
1 Introduction
What role does child care play in the process of children’s skill acquisition during early life? How long does early child care affect child development? Investigating these questions is important for several reasons. First, policies aiming to encourage female labor force participation via expanding the supply of institutionalized care need to take the potential effects on child development into account. Second, providing evidence on the specific role of care provided by the mother in comparison with center-based care is relevant for a better understanding of the process of skill acquisition early in life. Third, understanding the effects of early child care is important because early life conditions have long lasting effects on the formation of cognitive and non-cognitive skills (Heckman and Materov, 2007).
The focus of this study lies on the impact of shifting hours of care provided by the mother to hours of care provided by highly regulated center-based care on children’s short- and medium- run development. Center-based care may affect child development for at least two reasons. First, a child spends less time with her or his mother but more time in child care. Whether child care improves child development or not therefore critically depends on the relative quality of care.
Second, as stressed in current policy debates, child care frees up time for market work hence increases family income.
Clearly, whether substitution generates positive or negative effects for child development depends on the mother’s quality of care relative to the quality of care provided by the center.
This is why we are particularly interested in understanding who is affected by expanded access to highly regulated care. We distinguish between the following two different effects of care provided by highly regulated child care centers to children aged 0 to 2 years – i.e. during early childhood.
First, we compare children who spend some time in formal care to children who do not. This analysis provides information on the effect for the average child placed in child care. Second, we are interested in the causal effect of exposure to non-parental care on the skill acquisition of children who only attend center-based care once access is less restricted. This analysis provides information on the effect of child care on the extra child attending child care. These effects may differ depending on how child care slots are rationed.
Two empirical strategies serve to measure these effects. The first strategy compares children with some exposure to early child care to children who were never exposed to early child care.
This approach provides information on the effects of child care for the average child placed in child care, but it may be affected by selection bias. Children from advantageous socio-economic backgrounds are more likely to be exposed to early child care. We address selection by correcting for a large number of observed individual and regional characteristics. These characteristics are
effective in balancing the quality of care provided by the mother.1
The second identification strategy relies on variation in the amount of child care slots avail- able to parents. We use shocks to child care supply that cannot be predicted by a wide range of important predictors of demand of child care as an instrument for child care enrollment. Such supply shocks are arguably exogenous to child development because regional child care provision is a lengthy administrative process. In a first step, authorities at the county level project the need for child care in their region, then non-profit organizations submit their intentions to open child care centers, and finally authorities at the state level must approve the applications for new child care centers. This suggests that the actual number of slots available to any given birth cohort is never identical to child care demand (due to planning errors, delays in construc- tion or approval of child care centers). Our empirical analysis conditions on the region-specific growth in child care slots prior to a child’s birth as well as further regional characteristics which approximate the local demand for child care as projected by the regional authority. As a result, effects of child care on the marginal child are identified from unpredicted shocks in the supply of child care slots at the local level at childbirth.
Two important pieces of evidence are consistent with the assumption that shocks to local child care supply are exogenous. First, we find no systematic correlation between individual predictors of child development and regional child care offer rates once regional indicators have been taken into account. This finding suggests - without proving - that also unobserved in- dividual predictors of child development are not systematically related to the offer rate. And indeed, the child care offer rate is unrelated to indicators of the quality of care provided by the mother. Second, we test the robustness of our results to including regional dummies (East vs West), state dummies or state specific time trends. Results are fairly robust to these sensitivity analyses. Thus, the set of regional controls effectively balance child development potential.
The empirical analysis combines two different datasets that provide information for different cohorts. The first is the German Child Panel (GCP), which contains information on 686 children born 1996/97 who are interviewed at age 5 to 10. Moreover we use data recently collected by the German Socioeconomic Panel (GSOEP) on 762 children born between 2002 and 2005 and being surveyed when they are between 2 and 3 years old. Both datasets contain a rich set of child development measures such as independence, social skills, behavior and personality. The GSOEP also asks about children’s language and motor skills; the GCP contains information on children’s grades obtained in primary school. Moreover, both datasets provide information on the use of center-based care during early childhood as well as on the individual background.
1For instance, mothers who have a child in child care read and sing with their children more often than mothers who care for their child full-time. This difference disappears once we condition on observed characteristics.
Regional statistics are then used to add information on the local availability of center-based care as well as on features of the county where the family resided at childbirth. Hence, both datasets allow discussing the impact of center-based care prior to the age of 3 on a wide range of child development outcomes at age 2-10. Further information provided by the GSOEP, namely a rich module on the hours of non-maternal child care (i.e. provided by the partner, the extended family, informal child care and formal child care) as well as information on a broad range of mother-child interactions allows discussing the different channels through which exposure to center-based care may affect child development.
Our results suggest, first, that child care improves the overall development for the average child placed in child care. In the short run (at age 2-3 years) children in child care are more independent (+0.11 standard deviations) and have better social skills (+0.12 standard devia- tions) than their peers. Children exposed to child care also benefit in terms of language skills – the main cognitive skill we observe. However, in the medium-run (at age 5-10 years) these advantages are no longer present, children who were exposed to child care at age 0-2 even do worse in terms of their school grades in the core subjects at primary school. Second, children who enter child care because the supply of center-based care increases benefit even more strongly from child care than the average child (independence + 1.12 standard deviations, social skills + 0.71 standard deviations, personality +0.67 standard deviations). Marginal children also benefit in the medium-run: they are less dependent on their parents, and are likely to perform better in school at age 5-10 years. The underlying reason for this differential impact is that children who are only enrolled in center-based care once access is less restricted come from a more disadvan- tageous family background and thus have higher gains from center-based care. Third, we find little evidence that child care affects child development via changing household income. Child development is mainly improved because children spend more time in child care centers and time in child care substitutes for time mothers spend on activities with low educational value such as shopping or watching TV.
This paper is related to three main strands of the literature on child development. The first strand of literature investigates the effect of universally accessible child care on children’s skill acquisition. There is a large body of research on the effects of non-parental care during pre-school ages (4-6 years old) (e.g. Magnuson et al. (2004), Fitzpatrick (2008), Gormley Jr.
et al. (2008), Berlinski et al. (2009), Cascio (2009), Havnes and Mogstad (2011)). In contrast to this, the body of research on the effects of non-parental care during early ages (0-2 years) is fairly small.2Baker et al. (2008), for instance, evaluate the introduction of universal child care
2There is some earlier literature in the child development field providing descriptive evidence on the relation between non-parental child care during early childhood and children’s skill acquisition. Belsky et al. (2007), for
in the Canadian province of Quebec. Their findings indicate that lowering the out-of-pocket cost of child care increases the use of child care by 14.6 percentage points, but leads to more hostile parenting styles and lower child well-being (increases anxiety, decreases health). Datta- Gupta and Simonsen (2010) concentrate on the effect of exposure to different types of child care in Denmark and find that children benefit more from center-based care than from low quality (home) care). Using local variation in the availability of child care in Chile, Noboa Hidalgo and Urz´ua (2010) study the short-run effects of child care targeted to children aged 5-14 months.
Their findings indicate gains associated with child care, particularly with respect to motor and cognitive skills.
The second strand focuses on understanding the consequences of maternal employment, es- pecially employment during early childhood, on children’s achievement. While some of these studies show that maternal employment may improve intellectual performance through increas- ing household incomes (Blau and Grossberg, 1992), others have also shown that it is associated with lower outcomes among children (Baum, 2003; James-Burdumy, 2005). Still others suggest that the effects may depend on the characteristics of mothers and families (see Ruhm (2004) and Brooks-Gunn et al. (2002) for a full review of the literature). Several studies evaluate the effect of expanded parental leave on child development. Dustmann and Schoenberg (2009) study the long-run effects of expanding paid parental leave in Germany and find little support for the hypothesis that an expansion in maternity leave improves childrens outcomes. In a similar vein, Wuertz-Rasmussen (2010) finds no measurable effect on children’s development from an exten- sion of parental leave in Denmark (from 14 to 20 months). Baker and Milligan (2010) find no effects on child development outcomes of children aged 2 to 4 years from changes in parental leave in Canada. Finally, Carneiro et al. (2010) focusing on the long-run effects of expanding parental leave immediately after birth, find some positive effects of maternal care on educational and labor market outcomes at age 25 in Norway.
A third strand focuses on the effects of pre-school interventions targeted at disadvantaged children. These targeted interventions have demonstrated strong beneficial impacts on the development of the participating children (see the reviews on Head Start by Currie (2001), Blau and Currie (2006) or Heckman (2007)). Other examples are the Perry Preschool Project, The Abecedarian Program, and the Chicago Child-Parent Centers; for an overview please refer to Blau and Currie (2006) or Heckman and Materov (2007).
This paper complements the existing literature in at least three respects. First, we examine the impact of non-parental care on children’s short (age 2-3 years) and medium run (age 5-10
instance, find that the amount of time spent in non-parental care throughout a child’s first 4.5 years predicts higher vocabulary scores whereas Belsky (2001) finds that it is associated with development risks.
years) development with respect to a comprehensive set of cognitive and non-cognitive outcomes.
This allows us to shed some light on different dimensions of children’s development at the same time, and to address the open question if differences dissipate over time. Second, we explicitly distinguish between motherly care and center-based care and discuss the heterogeneity in the effects of center-based care depending on the quality of motherly care. In this vain, we highlight the tremendous differences in the effects of child care on the average child and the effect on the extra child attending child care only because access is less restricted. This is important in discussing the incidence of expanding high quality care on child outcomes. Third, we provide evidence on the mechanisms that lead to differential child development for children attending early child care.
The remainder of this paper is structured as follows. Section 2 discusses the institutional background. Section 3 provides information on the data sources and a set of key descriptive statistics. Section 4 discusses the main identification strategies. Section 5 presents the main results, and section 6 provides a summary and implications of our findings.
2 Institutional Background
This section discusses historical differences in the child care system across the two former German countries, the GDR and the FRG, provides information on the changes in the child care systems after German reunification, as well as key evidence on remaining regional differences.3
The central aspect of the German child care system is the strong regional variation in child care offer rates. This is mainly due to the different perspectives on family policy adopted in the two German countries prior to German reunification. In the former German Democratic Republic (GDR), the state was responsible to provide child care for children at all ages. Uni- versally accessible child care should facilitate both female labor force participation as well as guarantee national education. Given a high supply of child care (56% of children age 0-2 had a slot in public care) and the allowance for mothers to take one year off after childbirth, there was basically full coverage for children age 1-2 years. In contrast, the former FRG has until recently embraced the view that child care is the responsibility of each family and that mothers should best care for their children until they enter Kindergarten. In support of this view, Germany rapidly expanded parental leave from 3 months in 1979 to 36 months of job-protected, and 24 months of paid leave in 1992. Yet the supply of formal child care for 0-2 year olds remained very limited (1%).
3All facts about the institutional background are taken from OECD (2004), Dittrich et al. (2002) and Riedel et al.(2005)
After German Unification in 1990, the former East adopted all of West German family policy (parental leave, tax system) except for child care. Nevertheless, despite the agreement in the reunification contract to maintain the supply of child care, financial shortages led to a decline of the coverage rate for children aged 0-2 years in East Germany (41.3% in 1994, 36.4% in 1998 and 37.1% in 2002). In West Germany, however, due to a general concern about the relevance of universal child care, the amount of child care centers started slowly to increase (2.2% in 1994, 2.8% in 1998 and in 3.6% in 2002). Despite a slow convergence process, the German child care system is still characterized by enormous regional differences.
The supply of child care slots is determined at the county level. Each of the 439 count counties in Germany is in charge of organizing its own child care system. The process has three steps. In a first step, regional authorities assess the demand for child care once every three years. Authorities identify centers that need to close if current supply exceeds the desired supply. If supply is below demand, non-profit organization submit their proposals to set up new centers. Finally, authorities at the state level must approve the submitted proposals. Approval is necessary for receiving subsidies from the state, which allow providers of child centers to keep fees at a very low level.4 Given unsuccessful applications, delay of approvals or construction, realized supply does not necessarily meets identified demand, but is subject to important and unpredicted shocks.
Figure 1 displays a map shading counties (Kreise) according to the number of full time child care slots available per 100 children aged 0-2 years on Dec 31, 1998 – the offer rate. Clearly, there is striking regional variation in child care offer rates between regions located in East Germany and regions located in West Germany. Generally, offer rates are on the order of 40 % in the East, whereas they are below 10 % in the West. Yet, note that there is also substantial within region variation. Bavaria, the state with the lowest supply of child care in West Germany, offers on average only 1.4 slots per 100 children. There are counties within Bavaria that offer as few as 0.1 slots per 100 children and other counties that offer as many as 7.1 slots per 100 children.
Brandenburg, the state with the highest supply in East Germany, offers on average 52 slots to 100 children aged 0 to 2 years. Coverage varies from a minimum of 43 % to a maximum of 61%.
This shows that there is substantial local variation in child care offer rates. While the supply of child care is certainly related to the socio-economic situation of each county, above described supply shocks represent an important idiosyncratic component to the supply of child care. As a result, counties with the same observed socio-economic situation end up with different offer rates. The empirical analysis focuses on exactly this idiosyncratic component for identification.
4Fees vary across regions and depend on family income in a progressive manner but they are capped at 600 Euros per month for a fulltime slot – which corresponds to about one quarter of the operating expenses.
Figure 1: Slots in center-based care per 100 children aged 0-2
Evidence in Figure 1 and the background on the planning process suggests that regions with low child care offer rates are in a situation with excess demand. As explained above, expanding the supply of subsidized public child care is a lengthy political process. This leads to a situation where counties that would like to send more children to child care simply do not have enough slots available for the parents who apply for child care. To deal with the excess demand, centers operate with waiting lists giving preference to families who sign up their children early. Further criteria which allow children to jump the waiting lists are the working status of their parents, single mothers and siblings who are already enrolled in the child care center. But considering excess demand and the need to register very early, rationing favors children with an advantageous family background - a fact further discussed in Section 4. Parents whose children do not get a slot in center-based care face two options: they can either keep their child at home until Kindergarten starts (age 3 and up) or purchase very expensive informal care on the market.
The costs of the latter option exceed the fees of child care centers by far. As a result, informal care is only used in rare occasions (only 5.5% of the families in our data rely on child minders on a frequent basis).
Besides providing care, centers have a clear educational mission in two dimensions. Educa- tional goals concern developing skills related to pattern recognition, motor skill development, and language skill development. Center staff develop these skills using educational activities and playful activities in support of these skills. Moreover, center-based care is also expected to contribute to the development of social skills such as interacting with others, calling other people by their name, etc.
Does the quality of child care vary across regions? Center-based care for 0-2 year old chil- dren is tightly regulated but oversight is decentralized at the state level (16 states in total).
Regulations concern dimensions such as staff child ratios, space, but also special training and qualifications for the staff before being allowed to work in the sector. Since regulations are at the state level, quality differences could be important across states but not within states. This provides a rationale for assessing the sensitivity of our results to state fixed effects.
3 Data and Descriptive Evidence
The next subsection 3.1 describes the available data sources and the subsection 3.2 provides a descriptive analysis of the relationship between center-based care and child development and discusses potential channels through which early formal care might exert its effect on children’s
development.
3.1 Data sources
The empirical analysis draws upon two different datasets, providing information for different birth cohorts. The first dataset is the German Child Panel (GCP), which contains information on 686 children born 1996 and 1997. The parents of these children are interviewed up to 3 times when their children are between 5 to 10 years old. The second dataset contains data recently collected by the German Socioeconomic Panel (GSOEP) on 762 children born between 2002 and 2005 to women who are part of the GSOEP. Mothers of these children are interviewed when their children are 24 to 47 months old.
Both surveys cover a broad range of indicators of child development, adjusted to the age of the child. We group these indicators into the following key measures of non-cognitive child development outcomes are available:
• Independence: The GSOEP asks questions if the child eats with a spoon without making a mess, blows her nose without assistance, uses toilet to do number 2, puts on pants and underpants frontwards, brushes her teeth without assistance. The GCP asks a battery of questions how the child manages unexpected or problematic situations.
• Social skills: Questions contained in the GSOEP are if the child calls familiar people by their name, plays games with other children, participates in role-playing games, shows particular liking for certain playmates or friends, calls his/her own feelings by name. The questions contained in the GCP are about the relations the child has to children of his age group, if (s)he likes company, likes to meet kids, shows empathy, starts to fight, argue, bug or provoke other children, etc.
• Behavior: The GSOEP asks the mother to judge about the child’s behavior ranking it on a scale between characteristics such as focused/distracted, obstinate/obedient, quick in learning/need time to understand. The GCP provides similar information on a child’s degree of prudence, concentration, assertiveness, fidgetiness, quickness, etc.
• Personality: Both datasets include judgments about a child’s personality such as how happy, irritable, difficult to console, curious, active, communicative, empathic, shy, etc.
the child is.
• Motor skills The GSOEP contains additionally information on children’s motor skills at age 2 to 3 (e.g. if a child walks forwards down the stairs, uses the door handle to open
doors, climbs jungle gyms, uses scissors to cut paper, paints / draws recognizable forms on paper).
With respect to the cognitive development the following skill categories are asked:
• Language skills: The GSOEP provides information on a child’s communication skills such as whether the child understands brief instructions, forms sentences with at least 2 words, listens attentively to a story for at least 5 minutes or relates simple messages.
• School grades: The GCP informs about a child’s performance in a range of subjects in primary school: math, reading, writing, science, sports, arts and music. Grades are reported by mothers and refer to the final grade her child attained in the most recent grade transcript. Grades are available for children aged 8 and older because younger children receive a verbal assessment of their work. Grades measure academic performance on non-standardized tests which may not be comparable across different school classes.
Mothers assess child development using qualitative scales with four discrete categories: the statement is completely true (1), more or less true (2), less true (3) and last not at all true (4). Employing factor analysis separately on both datasets we create an index across all dif- ferent developmental dimensions (including all variables mentioned above among the categories independence, social skills, behavior and personality) as well as separate indices for each of the different categories of the non-cognitive (independence, social skills, behavior, personality and motor skills) and cognitive (language and school grades) development. All indices are scaled to have mean zero and standard deviation of one.
Interpretation of child development indices is straightforward for the domains that reflect skills or capacities. Higher values signify more developed or better developed skills. Interpre- tation of the indices measuring personality or behavior is not as clear cut. We assign positive scores to personality traits or behaviors which have positive connotation but recognize that this coding is not uncontroversial. We therefore provide not only effects on the aggregate child development index but also on its components.
All indicators are based on mother’s assessment of the child’s development. Self reports could be affected by a range of biases. First, mothers could overestimate their child’s abilities.
This is a bias that plausibly affects all mothers. Second, self-reports could also be affected by child care mode. For instance, mothers who use some formal care may evaluate their child relative to their children’s playgroup at the child care center whereas mothers who do not use child care evaluate their child compared to their own expectations. Yet, note that this bias is
unlikely to be important. Mothers respond to survey questions at home. Survey items target the capacities of the child under consideration and are not formulated in a relative fashion.
Nevertheless, we can not rule out this form of bias ex ante. But if this bias is important, it will affect ”subjective” measures (personality) more strongly than ”objective” (independence, motor skills, behavior) measures. Thus, a pattern of results that finds effects for ”subjective” skills only would be suspect. Third, mothers who use formal care may evaluate their children better in order to justify their absence. This bias is important for the cohort aged 2-3 years but less important for the cohort aged 5-10 years since attendance to early child care occurred back in time and hence any type of reporting bias should have disappeared. We will therefore provide detailed results by type of skill and in the short- and medium-run.
Taken together, this database provides information on a full range of skills characterizing children’s development at different ages (2-10 years old). They can be grouped into cognitive (language skills and school grades) as well as non-cognitive skills (independence, social skills, behavior, personality, as well as motor skills).5 We discuss whether only non-cognitive, cognitive, or both types of skills are affected by a shift from maternal care to formal care in the short- and/or medium-run. We maintain that broad focus because non-cognitive skills have been argued to be particularly important for later life success (Heckman and Krueger, 2004). Our data does not permit discussing the effects of childcare for later labor market success. Yet, note that existing literature suggests that this relationship is strong (Gregg and Machin, 1998).
Moreover, we provide evidence on intermediate outcomes – primary school grades – that are strongly associated with later labor market success (Chetty et al., 2011).
Our main analysis pools data from both sources in order to combine as many age cohorts as possible. Information on several age cohorts is central in order to assess sensitivity of results.
Pooling can, however, also be problematic in case the two dataset are not comparable. This is why we also provide separate estimates by data source.
3.2 Descriptive evidence
This section provides descriptive evidence on children with some exposure to center-based care compared to children with none.6 We also shed light on the underlying channels through which center-based child care might exert its effects. In particular, we discuss if mothers, whose children are not enrolled in formal care, rely on other sources of child care or interact in a significantly different way with their children.
5Schmiade et al. (2008) discuss child development measures in the GSOEP and refer to them as ”adaptive behavior”. For the purpose of comparison with the older cohort we categorize our skills in two groups and refer to them as cognitive and non-cognitive skills.
6See Table A5 for descriptive statistics on the full sample.
Allocation of treated and control children is based on two different types of information concerning child care exposure. The GSOEP contains a module informing on the hours of non-maternal care per week. We allocate children who have positive hours of care in the child care center into the treated group. The GCP contains information on whether or not the child attended child care when he or she was aged 0 to 2 years old. We allocate children with some exposure to early child care into the treated group. All remaining children are allocated to the control group. Note that the treatment is based on different information across the two data sources. The GSOEP provides a snap-shot of care arrangements during the interview week. We may therefore miss children who enter child care after the interview has taken place. In contrast, GCP information is retrospective and captures all children who ever attend child care.
How do children differ depending on their child care use? Table 1 indicates that children who attend some formal child care do better in terms of cognitive skills, such as language skills, as well as non-cognitive skills, such as independence, social skills, and motor skills. Neither are there differences in terms of personality and behavior nor in terms of grades obtained in school.
Overall, there is a 0.12 standard deviation difference in terms of the average skills and abilities for children attending a daycare center compared to children who never attend the daycare center.
Table 1: Child development and Child Care
Table 2 contains information on care. This information is only available in the GSOEP. The data do not contain information on hours of care provided by the mother. Yet weekly hours of non-maternal care along with the weekly hours of maternal market work are available. Note that the mother could also be present during some non-maternal care spells, for instance when the father or any other family member provides care. Results indicate that partners care for their child during 12.6 hours per week. The second most important source of care is formal care provided by a child care center during 10.0 hours per week. Next in line is the extended family.
Grandparents provide about 5.0 hours of care per week, siblings and relatives together provide 1.5 hours per week. Finally, child minders and nannies provide also 1.4 hours of care per week.
Table 2: Non-maternal Child care and Maternal Market Work
How do child care arrangements differ between children who are in child care and those who are not? Children with some formal care spend on average 22.5 hours in the child care center, and receive 1.4 hours less care provided by a child minder. There are no other important differences between children exposed to different modes of child care. Thus, child care predominantly
replaces maternal care, and, to a much lesser extent, informal care. Interestingly, mothers who place their child in child care work almost 13.5 hours per week whereas mothers whose child does not attend formal care work about 6.2 hours per week. This means that the additional formal care purchased by families with formal care do not translate into labor supply one-for- one. There are at least two explanations for this finding. First, mothers may purchase child care to free up time for non-market work or leisure. Second, hours of labor supply provide a strict lower bound on the amount of child care needed to work because hours of work do not count commuting time, etc..7
Child development may differ between children in formal care and informal care because of differences in pre-existing characteristics or because of selection into child care. Table 3 contrasts children by formal care in terms of individual and regional characteristics. Overall, Table 3 indicates that 21% of all children are born in East Germany; this share amounts to 31 % for children with formal care exposure. About 50 % of all children are boys, 9% are underweight at birth (less than 2500 grams at birth) with low birth weight being less prominent among children with exposure to formal care. Moreover, children with formal care exposure are younger and born into smaller families than children with no exposure to formal care. Mothers of children in formal care do not differ in terms of age ( about 30.4 years old) and cohabitation status. But mothers who place their child in formal care are better educated (28 % have a university degree vs 17 % for mothers who do not use formal care). Moreover, household income is significantly higher for children with some formal care (15 % income exceeding 4000 Euro) compared to children with no formal care (10 % with income exceeding 4000 Euro). These results indicate that there is clear selection into formal care based on mother’s education, household income, family size and child birth weight.
Table 3 also provides a detailed set of features of the county (Kreis). The key regional characteristic is the number of child care slots available to children aged 0 to 2 years – the child care offer rate at birth. This rate stands at 10.4 percent (with a standard deviation of 15 percentage points, see Table A5). Note that this rate is much lower than the percentage with some formal care in the sample (35 % in the GCP and 45 % in the GSOEP). This is due to few children using formal child care below the age of 1 and children sharing full-time slots. Table 3 also shows that the change in child care offer rates during the three years prior to birth is positive on average. Note, however, that this development masks an expansion in the West and
7Note that also parental leave policy has been shown to generate small substitution effects. Schoenberg and Ludsteck (2008) who study Germany and Lalive et al. (2010) who study Austria find less than one-for-one effects of expansions of parental leave on maternal labor supply. This is mainly for two reasons. First, an important share of women do not work even long after giving birth to their first child. Second mothers tend to shift between unemployment insurance and parental leave when parental leave is expanded.
a contraction in the East. Regional characteristics also indicate that children with exposure to formal care live in regions with higher unemployment, lower fertility, and a slightly older age structure. Estimates also need to take these regional differences into account.
Table 3: Individual and regional background characteristics
The GSOEP interviews mothers concerning the activities they have undertaken with their child during the last two weeks prior to the survey. We use this information to assess the quality of maternal care. We first grouped activities with respect to whether they stimulate cognitive abilities (singing to the child, reading to the child, painting/handicraft, looking at picture-books with the child), stimulate motor skills (walking with the child, going to the playground), and passive child care (visiting other families, going shopping with the child, watching TV with the child). Children who are in child care get to spend less time with their mothers. Child care should therefore lower the frequency with which mothers undertake these activities with their children. The effect should be particularly strong on the marginal activities, i.e. the activities that mothers do not consider central for their children’s well-being.
Table 4: Activities undertaken by mothers and children
Table 4 displays the proportion of mothers who indicate performing the mentioned activity daily.8 Results indicate that mothers who have their child in child care reduce time in home production or passive child care – mothers with some formal care go shopping less frequently with their child and watch TV less often. However, mothers who use some formal care also sing and read to their children more frequently than mothers who do not use formal care at all.
This fact is puzzling since mothers with formal care spend less time with their children. This finding rather reinforces the fact that mothers who place their child in formal care are positively selected. This spurious effect should therefore disappear once controls for mother education or household income or other variables listed in Table 3 are taken into account.
4 Conceptual Framework and Estimation
This section presents the framework for estimating the causal effect of formal care on child development, introduces the empirical framework, and discusses three main empirical strategies:
ordinary least squares (OLS), linear IV (IV), and reduced form (RF) analysis.
The effect of formal care on a particular child is just the difference in child development
8Information is available on whether the activity is performed daily, several times per week, once per week, less than once per week, or never. Results are similar if these responses are added into an index.
with some exposure to center-based care and child development in the absence of child care.
This individual causal effect may arise for at least two reasons. The first way in which child care can affect child development is direct. Since children aged 0 to 2 years need to be cared for throughout the week, any child care time generates a shift in the structure of early child care. We show above that children are cared for by four basic types of care providers: the mother, the center, the extended family, and providers of informal care. Placing a child in formal care necessarily entails a shift from the three remaining providers of care to center-based care. This shift in type of child care will produce a positive effect on child i if the quality of center-based care dominates other sources of care. This is particularly likely if time in the child care center substitutes for time spent on rather passive activities (see above). The second way in which child care can affect child development is indirect. Placing a child in child care frees up time for market work thus potentially raising household income. Moreover, mothers who place their children in child care have less time at their disposal and will not undertake the same activities with their children. Even though these channels are indirect, they are nevertheless a straightforward consequence of placing child i in center-based care and they are therefore part of the causal effect. Our primary objective is to identify the total effect of child care on child-development. Yet, we discuss the relevance of direct and indirect effects by providing information on how income and activities are affected by child care.
Let yiat measure skill or ability development of child i of age cohort a born in year t. Let Di= 1 if child i is currently enrolled in child care (GSOEP) or spent some time in center-based child care during ages 0 to 2 years (GCP), and Di = 0 otherwise.9 Our main objective is to identify the parameter δ in the following linear model of child development:
yiat= αa+ Xit0β+ Xct0 γ+ δDi+ iat (1)
Our first empirical strategy is ordinary least squares (OLS) taking into account individual level background characteristics (vector Xit), and a rich set of regional characteristics (vector Xct, c indexes German counties or Kreise).10 Note that both sets of characteristics are mea- sured at childbirth. The reason for this is the following. Current individual characteristics are endogenous, i.e. families may strongly change working arrangements after child birth. Thus, current characteristics are ”bad controls” (Angrist and Pischke, 2009). Moreover, current re- gional controls may be a result rather than a predictor of the birth cohorts which child i is a
9Note that by coding formal care in a binary fashion we focus on the extensive margin, i.e. being exposed to any care vs no care at all. While studying the intensive margin is interesting in its own right, we abstract from doing so because we have information about the intensive margin of exposure to center based care in the GCP.
10Regional characteristics include the unemployment rate, female employment rate and regional GDP, fertility, migration, age structure, and habitable area, and the growth in child care slots from 1994 to 2002.
part of. Regional characteristics are particularly important since German regions differ strongly not only in terms of offer rates of formal care, but as well in terms of their socio-economic development and demographic structure.
OLS compares children who are or have been exposed to center-based care early in their childhood to children who have never been in center-based care. OLS provides a weighted average of the underlying heterogeneous treatment effects. The weights are related to the variance of being in treatment (Angrist, 1998). This effect is similar to the average effect of treatment on the treated with the exception of the weights. To calculate the average effect of treatment on the treated one uses the probability of treatment rather than its variance.
Note, however, that OLS may also be affected by selection bias. First, families may base their decisions to purchase formal child care for their children on unobserved child ability. Second, these families may also differ in terms of other aspects that are correlated with children’s skills - for instance, in the way they interact with their children, as we have seen in Section 3. Yet, our analysis controls for a wide range of observed characteristics at both the individual and the regional level. Moreover, we assess the quality of the control variables by assessing whether they allow us to achieve balance in terms of the activities that mothers undertake with their children.
Descriptive evidence shows that mothers with a child in child care more frequently read and sing with their children. This difference disappears once we condition on controls. We therefore expect the bias due to unobserved characteristics not to be quantitatively important.
The second empirical strategy relies on variation in the amount of child care slots available to parents. The supply of child care is clearly endogenous to local efforts to promote maternal labor supply or to guarantee a well-functioning educational system. In other words, unconditional exclusion surely fails. We therefore condition on any regional characteristics on which the regional authority base their projections for the local demand for child care. Important predictors for the regional demand for child care are socio-economic conditions such as the regional female employment rate, the regional unemployment rate and the regional GDP, but also demographic characteristics such as the regional fertility rate, regional net migration and the regional age structure. Moreover, we consider the region-specific growth in child care slots prior to a child’s birth to approximate the general regional trend in opening child care centers. As a result, our instrumental variable strategy relies on the remaining unpredicted variation in the supply of child care slots.
Our first stage looks thus as follows
Di= πa,0+ Xit0π1+ Xct0 π2+ π3Zct+ νiat (2)
where Di is the binary variable indicating individual child care usage, Xit stands for the set of individual level background characteristics and Xct for the set of regional characteristics described above. Zct indicates the offer rate of slots in child care in the county c and year t when a child is born. Importantly, we measure the number of slots potentially available per 100 children rather than actual slots filled – a measure of the supply of child care slots. Moreover, we use the local supply of formal care as well as all regional characteristics at the time of childbirth, hence prior to parent’s actual decision to send their child to formal care in order to rule out any mechanical relationship between individual child care attendance and regional share of children in child care.
One may still doubt whether the remaining unpredicted variation in child care supply can be excluded from the model explaining child development. Exclusion may fail for two reasons.
First, regional differences in terms of the demand for child care may not only be based on observable aspects such as the local labor market or fertility decisions, but also on potentially unobservable aspects such as a mother’s attitude towards raising their children or educational systems. Second, mothers might also migrate to regions with more favorable supply of child care.
To address the latter claim, we assess whether families’ migration decision in the years around childbirth can be predicted by differences in the supply of slots in child care centers (Table A2).
Our findings suggest that families do not move to regions with better supply of child care.
Regarding the first threat to identification, we provide the following sensitivity checks. First, we assess this claim by relating the regional child care offer rate to the set of individual charac- teristics (Table A3, column 1). The child care offer rate is systematically related with important predictors of child development such as mother’s education and household income. Yet, once we include regional characteristics, none of the individual characteristics is systematically related with the child care offer rate (Table A3, column 2). This analysis shows that local child care offer rates are unrelated to observed characteristics of children, mothers, and their households – evi- dence that strengthens the case for local child care offer rates also being unrelated to unobserved characteristics.
In addition, we provide supplementary analysis that discusses the sensitivity of our baseline results to unobserved region effects. We first add a dummy that identifies former East German states. This strategy identifies within regional variation in offer rates. This is useful since East versus West is the most important regional divide in Germany. We also add state fixed effects for all Western states in addition to the East dummy. This is to account for the strong regional variation within West Germany in terms of education systems and regulations of child
care centers. We then add a separate time trend for each region. Taken together, these three strategies allow assessing whether our baseline strategy confounds child care effects with cultural differences, formal education systems, or time invariant differences in local demand for child care.
Thus, we are confident that we discuss whether the child care offer rate can be excluded in terms of omitted regional level and trend variables.
We use a standard two stage least squares approach and report the resulting IV estimates for formal care. This approach has also been termed ”linear IV” by Heckman and Vytlacil (2000). Linear IV estimate provides a weighted average of the marginal treatment effect (MTE) parameter. In the present context, the MTE parameter is the causal effect of center-based care for children who get access to care because the supply of child care slots marginally increases from a level p to p0. With a continuous instrument Zct, as in our context, linear IV measures a weighted average of the underlying heterogeneous MTEs for children with different propensities to attend center-based care. The weights are, again, related to the conditional variance of being in the treatment.
The two empirical strategies target different parameters: the effect of child care on children who are in child care (OLS) and the effect of child care on children who enter child care because the supply of child care is expanded (IV). How are these two effects related? This depends strongly on the selection mechanism. Suppose children of parents with high education get enrolled first – i.e. there is selection into child care based on parents’ intellectual background (see Figure 2). This selection mechanism is plausible given that centers use waiting lists placing children earlier whose parents have expressed a strong interest in a slot earlier. Children of mothers with high education would benefit more from spending time with their mothers rather than in child care. Thus, the gains from formal child care are smallest or even negative for children of skilled mothers. In contrast, children of mothers from a low educational background could benefit a lot from joining the child care center (low skilled mother). Consequently, in a situation with a low level of child care offered, the average effect of child care is small and potentially negative since most children in child care have highly skilled mothers. In contrast, the children entering child care when supply of child care is expanded are from a low skilled educational background. The marginal child will therefore exhibit a much more positive effect of being in child care. Since the linear IV estimate is a weighted average of MTEs, we expect the IV estimate to be larger than the OLS estimate under this selection mechanism. Moreover, the MTE should be especially strong for children who enter child care when the supply of child care is relatively high because the additional slots are going to children whose parents are from low socio-economic backgrounds. We therefore provide estimates of the MTE at low and high
Figure 2: A Plausible Selection Mechanism
levels of the offer rate.
Finally, we also provide estimates that relate child development to child care offer rates directly as follows
yiat= αa+ Xit0 β+ Xct0 γ+ νZct+ µiat (3)
The parameter ν captures the so-called reduced form (RF), i.e. the effect of increasing the child care offer rate by one unit. The parameter ν is the relevant policy parameter and could, for instance, be used for a cost-benefit analysis of the extension of early child care.
5 Results
This section discusses the effect of exposure to formal care using ordinary least squares (OLS) measuring the effect on the average child placed in child care, and using an instrumental variable method (IV) measuring the effect of child care on children that are in child care because the supply of child care is expanded. The section first presents the main result, then discusses sensi- tivity to region fixed effects and time trends, argues how to rationalize the main result, presents the short- and medium-run effects, and concludes with evidence on the likely mechanisms.
Main Result: Table 5 shows the main results regarding the aggregate non-cognitive skill factor for the sample pooling children at all ages (2-10 years old). This indicator reflects the mean over all the skills and abilities discussed in section 3.1, in particular independence, social skills, behavior and personality. All estimates are based on cluster robust standard errors that allow for clustering at the county level.11
Table 5: Baseline Results
We first provide estimates of the effect of formal care on children who are in formal care (Table 5 Column 1). Children with some exposure to formal care do on average better in terms of non-cognitive skills than children who are not taken care of by a center. The difference in child development is on the order of 0.105 standard deviations. How does this effect compare to other predictors of child development? All regressions contain a comprehensive set of regional
11Note that the data contain repeated observations for children in the German Child Panel. Pooled analysis of panel data requires that standard errors are clustered at the panel unit. Note, however, that clustering at the county level implies clustering at the child level since hardly any child leaves the county of residence between two survey waves.
and individual characteristics (see Table A1 for coefficient estimates). Regional background characteristics are not significantly related to the child development score (neither individually, nor jointly). In contrast, individual characteristics such as birth weight and gender predict child development in a statistically significant and important fashion. Children who are underweight at birth (birth weight being lower than 2500 grams) exhibit an overall measure of child de- velopment that is 0.27 standard deviations below that of children with normal birth weight.
This disadvantaged skill development for low birth weight children corresponds to the difference in scores between boys and girls (boys have 0.32 standard deviations lower scores than girls).
Moreover, household income is very strongly related to child development. Children belonging to lower income groups fare worse in terms of the child development measures we study. Mother’s education is positively but insignificantly related to child development. Neither the number of siblings nor the birth cohort are related to child development.
Does expanded offer of child care slots affect the probability that children can be placed in a child care center? Column 2 in Table 5 discusses the effect of the supply of child care slots on the probability that a child is in formal care. This is the first stage result equation in the IV analysis. Results indicate that increasing the child care offer rate by one percentage point increases the probability that a child is enrolled in child care by 0.9 percentage points, almost a one-for-one relationship. Note that the t-statistic on the child care offer rate is 5.3 (implying an F-statistic of 27.9) which is above the rule of thumb threshold for weak instruments. Thus, we do not face a problem of a weak instrument. Moreover, mother’s education and household income are very important in shaping the probability that children attend child care. A mother with secondary school degree is 18 percentage points more likely to send her child to child care compared to a mother without a degree; the corresponding effect is 29 percentage points for mothers with university degrees. This finding strengthens the claim that rationing favors children from advantaged backgrounds. Highly educated mothers may place their children early enough on the waiting list in order to place their children in formal care despite excess demand.
Hence, children with the lowest gains from formal care are placed first. Households in the income brackets 0 to 2250 Euros per month are significantly less likely to place their children in formal care. The remaining individual characteristics are not statistically significant.
Column 3 of Table 5 shows the effect of an increase in the available slots per 100 children in the county where a child is born; an increase in 1 slot per 100 children increases the overall non-cognitive skills by 0.01 standard deviations. This effect is non-negligible; in case a county would change from no child care facilities at all to full coverage (meaning a change from 0 to 100 slots per 100 children), overall skills would increase by 1.04 standard deviations. This reduced
form effect suggests that child care affects child development. More children attend child care if the supply of child care slots is extended; if children reap positive benefits, child development scores increase with supply of child care slots. However, the reduced form effect might also confound the effect of formal care with other unobserved county characteristics. We discuss alternative ways of accounting for these unobserved characteristics below in Table 6.
Turning to the IV estimates, Table 5 Column 4 provides the estimates of the effect of formal care exposure using unpredicted variations in local care supply as an instrument. Results in- dicate that enrollment into formal care following an increase in the local supply improves child development by 1.12 standard deviations. The estimate is statistically significant at any conven- tional level. Children who are enrolled in formal care once access is less restricted have by far larger gains from exposure to formal care than the children enrolled regardless the shortage of supply. As already discussed in Section 4, rationing may be based on advantaged backgrounds or put it differently on gains from exposure to high-quality non-parental care, with the children with the lowest gains being the ones enrolled first. We discuss this issue further below.
Sensitivity analysis: Regional differences in parenting styles, labor markets, etc. are poten- tially the key threat to the IV identification strategy. Yet note first that regional characteristics are not important predictors of child development (see Table A1 with full results). Moreover, results are quite robust to controlling for unobserved differences between regions and states.
Panel A in Table 6 reports the baseline result from Table 5. Panel B in Table 6 shows the results for the different estimations – OLS, RF, First Stage and IV – controlling for the set of individual and regional characteristics as well as for a dummy for residence in East Germany.
Results indicate that the effect of child care on children in child care (OLS) is positive but not significantly different from zero. Yet the effect of child care on the extra child (IV) remains pos- itive and statistically different from zero (despite a strong increase in the standard error). Once we control additionally for the states in West Germany (Panel C in Table 6), the point estimate of the IV estimation decreases, but is still significantly different from zero. Panel D in Table 6 presents results that allow additionally for separate linear time trends in all West German states plus East Germany. Results are, again, robust to the inclusion of separate time trends.
These results suggest that the baseline analysis is quite robust to allowing for endogeneity to time invariant state and county characteristics, or slowly moving state characteristics.
Table 6: Sensitivity analysis controlling for regional fixed effects
What explains the difference between OLS and IV? We have argued above that ra- tioning tends to favor mothers with extensive educational background. In other words, the differences between OLS and IV might be explained by differences in the quality of the alterna- tive care mode - here motherly care - available to the average child enrolled in child care and the marginal child attending child care once access is less restricted. Below we shed some light on this hypothesis by distinguishing between high and low educated mothers, East and West Germany – two regions characterized by very high and very low offer rates – and finally regions differing in the offer rate of child care.
Table 7 provides results by mother’s education and region of residence (East versus West).
This contrast allows assessing whether selection into child care differs by mother’s educational background. Moreover, the average effects of child care are also expected to differ since highly educated mothers offer child care that is similar or even dominates the quality of care provided by the center, while mothers with low education might care for their children in ways that are dominated by the center. Panel A in Table 7 provides results for children whose mother has completed primary or secondary education (OLS). The effect of formal care is positive and significant for the average child in formal care (OLS). Expanding the child care system improves child development significantly and quantitatively importantly (RF). This is because more children attend formal care as the number of slots in formal care is increased (First Stage).
Overall, the marginal child of a mother with low to intermediate educational background benefits strongly from attending child care (IV). The effect of attending child care is on the order of 1.13 standard deviations.
In contrast, children of highly educated mothers do not benefit from being placed in formal care (OLS Panel B in Table 7). Moreover, increasing the number of slots in center-based care does not lead to higher take up of child care among children of mothers with completed university education. Thus, the child care offer rate is a weak instrument for this specific subgroup and we therefore do not calculate IV estimates for children of mothers with a university degree. Note that the child care offer rate is not correlated with child development (RF). This suggests that child care offer rates affect child development via child care rather than reflecting a spurious correlation between unmeasured determinants of child development and child care offer rates.
The OLS results suggest that child development gains from child care differ by educational background of the mother. Moreover, the first stage results demonstrate that the marginal child entering childcare because supply is expanded is from a low educational background with corresponding large gains from being placed into child care.12
12Note, however, that the difference between the effects of child care in the IV and OLS estimates is larger than the difference of the OLS effects of child care for children of parents with low educational backgrounds. This
Table 7: Results by Motherly Education and Region
We also provide results stratified by region of residence (West vs East). Due to the absence of rationing in East Germany, the effect for the average child placed in child care should be higher in East Germany than in West Germany but the child care offer rate should not work as an IV in East Germany. Indeed, separate OLS regressions of the overall skill indicator on formal care usage indicate positive but insignificant point estimates for both East and West German children (Panels C and D in Table 7). Interestingly, increasing the offer rate by two slot per 100 children leads to almost one additional child getting access to formal care (First Stage) in the West. Consistent with this pattern, the reduced form evidence shows that child development improves in the West as the number of slots per 100 children increases. Overall, the effect of child care on the marginal child is significantly different from zero and on the order of more than two standard deviations (IV). In contrast, results for East Germany indicate that expanding the offer of child care does not lead to incremental take up of child care in East Germany (First stage Panel D). This is consistent with East Germany being characterized by excess supply of child care slots; we therefore do not report IV estimates for the East German sub-sample.
Moreover, consistent with the absence of the first stage effect, the RF results indicate that child development is not correlated with the child care offer rate.
Do the extra gains from child care differ between regions that offer lots of slots compared to regions offering fewer slots? If rationing into child care favors children from an advantageous background (with low positive or even negative effects) we should observe increasing returns for the extra child in regions with lots of slots compared to regions where slots are limited.
Table 8 explores estimation of the MTE using local IV. Local IV is IV estimation on data that is weighted to reflect the distance to a particular level of the instrument. For instance, the marginal treatment effect at a child care offer rate of 10 % is estimated using standard IV on data that puts most weight on observations with a 10 % offer rate and less weight on data that is far from 10 %. We use a normal kernel to weight the data. Bandwidths are chosen to reflect the left-skewed distribution of the offer rate, i.e. small bandwidth for low values of the offer rate, and large bandwidths for observations with a high offer rate.
Table 8: Local IV estimates
Table 8 shows estimates of the MTE – the effect of child care induced by increasing the number of slots by 1 in 100 – at child care offer rates of 3% (median), 5% (65th percentile), 10%
suggests that selection into child care is not only related to observed parental education but also to unobserved indicators that are correlated with child development.
(75th percentile), and 20% (80th percentile).13 Results indicate that the MTE varies with the level of child care. The MTE is not significantly different from zero in regions that offer child care to 3 % of all children, and the effect is positive and still insignificant at an offer rate of 5%. In contrast, the marginal treatment effect is strong and significantly different from zero in locations that expand child care from a level of 10 %, or 20 %. These results are consistent with rationing of child care favoring children who benefit from high quality home care.
Taken together, these results are consistent with rationing taking place based on family background. Children with a good child development potential are selected into child care in a situation with few slots available. Children with higher potential gains from center-based care enter child care as child care is expanded. This generates a situation with a weak average effect of child care on child development and a strongly positive effect on the marginal child. These findings indicate that high-quality care might contribute to decrease inequalities arising due to different socio-economic backgrounds and to weaken intergenerational transmissions of socio- economic status. In other words, child care at early childhood can help to ”level the playing field” (Havnes and Mogstad, 2010).
Detailed Results: How does formal care affect the different dimensions of child development?
Table 9 provides detailed results regarding skill dimensions. The results in column 1 indicate that children with some exposure to formal care benefit in terms of independence and social skills. There are no differences in terms of personality and behavioral development. Turning to column 2, the RF estimates, we can see that the ceteris paribus effect of offering 1 more slot in formal care per 100 children favors children in all dimensions, except the behavioral dimension:
in case a county would change from no child care facilities at all to full coverage (meaning a change from 0 to 100 slots per 100 children), the observed effects are strong, ranging form 0.66 standard deviations for social skills to 1.04 standard deviations in children’s independence.
Similarly, the effect of child care on the additional children entering care because child care offer rates are expanded are positive (column 3). There are improvements in children’s independence (1.12 standard deviations), social skills (0.71 standard deviations) and personality (0.67 standard deviations). The point estimate on behavior is positive but not significantly different from zero.
Table 9: Detailed Results
13We have also explored estimating the marginal treatment effect at an offer rate of 1% (25th percentile) and an offer rate of 45 % (95thpercentile). The first stage indicates that the offer rate is not an instrument for formal care at these percentiles. The lack of power for the 45% offer rate is consistent with our findings in Table 7 for East Germany. The lack of power for the 1% offer rate is perhaps due to slow take-up of child care in regions that do not have child care at all.
