Does Universal Child Care Matter?
Evidence from a Large Expansion in Pre-School
Education
Christian Dustmann
Anna Raute
Uta Schönberg
This Draft: October 22nd, 2012
Abstract
This paper studies the effect of a German universal child care program (aimed at 3- to 6-year-olds)
on the school readiness outcomes. We draw on unique administrative data for the entire population of
children who are about to start school in one large region. Our identification strategy exploits a federal
policy reform which entitles every child to a child care slot from her third birthday to school entry,
leading to a staggered expansion in child care facilities across municipalities. We find that public child
care attendance strongly and robustly reduces language and motor skill problems and improves overall
school readiness for children of immigrant ancestry (i.e., children who themselves or whose parents were
born outside Germany), but has no significant effects for children of native ancestry. We offer three
explanations for these findings. First, non-linear returns to child care attendance may lead to higher
overall effects for children of immigrant ancestry, as the reform increased attendance rates for these
children at a different margin than for children of native ancestry. Second, the two groups of children
differ with respect to the counterfactual care arrangements. Third, due to their lack of proficiency in
the host country language, children of immigrant ancestry may benefit more from public child care and
exposure to other children. Our findings suggest that universal child care programs help to narrow
the achievement gap between children of immigrant and native ancestry, potentially reducing life-time
inequalities between these two groups.
Correspondence: Christian Dustmann, Department of Economics, University College London, Email:
[email protected]; Anna Raute, Department of Economics, University College London, Email:
[email protected]; Uta Schnberg, Department of Economics, University College London and Institute
for Employment Research; Email: [email protected]. We are indebted to Dr. Johannes Dreesman
and the Lower Saxonian Health Ministry (NLGA) for providing us with the data and their valuable
data support. We thank the Norface programme on migration for financial support.
1
Introduction
Pre-school and early childhood programmes are often thought to be an important and effective way to
influence the development of children (see e.g. Ruhm and Waldfogel, 2011 and Almond and Currie, 2011
for excellent recent surveys). Not only is learning in early childhood particularly productive because of
the long time to reap the rewards (Becker, 1964) and because many skills are best learnt when young
(e.g. Shonkoff and Phillips, 2000), but there may also be important ”dynamic complementarities”
of early learning with acquisition of human capital at later stages (Heckman, 2007 and Cunha and
Heckman, 2007). To assess the benefits of early childhood programs, a large literature has focused on
early intervention programs, such as Head Start or the Perry Preschool Project, that target specifically
children from disadvantaged backgrounds and combine a variety of different services (see e.g. Blau
and Currie, 2006, for a survey). This literature finds strong positive effects of these programs in the
short-run, and conflicting evidence in the longer run.1
Less is known however about the benefits of universal child care programs targeted at all children.
In this paper, we study the effects of such a program that is aimed at 3- to 6-year-olds for the case
of Germany, where the quality of care is high, fairly homogenous and almost exclusively public. In
contrast to the typical early intervention program, where learning is primarily academically oriented,
learning in German public child care is mostly informal and play-oriented. A unique feature of the
German school system is that all children have to undergo an extensive battery of school readiness
tests, administered by pediatricians, before entering elementary school. This provides us with rare and
detailed administrative data on a series of school readiness indicators, such as fine or gross motor skill
problems or language deficits, for the entire population of children who are about to start school in one
large region. These indicators are, as recent research shows (Grissmer et al., 2010), stronger predictors
1
According to The Head Start Impact Study, Head Start improved cognitive and non-cognitive achievement measures
of its participants in the short-term which however quickly dissipate. Garces, Thomas and Currie (2002), however, find
that participation in Head Start increases high school completion and college enrolment among whites. Deming (2009)
finds large long-term gains of Head Start participation for disadvantaged children despite the fading-out of test score
gains. Using a regression discontinuity design, Carneiro and Ginja (2012) show that participation in Head Start reduces
the incidence of behavioral problems, serious health issues and obesity for adolescents and criminal activity among young
adults. Re-evaluating the Perry Preschool Project, Heckman, Moon, Pinto, Savelyev and Yavitz (2010a,b) find strong
positive effects of the program that last well into adulthood.
2
for later academic success than early cognitive test scores, which are typically used in the literature (see
Section 2 for a review). An important additional advantage of our data is that our outcome variables
are standardized assessments by health professionals as opposed to subjective assessments by parents
commonly used in the literature.
To deal with the non-random sorting of children into public child care, we make use of a federal
policy reform which entitles every child to a largely subsidized 4-hour child care slot from her third
birthday to school entry. In the region we study, this mandate lead to a staggered timing and intensity
of the construction of child care facilities across municipalities—which is the variation we exploit in our
instrumental variable estimation.
An important contribution of our work is to contrast the effects of time spent in public child care
for two groups of children: children of native ancestry, and children of immigrant ancestry, defined
as children who themselves or whose parents were born outside Germany.2 Children of immigrant
ancestry are more likely to come from a disadvantaged background than children of native ancestry,
perform significantly worse at school than children of native ancestry (see e.g., Dustmann et al., 2012)
and often do not speak German at home.3 We find very different effects for these two groups of children:
Whereas public child care attendance strongly and robustly reduces language and motor skill problems
and generally improves school readiness for immigrant children, it has no significant effects on school
readiness outcomes for native children.
We offer three explanations for these differential effects. First, they might be driven by non-linear
returns to public child care attendance with respect to age or length of attendance. Prior to the
expansion in child care facilities, the typical child of native ancestry entered public child care at the age
of 4 and thus attended public child care for two years before starting school at the age of 6. For these
children, the expansion in child care facilities therefore primarily increased the share of children who
enrol in public child care at the age of 3 and thus attend public child care for three years. In contrast,
2
For simplicity, we also refer to the first group as ”native children” and to the second group as ”immigrant children”.
See e.g. Casey and Dustmann (2008) who report significant language deficiencies for the largest traditional immigrant
communities in Germany (i.e., Turkey, Spain, Italy, former Yugoslavia and Greece).
3
3
prior to the expansion in child care facilities, many children of immigrant ancestry attended public child
care for only one year or not at all. For these children, the expansion in child care facilities therefore
increased both the 1-, 2-, and 3-year attendance rate. At the same time, we find some suggestive
evidence that for immigrant children the first and second year in public child care (when children are
between 4 or 5 years old) improves proficiency in the host country language and overall school readiness
by more than the third year in public child care (when children are 3 years old).
Second, the larger benefits of public child care attendance for immigrant than native children may
be explained by differences in the counterfactual child care arrangement. Our findings suggest that for
children of immigrant ancestry public child care attendance primarily crowds out maternal care while
for children of native ancestry it primarily replaces alternative child care like that by a child minder—
which is likely to be more similar to public child care than maternal care, in particular with respect to
the exposure to other children.
Third, immigrant children are, due to a lack of proficiency in the host country language, likely to
be particularly perceptive to the informal and play-oriented learning environment which public child
care provides, especially if the counterfactual care arrangement is the home environment. Host country
language proficiency has been shown to be an important determinant of success not only in elementary
school (see e.g., Dustmann, Machin, and Schönberg, 2010, Bleakley and Chin, 2008) but also for other
aspects of life (see e.g., Bleakley and Chin 2010 on social assimilation). The exposure to other children
in public child care is likely to quickly reduce this deficit. As host country language proficiency may
also be complementary to learning in general, it might lead to positive spillover effects, thus explaining
the positive effects of child care attendance on other measures of early achievement.
Our findings thus emphasize that the benefits of large scale, publicly provided child care programs
are far from uniform. In our study, they are concentrated among a specific group of disadvantaged
children, suggesting that universal child care programs are successful at narrowing the achievement
gap between children of native and immigrant ancestry at school entry (and possibly beyond), and
hence may help children of immigrant ancestry to assimilate into the host society. Our findings further
4
highlight that the welfare effects of universal child care programs crucially depend on which children
the program draws into child care, which type of child care arrangement public child care replaces, and
which margins of the child care attendance distribution (i.e., the one-, two-, or three-year attendance
rates) are affected.
The structure of the paper is as follows. In the next section, we briefly discuss the existing literature,
and how our paper contributes to it. In Section 3.1, we illustrate the possible channels through which
changes in child care supply may affect child outcomes. We then outline the German system of public
child care and the child care reform that we analyze (Sections 3.2 and 3.3). In Section 4, we describe our
data and provide a descriptive analysis on the differences in outcomes between children of immigrant and
native ancestry and the potential selection into child care. We then discuss our identification strategy
and investigate in detail the validity of the assumptions we impose (Section 5). We report results in
Section 6 and conclude with a brief discussion of our results in Section 7.
2
Contribution to Existing Literature
Our paper contributes to the small but growing literature on the evaluation of large scale, publicly
provided universal child care programs (see Baker, 2011 for a comprehensive survey).4 Much of this
literature has found that such programs have positive effects on children. For instance, Berlinski, Galiani
and Gertler (2009) find positive effects of an expansion in pre-elementary education in Argentina on
performance in elementary school. Similarly, Gormley and Gayer (2005) show a positive effect of a
prekindergarten-program for 4—5 year-olds in the U.S. on cognitive, motor and language skills measured
shortly after the program. There is also some evidence on longer term effects: Havnes and Mogstad
(2011a) find that an expansion of subsidized child care for 3—6 year-olds in Norway had strong positive
4
Related research evaluates the effects of attending center-based care that is typically privately provided, with a large
variance in quality. For instance, Loeb, Bridges, Bassok, Fuller, and Rumberger (2007) and Magnuson, Lahaie, and
Waldfogel (2007) document a positive association between attending center-based care and short-run cognitive test scores,
but a negative assocation between attending center-based care and bahavioral outcomes. To deal with the endogenous
selection into center-based care, Bernal and Keane (2011) exploit exogenous variation in welfare policy rules, and find—
contrary to the observational studies described above—that an additional year in child care reduces child test scores.
5
effects on long-run adult outcomes, such as educational attainment and labor market participation.
Felfe, Nollenberger and Rodrı́guez-Planas (2012) analyze the introduction of high-quality universal child
care for 3 year-olds across Spanish states and find improvements in children’s cognitive development at
the age of 15. Berlinski, Galiani and Manacorda (2008) also report positive effects of a rapid expansion
of pre-school child care facilities in Uruguay on completed education, while Cascio (2009a), exploiting
variation in the introduction of Kindergarten programs for 5-year-olds into public schools across different
states in the U.S., finds a decrease in high school drop-out rates and institutionalization rates as adults
for white, but not for black children. Evidence of positive effects is not unequivocal, however: Baker,
Gruber and Milligan (2008) study the introduction of highly-subsidized universal child care in Quebec
and find negative short-run effects on behavioral and health outcomes, while Datta Gupta and Simonsen
(2010) show that enrollment in pre-school in Denmark does not improve child outcomes at the age of
seven.
We add to this literature by focusing on a particularly interesting group of children for which the
benefits of public child care may be high: children of immigrant ancestry who themselves or whose
parents were born outside Germany. While several studies analyze the differential effects of universal
or targeted child care programs for disadvantaged and advantaged children or across racial and ethnic
groups5 , children of immigrant ancestry—who often do not speak the language of the host country at
home—have received relatively little attention, despite their growing share in the population.6
We further contribute to the literature by investigating possible non-linearities of the effects of child
care attendance with respect to years or age of attendance. The few studies that have investigated this
issue have reached conflicting conclusions. While Berlinski, Galiani and Manocorda (2008) find that a
5
For instance, Havnes and Mogstad (2011a) find larger long-term benefits of public child care attendance for children
of low-educated parents. Gormley and Gayer (2005) show that Hispanics experienced the biggest effects on cognitive,
motor skills and language scores followed by blacks, whereas the effects for whites were mostly small and statistically
insignificant. Evaluating an intervention program targeted at disadvantaged children (Head Start), Currie and Thomas
(1995, 1999) show that Hispanic and white children benefit more from the program with respect to schooling outcomes
than black children.
6
Studies which specifically report findings for immigrant children include Magnuson, Lahaie and Waldfogel (2006) who
show that participation in head start is associated with higher gains for children of low educated immigrant mothers than
for children of low-educated non-immigrant mothers, and Gayer (2008) who finds that a pre-kindergarten program in the
U.S. improved pre-reading, pre-writing, and pre-math skills in particular for Hispanic children whose parents speak Spanish
at home or whose parents were born in Mexico. Drange and Telle (2010) report that the introduction of a free pre-school
program for five-year-olds in Oslo improved test scores of immigrant girls.
6
preschool program in Uruguay positively affects school attainment primarily at the extensive margin,
with small and statistically insignificant effects at the intensive margin, Behrman, Cheng and Todd
(2004) report that a Bolivian Preschool program boosted child outcomes only for children with at least
seven months of exposure, and find increasing marginal impacts with longer program attendance.7 We
find suggestive evidence that for children of immigrant ancestry, the first and second year in child care
reduces the language deficits and deferment by more than the third year in child care.
Finally, one channel through which an expansion in child care facilities may affect children is through
a change in maternal labor supply. Most studies, including Baker, Gruber and Milligan (2008) for
Canada, Havnes and Mogstad (2011b) for Norway and Berlinski and Galiani (2007) for Argentina, find
that increases in the supply of child care places have much smaller effects on maternal labor supply
than on child care utilization.8 In line with this literature, we show that the expansion in child care
facilities had little effect on the labor supply of mothers of both native and immigrant children. Hence,
in our setting, the effect of child care attendance on child outcomes through an increase in maternal
labor supply is small for both groups of children.
3
Background
3.1
Child Care and Child Development: Channels
The following discussion illustrates the channels through which public child care attendance may affect
child outcomes. For simplicity, we distinguish two age periods only; period 1 corresponds to the years
when the child is 3 or 4 years old, while period 2 corresponds to years when the child is 5 or 6 years
7
Loeb, Bridges, Bassok, Fuller and Rumberger (2007) find the greatest association of center-based care with reading
and maths scores for children who start at age 2 or 3 rather than at younger or older ages and negative effects on behavior
to be greater the younger the starting age.
8
In a similar vein, Gathmann and Saß (2012) find that a subsidy given to parents who do not enrol their child in formal
child care lead to a much larger reduction in child care attendance rates than in maternal employment rates. González
(2012), in contrast, finds that a universal child benefit reform lead to an increase in maternal time at home as well as a
decrease in day-care utilization during the first year after birth. Gelbach (2002), using age of school entry cut-off rules, finds
sizeable effects of child care availability on the labor supply of married women, and particularly for single women without
younger children - a result that is similar to Cascio (2009b) who finds that the introduction of Kindergarten programs
for 5-year-olds lead to sizeable employment responses for single mothers. See Blau and Currie (2006) for a comprehensive
survey.
7
old. In each period, parents can make four types of investment into their child: time investments by the
parents themselves, time investments by alternative care givers such as child minders or grandparents,
time investments in public child care, and monetary investments such as toys or books.
Suppose that initially, public child care is rationed (as it was the case in Germany in the early
1990s). Parents optimally choose investments by maximizing a combination of their own and their
child’s utility (which is a function of the child’s achievement), subject to a time and budget constraint,
as well as subject to the slot constraint in public child care. Now suppose that the government expands
public child care for children in period 2 (i.e., when children are 5 or 6 years old). Consider a child
who was not offered a public child care slot in period 2 in the absence of the expansion, but now is.
Panel A of Figure 1 illustrates the channels through which child care attendance in period 2 may affect
child development at the end of period 2. First, public child care attendance may directly reduce time
investments by the mother as well as by alternative care givers.9 Public child care attendance may
also lower monetary investments if public child care attendance is costly. Second, child care attendance
may increase maternal labor supply, as it frees up some of the mother’s time. This in turn may further
decrease time investments by the mother, but may increase time investments by alternative care givers as
well as monetary investments through an increase in household income. These arguments highlight that
the impact of public child care attendance on child development depends on the extent to which public
child care replaces care by parents as opposed to care by others, and whether child care attendance is
accompanied by an increase in the mother’s labor supply and income. To ease the interpretation of our
estimates, we shed some light on these issues in Section 6.6.
Next, suppose that the government expands public child care also for children in period 1 (i.e.,
when children are between 3 and 4 years old). Consider a child who was not able to attend public
child care in period 1 in the absence of the expansion, but now is able to do so. Panel B of Figure 1
illustrates the dynamic effects which child care attendance in period 1 may have on child development
9
Here, it is important to emphasize that one additional hour in public child care does not necessarily reduce time
investments by the mother or by alternative care givers by the same amount. For instance, the mother may use the time
the child spends in child care to do much of the household work and hence not reduce the active time she plays with the
child; see e.g. Felfe and Lalive (2011) for empirical evidence along these lines.
8
at the end of period 2. First, child care attendance in period 1 will affect the child’s achievement at
the end of period 1—which, as discussed above, depends on the relative marginal returns of the different
types of investments, and the extent to which public child care attendance crowds out or complements
these investments. Note that the effect of child care attendance in period 1 on child achievement in
that period may differ from that in period 2, most notably because the relative marginal returns of the
different types of investments may differ by the child’s age when she starts child care. In particular, the
psychology literature emphasizes that child care attendance may be more beneficial for older children, as
for young children the exposure to multiple care givers may lead to behavioral anomalies and difficulties
in social relations with others (e.g. Belsky, 1988; Belsky and Eggeben, 1991; Bowlby, 1969).
In addition, investments in public child care in period 1 may affect the marginal returns to investments in period 2, leading to dynamic effects. These dynamic effects are in principle ambiguous. On
the one hand, as emphasized for instance by Cunha and Heckman (2007), a higher stock of skills at
the end of period 1 may increase the marginal return to investments in period 2, which in turn may
induce parents to increase their investments in period 2. Such dynamic complementarities in the form
of ”skills beget skills” imply that investments in public child care are more productive in period 1 than
in period 2. On the other hand, the return to investments in public child care may be decreasing in
the number of years the child has already spent in child care, so that the third and fourth year in child
care is less productive than the first or second year in child care. Since child care is an absorbing state
(i.e., once the child enters child care, she typically stays in child care until entering elementary school),
it is impossible to disentangle such dynamic effects from heterogeneous effects of child care attendance
by age. We discuss these dynamic or heterogeneous effects in Section 6.5.
3.2
Child Care Provision in Germany
In Germany, formal child care provision of children aged between 3 and 6 is mostly public.10 Child care
provision is further characterized by strict country-wide quality standards: in all child care institutions,
10
For instance, in 2007 only 21 out of the 4264 institutions were private in the region we consider.
9
the teacher-child ratio must not exceed 2 teachers for 25 children. Moreover, teachers must have
completed a three-year vocational programme, and are certified by the state. Further regulations exist
regarding the space provided by the institution for each child. As a consequence, the quality of child
care is relatively homogeneous.
The majority of children in Germany (i.e., more than 90%11 ) attend child care part-time, for 4
hours in the morning. In terms of educational content, Germany follows the social pedagogy tradition
according to which learning is mostly informal and play-oriented. While children engage in specific
activities depending on their age, such as circle time, painting, messy play, and physical education
classes, the development of social, language and physical skills is mostly carried out in the context of
day-to-day social interactions between children and teachers.
In Germany, public child care provision is highly subsidized, with parental fees covering only about
10% of the overall child care costs.12 By the early 1990s, the heavy subsidization lead to a severe
rationing of child care slots, as reflected by the long waiting lists. Rationing was most severe for 3-year
old children and children of non-employed mothers, as the age of the child and the employment status of
the mother were the key admission criteria. To eliminate rationing, the federal government introduced
in August 1992 a mandate which, by January 1st 1996, entitled every child to a subsidized 4-hour child
care slot from her third birthday until school entry. This lead to a rapid expansion in child care facilities
across (West) Germany. The construction and financing13 of the additional child care slots needed was
the responsibility of the municipalities.
3.3
Expansion in Child Care Facilities
In the following, we describe how the federal mandate affected the child care attendance rates of children
of native and immigrant ancestry in the particular region we consider in our empirical analysis (i.e.,
11
The calculation is based on data from the Statistical report on child care institutions from the Lower-Saxonian Statistical office (Niedersächsisches Landesamt fuer Statistik, 2004, p. 19).
12
A typical fee paid by parents is, depending on the parent’s income, between 54 and 129 Euros per month for a 4-hour
slot in 2006.
13
Municipalities received a subsidy of roughly 25% of the overall construction costs from the state. The regions also
contributed to the construction costs. Additionally, 20% of personal expenditure is subsidized by the state.
10
the region for which we have access to the school entry examinations). This region, the Weser-Ems
region, is part of the state of Lower-Saxony, and shares a border with the Netherlands. It is mostly
rural; the largest cities are Osnabrueck (around 270,000 inhabitants), Oldenburg (around 160,000 inhabitants), Wilhelmshaven (around 80,000 inhabitants), and Emden (around 50,000 inhabitants). The
region consists of 139 municipalities, with an average population size of 17,825.14 In 2004 the region
counted 2,473,998 inhabitants in total.
In this region, the construction of child care slots imposed too strong constraints on municipalities
so that providing a child care slot to every 3-year-old by January 1996 was not realizable. The state
government of Lower Saxony therefore allowed for exceptions from the universal legal claim to a child
care slot until December 31, 1998. This lead to a staggered timing and intensity of the construction of
child care slots across municipalities—which we will exploit in our empirical analysis.
Overall, between 1994 and 2002 more than 14,600 new child care slots for children aged between 3
and 6 were constructed in the region we consider, increasing the ratio between the number of available
child care slots and the number of children aged 3 to 6 from 62% to 78%.15 We provide more details
in Figure 2 where we plot the average number of years children spent in public child care by birth
cohort, separately for immigrant and native children, for our estimation sample (see Section 4 for a
data description). Prior to the expansion in child care facilities, children of native ancestry spent on
average 0.7 more years in public child care than children of immigrant ancestry (i.e., 2.12 years vs 1.43
years for the 1988 birth cohort). For native children, average attendance is approximately 0.3 years
higher for the 1998 birth cohort which fully benefited from the expansion in child care facilities than
for the 1988 birth cohort which was not affected by the expansion. The increase in the number of years
in child care between the 1988 and 1998 birth cohort was considerably larger for immigrant than for
native children (0.7 years vs 0.3 years), but even for the 1998 birth cohort their average attendance
lacks behind that of native children (2.48 vs 2.15 years for the 1998 birth cohort).
14
Calculations based data on total number of inhabitants in 2006 provided by the Lower-Saxonian Statistical Office.
Calculations based on data from the Statistical reports on child care institutions and population from the LowerSaxonian Statistical Office (1994 and 2004).
15
11
In Figure 3, we break down the average number of years in child care into the share of children who
do not attend child care at all, and who attend child care for 1, 2 or 3 years. For children of native
ancestry, the increase in average attendance is primarily driven by an increase in the 3-year attendance
rate, which increased from 34% for children born in 1988 to about 56% for children born in 1998. The
increase in the 3-year attendance rate was accompanied by a decrease in predominantly the 2-year
attendance rate, although the 1-year attendance rate also slightly declined. This reflects that for native
children, the share of children attending child care for at least one or at least two years was already
high before the expansion in child care slots (95% and 81%, respectively for the 1988 birth cohort),
suggesting that for these children, the rationing of child care slots mostly affected 3-year-old children,
and not 4- and 5-year-old children.
The picture for children of immigrant ancestry is different. Similar to children of native ancestry, the
3-year attendance rate is markedly higher for children born in 1998 than for children born in 1988 (44%
versus 18%). In contrast to native children, the increase in the 3-year attendance rate was accompanied
by a decrease in both the 2-year, 1-year, and the non-attendance rate. This reflects that before the
expansion in child care slots, the share of children attending child care for at least one or at least
two years was markedly lower for immigrant than for native children (78% vs 95% and 45% vs 81%,
respectively, for the 1988 birth cohort). One explanation for the strong increase in the 1- and 2-year
attendance rates for immigrant children is that, due to the lower employment rates of their mothers (see
also Table 10), 4- and 5-year-old immigrant children were more affected by the rationing of child care
slots than 4- and 5-year-old native children. An alternative explanation is that the federal mandate
increased awareness among immigrant parents (relative to native parents) about the availability of
highly subsidized public child care and also improved the knowledge about the process of how to apply
for child care places.
12
4
Data and Descriptive Evidence
4.1
School Entrance Examination
A unique feature of the German school system is that all children undergo a compulsory school entry
examination before starting elementary school. The main purpose of the examination is to assess
children’s school readiness. The examination typically takes place in the nearby elementary school of
the child’s municipality of residence in the months between February and June before school entry in
August, lasts about 45 minutes, is conducted by government pediatricians, and includes a battery of
tests of gross and fine motor skills as well as an assessment of the child’s speech development. Our
empirical analysis draws on administrative data reporting results of these examinations for one large
region in West Germany (the Weser-Ems region; see Section 3.3 for a brief description of this region)
for the years 1994 to 2006.
We focus on four test outcomes: Language or speech difficulties for children of native ancestry and
proficiency in the host country language for children of immigrant ancestry; fine motor skill problems;
gross motor skill problems; as well as an overall measure for school readiness. Similar indicators are
commonly used to assess the child’s school readiness in the U.S.16 , and have been found to be strong
predictors of later academic success (e.g. Duncan et al., 2007; Grissmer et al., 2010 and Pagani et al.,
2010). In particular motor skills are influential for cognitive development early on (e.g. Pagani and
Messier, 2012) and research suggests that they are a stronger overall predictor for later test scores than
even early reading or math test scores (Grissmer et al., 2010). An important further advantage of our
data is that our outcome variables are standardized assessments by health professionals as opposed to
subjective assessments by parents, which may be subject to a number of sources of bias.17
Language and speech difficulties are identified in our data through the conversation between the
pediatrician and the child during the examination, as well as through three tests where the child is
16
The National Education Goals Panel in US defines school readiness by physical well-being and motor development;
social and emotional development; approaches towards learning; language development; and cognition (National Education
Goals Panel, 1995). Many of these are measured in the school entry examinations.
17
In an appendix, Baker, Gruber and Milligan (2008) provide a detailed discussion on this issue.
13
asked to articulate words correctly, to describe a picture, and to repeat sentences. Since these elaborate
tests are often conducted only among native speakers, we use this outcome only for native children. The
outcome we use for immigrant children instead is an indicator whether the pediatrician recommends
German training because the child’s German and speech comprehension is insufficient.
The diagnosis of fine motor skills problems is based on drawing tasks (e.g. drawing of shapes, such
as triangles or crosses, or of a person) testing finger dexterity, as well as on tasks testing the hand-eye
coordination and diadochokinesia (the capacity to bring a limb alternately into opposite positions),
for example by asking the child to demonstrate the movement of trying a door knob or screwing in
a light bulb and performing the finger-thumb-test, in which the child touches each finger of one hand
to the thumb of the same hand in an alternating pattern. Tests that check for gross motor skill and
coordination problems include standing on one leg for at least 10 seconds without shaking and holding
on to something, jumping on one leg, astride and scissor jumps, hopping to the side, balancing on a
line, and exercises with a ball.18
Based on all tests the pediatrician gives a recommendation to parents and the elementary school
whether to hold the child back from immediate school attendance. This outcome may thus be seen as
an assessment on the overall school readiness of the child. It is an interesting outcome variable also
because deferment from school entry may lower lifetime earnings, due to delayed entry into the labor
market.19
In addition to test outcomes, our data include information on the number of years the child spent in
public child care. Other than migration background, parental background characteristics are included
in our data only from 2002 onwards. We merge to this data source the number of inhabitants, social
welfare recipients and unemployed in the municipality, obtained from the statistical office in Lower
Saxony, as well as characteristics of the local labor market in the municipality (such as the median wage
or the share of university graduates in the workforce), obtained from Social Security Records.
18
In our data, motor skill problems and language and speech difficulties take four values depending on the severity of
the anormality. As very severe levels are a rare outcome and the multi-valued outcome variable is lacking a meaningful
cardinal scale (see Cunha and Heckman, 2008), we have transformed variables into binary outcome variables.
19
Deming and Dynarski (2008) find that in the U.S. ”red-shirting”, i.e. the delayed entry into first grade or Kindergarten,
reduces educational attainment and lifetime earnings.
14
4.2
Sample and Test Outcomes
From this data base, we select school entry cohorts who were most affected by the expansion in child
care slots, which are children examined between 1994 to 200220 . We restrict the sample to children
born between January and June. We impose this restriction because children born between July and
December may opt to start school one year earlier than implied by the age of school entry cut-off rule
(which foresees that every child that turns 6 before the 30th of June enters elementary school in the fall
of the respective year, whereas every child that turns 6 after that cut off enters elementary school in
the following year). As a consequence, some children born between July and December are examined
in the year they turn 6, while others are examined in the year they turn 7. In contrast, all children
born between January and June are examined in the year they turn 6.21 In addition, we exclude 21
municipalities which, due to IT problems, provided the data only partially. Our baseline sample consists
of 118 municipalities.
Table 1 provides a first overview of our sample and outcome variables, separately for children of
native and immigrant ancestry. Overall, our sample consists of about 78,000 native children and about
10,000 immigrant children, although the number of observations slightly varies according to the outcome
variables. Children from the former Soviet Union form the largest immigrant group, making up more
than 40% of immigrant children in our sample. Most of these children are ethnic Germans and their
parents arrived in Germany predominantly in the early 1990s, starting in the late 1980s, after the
breakdown of the communist regimes in Eastern Europe. Children of Turkish descend form the second
largest immigrant group, making up roughly one quarter of immigrant children in our sample. The
Turkish arrived in Germany predominantly in the 1960s and 1970s. Hence, the parents of children of
Turkish descend have resided in Germany much longer than the parents of the children who arrived
from the former Soviet Union, but nevertheless often do not use German at home (see footnote 3 for
evidence).
20
The 2002 examination cohort was 3 years old in 1999, which is the first year, when the legal claim was fully binding.
Including children born between July and December, and controlling for age at test in a flexible manner, has little
impact on our results.
21
15
Developmental delays are common in our sample. 22.2% of native children show language and speech
difficulties, and pediatricians recommend German language training to more than one third of immigrant
children. While fine and gross motor skill problems are equally common among native and immigrant
children (roughly 17% of children experience some problems), recommendations for deferment from
immediate school entry are almost twice as likely for immigrant than for native children (12.7% versus
22.6%). In the last row, we compute a summary measure that is equal to 1 if the child either has some
fine or gross motor skill problems or shows some speech or language difficulties (native children) or is
recommended for German training (immigrant children). Overall, 38.7% of native children and 46.4% of
immigrant children experience any kind of problem, indicating that the different outcome measures are
not perfectly correlated. With the exception of German training for immigrant children, developmental
problems are more pronounced for boys than for girls for both native and immigrant children.22
4.3
Parental Background and Child Care Attendance
Table 2 provides additional information about child care attendance and parental characteristics of the
two groups of children. Since parental background characteristics are included in our data only from 2002
onwards, numbers refer to the years 2002 to 2006 and hence after the expansion in child care facilities.
On average, children of immigrant ancestry children come from a more disadvantaged background than
children of native ancestry: Mothers and fathers of immigrant children are less educated, younger, have
more children, and are less likely to work than mothers and fathers of native children. For native
children, there is a strong positive selection into child care attendance in terms of observable parental
background characteristics, in particular with respect to parental education: While the share of mothers
and fathers with no post-secondary education is 16.7% and 6.0%, respectively, for children who do attend
child care for one year only, it is only 5.6% and 2.5%, respectively, for children who attend child care
for three years or more. Similarly, while only 7.3% and 12.6% of mothers and fathers of children who
attend child care for one year only have some tertiary education, the share is 18.7% and 29.5% for
22
Deming and Dynarski (2008) also find for the U.S. that red-shirting is more common for boys than for girls.
16
mothers and fathers of children who spend three years or more in child care.
Interestingly, selection into child care, in particular with respect to parental education, is much less
pronounced for immigrant children. For instance, the share of mothers and fathers with some tertiary
education of children who attend child care for one year is 12.5% and 13.9%, respectively, compared to
14.6% and 17.6% for mothers and fathers of children who spend three years or more in child care. This
will be important for the interpretation of our results below.
5
Identification Strategy
Next, we describe our identification strategy to obtain causal estimates of the impact of child care
attendance on child outcomes.
5.1
Baseline Estimates
Consider a linear causal model that links outcomes of child i in municipality m and school examination
cohort t (Yimt ) to the number of years the child has spent in child care (Careimt ), and allows the impact
of child care attendance on child development to differ for the two groups of children (which we we
index by subscript g) :
′
g
Yimt = ag0 + ag1 Careimt + Ximt
ag2 + fm
+ γ gt + A′imt β g2 + uimt .
(1)
g
Here, Ximt is a vector of observed child characteristics,23 fm
and γ gt denote municipality and school
examination cohort fixed effects, A′imt is a vector of unobserved (to the econometrician) factors that may
affect child outcomes and child care attendance alike, and uimt is an i.i.d. error term. The coefficient
of interest is ag1 , which captures the impact of one additional year spent in public child care on child
outcomes. Note that this specification restricts the impact of child care attendance to be linear in child
care duration. This does not only rule out that the impact of public child care on child development
varies by age, but also ignores possible dynamic effects. We relax this assumption in the next section.
23
We control for gender, age at examination, month of birth dummies, and (for immigrants) ethnic group.
17
The problem with estimating this regression by OLS is the potential selection of children into child
care based on unobservable characteristics A′imt , leading to a potential bias in our estimates. From
Table 2, we expect, due to the strong positive selection into child care, the OLS estimates to be upward
biased for native children. The bias is likely to be smaller for children of immigrant ancestry for whom
the positive selection into child care is much less pronounced than for children of immigrant ancestry.
To obtain plausibly causal estimates, we exploit the staggered timing and intensity of the construction of child care slots across municipalities, induced by the federal mandate which entitles every child
to a 4-hour child care slot from her third birthday up to school entry. We implement our estimator
as an instrumental variable estimator, where we instrument individual-specific child care attendance
Careimt in equation (1) with the aggregate child care attendance rate for children in municipality m
who are examined in year t, which we compute separately for children of immigrant and native ancestry:
¯ gmt .24 Thus, compared to OLS estimation, IV estimation discards all child-specific variation in child
Care
care attendance within municipalities and cohorts—which is likely to be driven by endogenous selection
into child care—, and exploits only variation in the aggregate child care attendance rate of native and
immigrant children within municipalities across cohorts—which, due to the severe rationing of child care
slots at the beginning of our sample period, is likely to be driven by the construction of new child care
places induced by the federal policy reform that differed in intensity and timing across municipalities.
We discuss the key identification assumptions in detail in Section 5.3.
We compute aggregate attendance rates separately for children of immigrant and native ancestry
because, due to the lower initial attendance rates of the former, these children benefited from the
expansion in child care slots more than children of native ancestry (see Section 3.2 and Figures 1 and
2). Consequently, the aggregate attendance rates at the municipality-cohort level of immigrant and
native children, conditional on child characteristics as well as municipality and cohort fixed effects, are
with a correlation coefficient of 0.12 only weakly correlated.
24
Alternatively, we could use the full set of interactions between municipality and school entry cohort effects as an
instrument for individual child care attendance. In practice, this yields very similar results to those reported here. Moffitt
(1996a, 1996b) points out that in the absence of covariates, this IV strategy is formally equivalent to OLS estimation on
data aggregated up to the municipality and school entry cohort level.
18
In order to increase the number of children per municipality and school examination cohort, and
since children are most likely to compete for child care slots with other children in their school entry
cohort, we compute aggregate attendance rates for the whole sample (i.e. all children who are examined
in year t), although our estimation sample includes only children born between January and June only
(see Section 4).25 For native children, the average and median numbers of children per municipality
and year of school examination cohort are 374 and 231, compared to 100 and 53 for immigrant children.
In order to make sure that our findings are not driven by very small municipalities, we exclude these in
our robustness checks (see the first column in Tables 6a and 6b). Throughout the empirical analysis, we
cluster standard errors at the municipality level to allow for an arbitrary auto-correlation of residuals
within municipalities.
We report first stage estimates of the impact of the aggregate attendance rate on the individual
attendance rate, conditional on child characteristics as well as municipality and cohort fixed effects, in
Panel A of Table 3. Since we compute aggregate child care attendance rates for all children, but restrict
the estimation sample to children born between January and June, the first stage coefficient is slightly
lower than 1 (0.954 for native children, and 0.920 for immigrant children). For both groups of children,
the first stage is highly significant, with an F-statistic above 1000 for native children, and above 400 for
immigrant children. For immigrant children, the aggregate attendance rate explains 3.9% of the total
variation in individual child care attendance rates, compared to only 1.4% for native children.
5.2
Non-linear Returns to Public Child Care Attendance
Equation (1) restricts the impact of child care attendance to be linear in child care duration and to be
homogenous by age. Next, we relax this assumption. To keep the notion simple, we drop the superscript
g. Let Y l − Y l−1 denote the marginal impact of attending child care for l instead of l − 1 years. Due to
the near perfect correlation between the number of years spent in child care (l) and the age at which the
child starts child care (since child care attendance is an absorbing state), Y l − Y l−1 also refers to the
25
Computing aggregate attendance rates based on our estimation sample has little impact on our findings.
19
impact of attending child care when the child is between 6 − l and 6 − l + 1 years old. The IV estimate
of a1 in equation (1) may then be interpreted as a weighted average of these marginal effects (see also
Løken, Mogstad, and Wiswall, 2012 and in particular Mogstad and Wiswall, 2011):26
plim âIV
1
=
5
(Y l − Y l−1 )wlIV ,
(2)
l=1
wlIV
=
˜¯ )
l , Care
Cov(Dimt
mt
˜
¯
Cov(Careimt , Caremt )
l
where Dimt
is an indicator variable that is equal to 1 if the child spends at least l years in public child
˜¯
¯
care (or started child care at the age 6 − l), and Care
mt is the residual from a regression of Caremt
on municipality and school examination cohort fixed effects and the same set of control variables as in
equation (1). The weights sum up to 1 and, provided that the monotonicity assumption holds27 , are nonnegative. The IV estimator places more weight on the marginal effects at years of child care attendance
(or at the age at which the child attends child care) that are most affected by the instrument.28
We estimate these weights and provide evidence on the possible non-linearity of returns to public
child care attendance in Section 6.5. To do this, we replace in equation (1) the number of years the
child has spent in child care (Careimt ) with three indicator variables that are equal to 1 if the child has
spent at least 1, at least 2, or at least 3 years in child care (Care≥j
imt , j = 1, 2, 3):
Yimt =
ag0
+
3
g
g
g
′
g
′
agj Care≥j
imt + Ximt a2 + fm + γ t + Aimt β 2 + uimt .
(3)
j=1
The coefficients on these variables, agj , pick up the marginal impact of one additional year in child care,
when the child is 5, 4, or 3 years old, respectively. We then instrument these variables with the aggregate
26
Løken et al. (2012) use a generalized decomposition of the linear IV estimand, which was originally developed by
Angrist and Imbens (1995) for the case of a binary instrument.
27
l
˜
¯ mt ) ≥ 0 for all l.
To ensure that all weights are non-negative, we require Cov(Dimt
, Care
28
The OLS estimator may also be interpreted as a weighted average of the marginal effects, where the weights equal:
wlOLS =
l
˜ imt )
, Care
Cov(Dimt
˜ imt )
Cov(Careimt , Care
˜ imt is the residual from a regression of Careimt on the same set of control variables as above.
Here, Care
20
share of immigrant and native children in the municipality and school examination cohort who spend
¯ ≥j,g
at least 1, 2, or 3 years in child care (Care
mt , j = 1, 2, 3). We report the first stage estimates in Panel
B of Table 3. For both groups of children, all three first stages are highly significant, with an F-statistic
between 328 and 400 for native children, and above 120 for immigrant children; aggregate attendance
rates explain between 1.2-1.5% and 3.6-4.4%, respectively, of the total variation in individual child care
attendance rates.
5.3
Identifying Assumptions
Our baseline linear IV estimates (equation (1)) identify the causal impact of child care attendance on
child development if the native- and immigrant-specific aggregate child care attendance rate in the
g
¯ mt , is uncorrelated with the unobserved child characmunicipality and school examination cohort, Care
teristics, Aimt , conditional on observable child characteristics, Ximt , as well as municipality and school
g
and γ gt . In the non-linear specification (equation (2)), we require
examination cohort fixed effects, fm
the slightly stronger assumption that the share of native and immigrant children in the municipality
and school examination cohort who attend child care for at least 1, 2, or 3 years be uncorrelated with
the unobserved child characteristics Aimt , conditional on observable child characteristics, Ximt , as well
as municipality and school examination cohort fixed effects.
Changes in Composition The first reason why these assumptions might be violated is that
the variation in the aggregate child care attendance rate across school examination cohorts within
municipalities is driven by changes in the composition of children and their parents, and not by the
expansion in child care slots. This could be because of moves of children or their parents to the
municipality. Among native children, moves to the municipality are rare: only 3.08% of children have
moved to the municipality (either from abroad or another municipality) over the last 2 years. Among
children of immigrant ancestry, in contrast, 12.92% did not live in the same municipality 2 years earlier.
For 70% of these, the reason is that they arrived from abroad. The share of new arrivals is particularly
high for children from the former Soviet Union (19.28%), while new arrivals among Turkish children
21
is only slightly higher than that among native children (4.72%). In our robustness checks in Section
6.2, we first drop from our sample all municipalities that between 1994 and 2002 experienced an inflow
of immigrants into the workforce above the median, as well as all municipalities where the share of
immigrant children increases by at least 5 percentage points between two school examination cohorts
(see columns (2) to (4) of Tables 6a and 6b). As a further robustness check, we also report in Section
6.4 results separately for children of immigrant ancestry from Turkey and the Soviet Union (see Panel
B of Table 8).29
Exogenous Expansion in Child Care Facilities Our identification strategy would further be
violated if regions which differ with respect to the intensity and timing of the expansion in child care
facilities experience different time trends with respect to child development. Note that unlike most of
the existing literature, which exploits variation in the expansion of child care facilities across broadly
defined local labor markets, we exploit regional variation at a much finer local level, within one large
local labor market (see Section 3.3 for statistics on the size of the region). The common time trend
assumption therefore is more plausible in our study than in many existing studies.
Nevertheless, even in our study the intensity and timing of the expansion in child care slots could
be non-random and driven by events specific to the municipality that themselves have a direct impact
on child outcomes. In order to get a first idea which municipalities in our sample experienced an
above average expansion in child care slots between 1994 and 2002, we regress in Table 4 the change
in the aggregate child care attendance rate between 1994/95 and 2001/2002 on the initial child care
attendance rate in 1994/95, separately for native (column (1)) and immigrant children (column (2)), as
well as on a number of baseline municipality characteristics. By far the most important determinant of
the increase in the native- and immigrant-specific aggregate child care attendance rate in a municipality
29
Alternatively, the plausibility of the assumption that changes in the aggregate child care attendance rate across school
examination cohorts within municipalities are not driven by compositional changes could be assessed by regressing the
native- and immigrant-specific aggregate child care attendance rates on parental characteristics, controlling for observable
child characteristics, municipality and school examination cohort fixed effects. Our data includes parental characteristics
only from 2002 onwards. For the period 2002 to 2006, we find no evidence for a correlation between observable child
characteristics and aggregate child care attendance rates, conditional on municipality and cohort fixed effects. The results
are available from the authors upon request.
22
is the initial attendance rate: An increase in the initial aggregate child care attendance rate in 1994/95
by one year reduces the increase in the average child care attendance rate between 1994/95 and 2001/02
by 0.285 years for native children and by 0.762 years for immigrant children. Interestingly, the baseline
aggregate child care attendance rate for native children has a strong positive impact on the change in
the aggregate child care attendance rate for immigrant children (a coefficient of 0.58), suggesting that
a larger share of the newly created child care slots falls to immigrant children if the initial aggregate
child care attendance rates of native children is higher and their demand for additional child care slots
is thus lower. We do not observe a similar pattern for native children.
Importantly, neither the median wage, the share of medium- and high-skilled individuals, the share
of individuals on social benefits, nor the number of inhabitants or the share of immigrants in the
workforce, help to predict the change in the aggregate attendance rate for native or immigrant children.
Only one of the 14 variables we include, the unemployment rate in the regression for native children,
has a statistically significant coefficient at the 5% level.
In order to further investigate whether the timing and intensity of the construction of new child care
slots is quasi-random, we regress in Tables A1a and A1b the native- and immigrant-specific aggregate
child care attendance rate as well as the share of native and immigrant children who attend child care
for at least 1, 2, or 3 years on the same municipality characteristics as in Table 4, while controlling
for child characteristics, municipality and cohort fixed effects. The municipality characteristics refer
to the year when the child is 5 and 2 years old, i.e. to 1 and 4 years before the school entry exam.
Focusing first on the aggregate child care attendance rate in years (column (1)), all but one municipality
characteristics (the number of inhabitants at age 5) are statistically insignificant for both immigrant
and native children. Turning to the detailed attendance rates based on the share of children who
attend child care for at least 1, 2, or 3 years, only 3 out of 39 municipality characteristics for native
children, and 2 out of 39 municipality characteristics for immigrant children, help to statistically predict
aggregate attendance rates. This supports our assumption that the size and the timing of the expansion
in child care slots is unrelated to underlying trends in child outcomes. Nevertheless, to make sure that
23
our estimates reflect causal effects, we conduct a battery of robustness checks, such as controlling for
observable municipality characteristics or allowing linear municipality-specific time trends. We describe
the robustness checks in detail in Section 6.2 and report them in columns (5) to (8) in Tables 6a and
6b.
Spillover Effects of the Expansion in Child Care Coverage A third reason why our identification strategy might be violated is that the expansion in child care slots lead to a deterioration in child
care quality and hence did not only affect children who are pulled into child care because of the reform,
but also children whose child care attendance is unaffected by the reform. This is discussed by Baker
et al. (2008) as a possible reason for the large negative effects on child development of the introduction
of subsidized universal child care in Quebec. Unlike in the Canadian setting, a sharp decline in quality
seems unlikely in our case, since strict quality standards, like the teacher-child ratio and teacher qualifications, were in place throughout our observation period (see Section 3.2). Moreover, municipalities
were given sufficient time to meet the mandate, precisely in order to avoid a large reduction in child care
quality. If any deterioration in child care quality took place, we would expect the decline to be most
pronounced in municipalities in which the construction of new child care slots happened particularly
fast. As an additional robustness check, we therefore exclude all municipalities with a particularly large
increase in the child care attendance rate between two school examination cohorts; see Panel A of Table
7 in Section 6.3.
Another reason for spillover effects to children whose child care attendance is unchanged by the
expansion in child care facilities are peer effects: Since the expansion increased child care attendance
rates of immigrant children by more than that of native children, and primarily drew in 3-year olds into
public child care (see Figures 2 and 3), native children in child care are exposed to more immigrant
children as well as to younger children after than before the expansion, with possible consequences to
their development. We investigate these peer effects in Panels B and C of Table 7 in Section 6.3.
24
6
Results
6.1
Public Child Care and Child Development: Baseline Results
How does public child care attendance affect child development? We report our baseline results in
Table 5, separately for native (Panel A) and immigrant children (Panel B). In each panel, we display
OLS estimates based on equation (1) in the first row, and IV estimates where we instrument individual
child care attendance with the native- and immigrant-specific aggregate child care attendance rate,
¯ gmt , in the second row. For native children, OLS estimates consistently indicate that public child
Care
care attendance improves child development: One additional year in child care lowers the probability
of deferment by 5.1 percentage points, and reduces the probability of speech and language difficulties
and fine and gross motor skill problems by 2.5, 2.9 and 2.4 percentage points, respectively. However,
the positive impact of child care attendance on child development disappears if we exploit the staggered
timing and intensity of the construction of child care slots across municipalities in the IV estimates in
the second row. In fact, in contrast to the OLS estimates, the IV estimates indicate that one additional
year in public child care may increase developmental problems—although the effects are imprecisely
estimated and not statistically significant. In line with the evidence in Table 2, the difference between
OLS and IV estimates points to a positive selection of native children into child care.30
For immigrant children, in contrast, OLS and IV estimates are similar in magnitude. This is once
again consistent with the evidence in Table 2 which shows that the selection of immigrant children into
child care is small in terms of observable family background characteristics. Both OLS and IV estimates
consistently suggest that public child care attendance improves child outcomes. According to our baseline IV estimates, one additional year in child care reduces the probability of receiving German training
by 14.1 percentage points, lowers the probability of fine and gross motor skill problems by 8.4 and 2.1
percentage points, and decreases the probability of deferment by 8.4 percentage points. Evaluated at
the baseline, these effects are large and imply that a one-year increase in the child care attendance rate
30
Alternatively, the two estimators may place different weights on the marginal effects. However, our calculations show
that OLS and IV weights are very similar (see Table 9, column (1)). The difference between the OLS and IV estimates in
Table 5 are therefore largely due to a positive selection bias in the OLS estimates.
25
lowers the probability of receiving German training by 41.1 percent, reduces the probability of fine and
gross motor skill problems by 51.7 and 14.8 percent, and decreases the probability of deferment by 37.2
percent.
6.2
Composition and Exogenous Expansion
We report a number of robustness checks in Tables 6a (children of native ancestry) and 6b (children
of immigrant ancestry). To make sure that our findings are not driven by small municipalities with
few children, we drop in column (1) all municipality-school examination cohort cells with less than 100
native children (Table 6a) or less than 20 immigrant children (Table 6b). Results are similar to our
baseline findings in Table 5.
In columns (2) to (4), we investigate the assumption behind our IV estimates that the variation in
the aggregate attendance rate, conditional on municipality and school examination cohort fixed effects,
is not driven by variation in the parental composition of children. We first drop movers from our sample
(column (2)), and then exclude all municipalities that experienced an inflow of immigrants (measured as
the percentage point change in share of immigrants in the working population) between 1991 and 2002
above the median (column (3)), as well as municipalities in which the share of children of immigrant
ancestry in the school entry examinations increases by at least 5 percentage points from one year to the
next (column (4)). Our results are largely unchanged.
In columns (5) to (8), we investigate the assumption behind our IV estimates that the intensity and
timing of the construction of new child care slots was quasi-random. Results in column (5) correspond to
our baseline specification, but we now add a wide array of time-varying municipality characteristics (the
same ones as in Tables A1a and A1b) referring to 1 and 4 years before the school entry examination when
children were 5 and 2 years old, respectively. The specification in column (6) allows for municipalityspecific linear time trends. In column (7), we allow for a fully flexible time trend with respect to
the child care attendance rate in the baseline school examination cohort (the average group-specific
aggregate child care attendance rate over 1994/1995), the key determinant for the construction of new
26
child care slots (see Table 4). In a similar spirit, in column (8) we allow for a fully flexible time trend
with respect to observable municipality characteristics (the same ones listed in Table 4) in the base
year. All specifications yield similar results: For native children, our findings point to a small negative
and mostly insignificant effect of child care attendance on child development. For immigrant children,
findings based on the robustness checks are close to our baseline estimates in Panel B of Table 5 and
indicate that one additional year in child care significantly reduces the probability of developmental
problems.
In sum, we conclude that our findings are robust to alternative specifications, and that potential
biases due to changes in parental composition and non-randomness in child care slot expansion are
small.
6.3
Changes in the Quality in Child Care and Spillover Effects
We now investigate the question whether the expansion in child care facilities also affected children whose
child care attendance was unchanged by the expansion. We first analyze whether the expansion in child
care facilities lead to a general deterioration in the quality of child care, by excluding municipalities
which experienced a particularly large increase in aggregate child care attendance rates between two
years and for which the decline in the quality of child care should be most pronounced (Panel A of Table
7). This has hardly any impact on our estimates for both immigrant and native children, suggesting
that the construction of new child care facilities did not lead to a general decline in the quality of child
care.
Next, we analyze whether the increased exposure to immigrant children increased developmental
problems of native children. We do this by including the aggregate attendance rate of immigrant
children in years as an additional control variable in our baseline IV regressions (Panel B). This variable
has only a small impact on developmental problems of native children, and its inclusion has little impact
on our IV estimates for native children. Spillover effects from immigrant to native children therefore
appear to be small.
27
Another reason why children whose child care attendance rates are unchanged by the expansion in
child care slots are nevertheless affected by the reform could be their increased exposure to younger
(i.e., 3-year old) children. We investigate this possibility by including the (approximated) share of 3year-olds in child care in the municipality and cohort that is examined one year later (i.e. the cohort
that is one year younger and attends the same child care facilities as the older cohorts) as an additional
control variable in our baseline IV regressions (Panel C). For native children, an increased exposure
to 3-year-olds generally increases the probability of a developmental problem. This effect is, however,
statistically significant for only one outcome variable (gross motor skills). The inclusion of the share of
3-year-olds also reduces the magnitude of the negative point estimate of public child care attendance
on language and gross motor skill problems, pointing to the possibility that the negative estimated
coefficient for native children may partly be driven by negative spillover effects from younger to older
children. This interpretation is however only suggestive, as the differences between these estimates and
our baseline estimate are not statistically significant.
For immigrant children, an increased exposure to 3-year-olds strongly (and significantly) increases
the probability of a German training recommendation, but has little impact on the other outcome
variables. Furthermore, the inclusion of the share of 3-year-old attenders in the younger cohort has only
a very small impact on the IV estimates for all outcome variables. Thus, these findings suggest that for
immigrant children, the spillover effects from younger to older children did not lead to a large bias in
our estimates.
6.4
Heterogenous Effects by Gender and Ethnicity
Do the effects of child care attendance differ by gender? In Table 8, we report results based on our
baseline specification (equation (1)) estimated separately by gender, where we instrument the individual
child care attendance rate with the native- and immigrant-specific overall aggregate child care attendance rate. For native children, there is some suggestive evidence that child care attendance increases
the probability of deferment for boys, but not for girls. For all other outcomes, point estimates are
28
similar for boys and girls.
For immigrant children, the absolute impact of spending one additional year in child care is somewhat larger for boys than for girls. For instance, while one additional year in child care reduces the
probability that the pediatrician recommends German training by 16.3 percentage points for boys and
10.1 percentage points for girls. Recall, however, from Table 1 that developmental problems (with
the exception of German training) are more pronounced for boys than for girls, which mitigates the
difference in effects between boys and girls.31
Our sample of children of immigrant ancestry includes diverse groups, with children from the former
Soviet Union whose parents recently migrated to Germany and of whom 19.28% did not live in the same
municipality two years before the school entry examination, and from Turkey whose parents, or even
grandparents, migrated to Germany primarily in the late 1960s and 1970s and of whom only 4.72% did
not live in the same municipality two years before the school entry examination, forming the two largest
groups. At the bottom of Panel B in Table 8, we report results based on our baseline specification
(equation (1)) estimated separately for these two groups, where we instrument the individual child care
attendance rate with the overall immigrant-specific aggregate child care attendance rate. Interestingly,
public child care attendance reduces the incidence of developmental problems, including the incidence
that the pediatrician recommends German training, for both groups of immigrant children, suggesting
that child care attendance does not only benefit children who recently migrated to Germany, but also
children whose parents were born in Germany or migrated to Germany many years ago.
6.5
Non-Linear Effects
Our results so far restrict the impact of the time spent in child care to be linear in child care duration
and to be homogenous by age. In Section 5.2, we have shown that our baseline IV (and OLS) estimates
in Table 5 (based on equation (1)) are a weighted average of the marginal treatment effects (see equation
31
Unlike our findings here, several studies, in particular those studying the effects of specific intervention programs (see
e.g. Anderson, 2008), often find larger improvements in cognitive outcomes for girls than for boys. One reason why we
find somewhat larger effects for boys than girls could be that in our setting, the learning environment is informal and playoriented and focuses on school readiness in terms of physical and language skills rather than on early academic training.
This may be an environment which is particularly beneficial for immigrant boys.
29
(2)). In the first column of Table 9, we report the weights given to each marginal treatment effect for
immigrant and native children in the OLS and IV estimation. Both estimation methods yield similar
weights. In line with Figure 3, the marginal return of attending child care for more than 2 versus
2 years receives the highest weight for native children (53.9% in the IV estimation), followed by the
marginal return of attending child care for 2 versus 1 year (29.93%) and for 1 versus 0 years (16.17%).
For immigrant children, in contrast, the IV estimator places larger weights on the marginal return of
attending child care for 2 versus 1 year (39.11%) and for 1 versus 0 years (22.56%), suggesting that
the expansion in child care slots affected immigrant children more often at the extensive margin (1
versus 0 years) and at a low level of the intensive margin (2 versus 1 year) of the child care attendance
distribution than native children. Hence, if the benefits of public child care attendance are decreasing in
child care duration (or larger when the child is older), this may help to explain why the overall benefits
of public child care attendance are larger for immigrant than for native children.
Next, we investigate whether the returns to public child care attendance are non-linear. We first
report OLS estimates and the IV estimates based on equation (3), where we instrument the individual
probabilities that a child attends child care for at least 1, 2 or 3 years with the aggregate share of native
and immigrant children in the municipality and school examination cohort who attend child care for at
least 1, 2 or 3 years. For native children, OLS estimates suggest that the first, second, and third year in
child care reduces developmental problems by roughly the same amount. Turning to the IV estimates,
no clear picture emerges and returns are imprecisely estimated.
For immigrant children, OLS and IV estimates tend to be similar in magnitude, confirming once
again that for these children the bias due to selection into child care is small. For two outcomes—
recommendation of training in the German language and deferment— both OLS and IV estimates are
suggestive of decreasing marginal returns to child care attendance. While the first and second year
in child care (i.e. when the child starts child care at the age of 4 or 5) reduces the probability of a
recommendation in German training by about 20 percentage points, the third year (i.e. when the child
starts child care at the age of 3) lowers the probability by only 9 percentage points and the effect is
30
statistically insignificant. This is consistent with the hypothesis that children younger than the critical
period for language acquisition (around puberty, see e.g. Lenneberg, 1967) learn a new language very
easily, and that two years of native language exposure prior to school is sufficient to avoid language
deficits. Similarly, while the first and second year in child care decreases the probability of deferment by
about 14 and 10 percentage points, the third year reduces the probability only by about 3.7 percentage
points, an effect that is again statistically insignificant. Hence, for these two outcomes, non-linear
returns to public child care attendance coupled with different weights given to the marginal treatment
effects help to explain why immigrant children benefit more from spending an additional year in public
child care than native children.
For fine motor skill problems, in contrast, the marginal effect of attending an additional year in
child care attendance is highest in the third year when children are between 3 and 4.32 One explanation
for this finding is that children’s fine motor skills rapidly progress around this age. Hence, it may be
particularly important to encourage fine motor skills at that age to prevent problems later on.33
6.6
The Expansion in Child Care Coverage on Mothers’ Labor Supply
Our discussion in Section 3.1 highlights that an important indirect channel through which child care
attendance may affect child outcomes is maternal labor supply (see also Panel A of Figure 1). Next, we
investigate the impact of the expansion in child care facilities on maternal employment, using data from
the Microcensus on all mothers residing in West Germany.34 Figure 4 gives a first impression of the link
between the expansion in child care slots and mothers’ employment rates for native mothers. In Panel
A of the figure, we plot child care attendance rates of native children who in April were 3 or 4 years old
32
We have also run alternative specifications with quadratic terms, using the individual attendance and its square,
instrumented by the attendance rate and squared attendance rate for each group. These estimates paint a similar picture
as the results reported above. In particular, for minority children the coefficient on the quadratic term is positive and
statistically significant for German training recommendation and deferment (pointing towards decreasing marginal returns),
and positive and statistically significant for fine motor skills (pointing towards increasing marginal returns).
33
Gallahue and Ozmun (2005) stress that early childhood is the fundamental phase of motor development, where motor
skills develop fast and children initially develop and refine basic movement skills, which are the basis for developing more
complex movement skills at later childhood.
34
To increase the sample size, we include, unlike in the school entry examination data, children born in the second half
of the year in our sample. We find similar results when we restrict the sample to children born in the first half of the year.
When we restrict our sample to Northern German states to make our sample more similar to our sample on children, we
also obtain very similar results.
31
over time. As expected, the attendance rate of 3- and 4-year-olds sharply rose between 1993 and 2001,
by about 20 percentage points, respectively. In Panel B of the figure, we plot the corresponding labor
force participation rates of mothers with 3- or 4-year old children (”treatment”) and mothers whose
youngest child is 11- or 12-year old (”control”).35 Labor force participation rates of mothers with a 3- or
4-year old child (for whom child care attendance rates increased sharply) rose by 8.7 percentage points
between 1993 and 2001. However, labor force participation rates of mothers whose youngest child is
11- or 12-year old (who were thus unaffected by the construction of new child care slots) increased by
almost the same amount over the same time period. This suggests that the increase in the labor force
participation rates of German mothers with 3- or 4-year old children is mostly due to an aggregate time
trend, and not caused by the expansion in child care facilities.
We provide additional information in Table 10 based on the same data set separately for native
(Panel A) and immigrant mothers (Panel B). For native mothers, the difference-in-difference (”DinD”)
estimate, which calculates the differential increase in employment rates between the treatment group
over the control group, points to a small positive, but statistically insignificant, effect of the expansion
in child care slots on maternal employment: a 10 percentage points increase in the attendance rate of
3- to 4-year-olds leads to an increase in employment rates of their mothers by 0.63 percentage points
(0.012/0.19·10). The findings in Panel B of Table 10 further indicate that the expansion in child care
slots also had only a small effect on labor force participation rates of immigrant mothers: While the child
care attendance rate of 3- or 4-year old immigrant children rose by 23 percentage points between 1993
and 2001, the labor force participation rates of their mothers, as well as the labor force participation
rates of mothers with a 11- or 12-year old child, remained roughly constant.
Hence, in our setting, the effect of child care attendance on child outcomes through an increase in
maternal labor supply is small for both native and immigrant children. Since public child care is heavily
subsidized, with parental fees contributing to only 11% of the total costs of child care (see Section 3.2),
35
We use mothers of children, whose youngest child is 11 or 12 years as our control group to ensure that the mothers
were not affected by the increase in public child care supply through their younger children. Using mothers with children
in a similar age range as a control group leads to similar results.
32
we also expect the impact of child care attendance on monetary investments to be small.36 Public child
care attendance therefore affects child outcomes primarily directly, as well through its direct effects on
time investments made by the mother and by alternative care givers (see Section 3.1 and Panel A of
Figure 1).
Does public child care primarily replace maternal care or alternative care? In Panel C of Table 10, we
provide some evidence on the counterfactual care arrangement, by separately reporting the 2001-1993
change in the share of mothers with a 3- or 4-year old child who are working and use child care—these
mothers are likely to have used alternative forms of child care prior to the expansion in child care slots
and switched to public child care because of the expansion—, and the change in the share of mothers
who are not working and yet use child care—these mothers are likely to have taken care of their child
themselves prior to the expansion in child care slots. The results reveal an interesting difference between
native and immigrant mothers. For native mothers, the share of children in child care increases much
more for working than for non-working mothers, suggesting that for native children public child care
primarily replaces alternative care. For immigrant mothers, in contrast, the increase in the share of
mothers who are not working and yet use child care is twice as large as the increase in the share of
mothers who are working and use child care. Hence, for immigrant children public child care primarily
crowds out maternal care. This partially reflects the much lower labor force participation rates of
immigrant mothers than native mothers (see Panel A of Table 10).37
36
Black, Devereux, Løken and Salvanes (2012) use a regression discontinuity approach to analyze the impact of child
care subsidies on child care utilization rates and mothers’ labor force participation rates and find small effects on both.
The positive effects of the subsidy on child outcomes found in this paper are therefore driven by the income effect of the
subsidy.
37
Unfortunately, our data do not allow us to directly investigate the impact of the expansion in child care slots on time
investments made by parents and alternative care givers. Even if immigrant children who are pulled into public child care
due to the expansion in child care slots would otherwise have been cared for by their mother, public child care attendance
might nevertheless increase time investments by the mother. In line with this hypothesis, Gelber and Isen (2011) find that
Head Start attendance caused a substantial increase in parents’ involvement with their children. Recent research by Aizer
and Cunha (2012) further investigates these adjustment mechanisms.
33
7
Discussion and Conclusion
In this paper, we study the impact of time spent in public child care on children who are about to
enter school. We do this in a setting where the quality of care is high, fairly homogenous, and almost
exclusively public, and where learning is mostly informal, through the social interactions with teachers
and other children. In our empirical analysis, we draw on rare administrative data on the population of
all children for one large region in Germany. Our outcome variables include indicators for developmental
problems (such as language deficits or gross or fine motor skill problems) that may interfere with the
child’s success at school. Similar indicators have been found to be strong predictors of later academic
success (e.g., Duncan et al., 2007).
To deal with the possibly non-random selection of children into public child care, we make use of a
federal policy reform which entitles every child to a largely subsidized 4-hour child care slot from her
third birthday until school entry. In the region we study, this mandate lead to a staggered timing and
intensity of the construction of child care slots across municipalities—which is the variation we exploit
in our analysis.
We study the effects of public child care attendance separately for children of immigrant and native
ancestry, and find very different effects for the two groups. While public child care attendance strongly
and robustly reduces language and motor skill problems and improves school readiness for immigrant
children, it has no significant effects for native children. We explore a number of possible explanation
for these differences. A first reason may be non-linearities in returns to child care attendance. Due to
the lower initial child care attendance rates of children of immigrant ancestry, the expansion in child
care slots increased both the one-, two-, and three-year attendance rate for immigrant children, but
affected primarily the three-year attendance rate for children of native ancestry. At the same time, we
find for some outcomes, like language proficiency and overall school readiness, some suggestive evidence
for decreasing marginal returns to child care attendance for immigrant children.
A second reason may be different counterfactual care arrangements for immigrant and native children. Our findings suggest that for immigrant children, public child care attendance primarily crowds
34
out maternal care, while for native children it primarily replaces alternative care—which is likely to be
more similar to public child care than to maternal care, in particular with respect to the exposure and
interaction with other children.
A third explanation for the larger benefits of public child care attendance for immigrant than for
native children is that the returns to investments in public child care, relative to investments made by the
mother or alternative care givers, are higher for immigrant children. In our setting, immigrant children
are more likely than native children to come from a disadvantaged background, and thus from a group
for which several studies have found the benefits of formal child care attendance to be particularly large
(see e.g. Havnes and Mogstad, 2011a). Even among disadvantaged children, immigrant children may be,
due to the specific deficiencies they face, particularly perceptive to the learning environment provided
by public child care. Foremost is here a lack in proficiency in the language of instruction. Deficiencies
in language proficiency may lead to serious under-performance at early elementary education, with
potentially damaging consequences not only far into the educational curriculum, but also for many
other aspects of life (see e.g. Bleakley and Chin 2008, 2010). The exposure to other children in public
child care, especially children whose first language is the host country language, is likely to quickly
reduce this deficit. Moreover, host country language proficiency may facilitate learning in other areas
and thus lead to complementarities and positive spillover effects—hence explaining the positive effects of
child care attendance on other measures of early achievement besides on language proficiency.
Overall, our results emphasize that the benefits of universal child care programs may not be uniform
but concentrated among specific groups of disadvantaged children who are particularly perceptive to
the informal learning environment which public child care provides. This suggests that such programs
may be successful at ”leveling the playing field” at school entry (and possibly beyond), and may thus, as
stressed by Currie (2001), be a more cost-effective way for governments to reduce life-time inequalities
rather than compensating for unequal outcomes later in life.
35
References
[1] Aizer, Anne, and Flavio Cunha. 2012 ”The Production of Child Human Capital: Endowments,
Investments and Fertility.” Brown University, mimeo.
[2] Almond, Douglas, and Janet Currie. 2011. ”Human Capital Development before Age Five.” In
Handbook of Labor Economics, Elsevier: 1315-1486.
[3] Angrist, Joshua D., and Guido Imbens. 1995. ”Two-Stage-Least Squares Estimations of Average
Causal Effects in Models with Variable Treatment Intensity.” Journal of the American Statistical
Association 90 (430): 979-1014.
[4] Baker, Michael, Jonathan Gruber, and Kevin Milligan. 2008. “Universal Childcare, Maternal Labor
Supply, and Family Well-Being.” Journal of Political Economy 116(4):709—45.
[5] Baker, Michael. 2011. ”Innis Lecture: Universal early childhood interventions: what is the evidence
base?” Canadian Journal of Economics/Revue canadienne d’économique 44(4):1069—1105.
[6] Becker, Gary S. 1964. Human Capital: A Theoretical and Empirical Analysis, with Special Reference to Education. Chicago: University of Chicago Press.
[7] Behrman, Jere R., Yingmeim Cheng, and Petra E. Todd. 2004. ”Evaluating Preschool Programs
When Length of Exposure to the Program Varies: A Nonparametric Approach.” Review of
Economics and Statistics, 86(1): 108—132.
[8] Belsky, Jay. 1988. ”The ‘effects’ of infant day care reconsidered.” Early Childhood Research Quarterly, 3(3): 235-272.
[9] Belsky, Jay, and David Eggebeen. 1991. ”Early and Extensive Maternal Employment and Young
Children’s Socioemotional Development: Children of the National Longitudinal Survey of
Youth.” Journal of Marriage and Family, 53(4):1083—1098.
[10] Berlinski, Samuel, and Sebastian Galiani. 2007. ”The effect of a large expansion of pre-elementary
school facilities on preschool attendance and maternal employment.” Labour Economics, 14(3):
665—680.
[11] Berlinski, Samuel, Sebastian Galiani, and Paul Gertler. 2009. ”The effect of pre-elementary education on elementary school performance.” Journal of Public Economics, 93(1—2): 219—234.
[12] Berlinski, Samuel, Sebastian Galiani, and Marco Manacorda. 2008. ”Giving children a better
start: Preschool attendance and school-age profiles.” Journal of Public Economics, 92(5—6):
1416—1440.
[13] Bernal, Raquel, and Michael P. Keane. 2011. ”Child Care Choices and Children’s Cognitive Achievement: The Case of Single Mothers.” Journal of Labor Economics 29(3): 459 — 512.
[14] Black, Sandra, Paul J. Devereux, Katrine Løken and Kjell G. Salvanes. 2012. ”Care or Cash?
The Effect of Child Care Subsidies on Student Performance,” CEPR Discussion Papers 8981,
C.E.P.R. Discussion Papers.
[15] Blau, David, and Janet Currie. 2006. ”Chapter 20 Pre-School, Day Care, and After-School Care:
Who’s Minding the Kids?” In Handbook of the Economics of Education. Elsevier: 1163—1278.
36
[16] Bleakley, Hoyt, and Aimee Chin. 2008. ”What Holds Back the Second Generation? The Intergenerational Transmission of Language Human Capital Among Immigrants.” Journal of Human
Resources 43(2):267—298.
[17] Bleakley, Hoyt, and Aimee Chin. 2010. ”Age at Arrival, English Proficiency, and Social Assimilation
among US Immigrants.” American Economic Journal: Applied Economics, 2(1): 165—92.
[18] Bowlby, John. 1969. Attachment and Loss. London: The Hogarth Press and the Institute of PsychoAnalysis.
[19] Carneiro, Pedro and Rita Ginja. 2012. ”Preventing Problem Behaviors in Childhood and Adolescence: Evidence from Head Start.” University College London, mimeo.
[20] Cascio, Elizabeth U. 2009a. ”‘Do Investments in Universal Early Education Pay Off? Long-term
Effects of Introducing Kindergartens into Public Schools.” National Bureau of Economic Research Working Papers 14951.
[21] Cascio, Elizabeth U. 2009b. ”Maternal Labor Supply and the Introduction of Kindergartens into
American Public Schools”. Journal of Human Resources 44(1): 140—170.
[22] Casey, Teresa, and Christian Dustmann. 2008. ”Intergenerational Transmission of Language Capital
and Economic Outcomes.” Journal of Human Resources 43(3): 660—687.
[23] Cunha, Flavio, and James J. Heckman. 2007. ”The Technology of Skill Formation.” American
Economic Review, 97(2): 31—47.
[24] Cunha, Flavio, and James J. Heckman. 2008. ”Formulating, Identifying and Estimating the Technology of Cognitive and Noncognitive Skill Formation.” Journal of Human Resources 43(4):
738-782.
[25] Currie, Janet. 2001. ”Early Childhood Education Programs.” Journal of Economic Perspectives
15(2): 213—238.
[26] Currie, Janet, and Duncan Thomas. 1995. ”Does Head Start Make a Difference?” American Economic Review 85(3): 341—64.
[27] Currie, Janet, and Duncan Thomas. 1999. ”Does Head Start help hispanic children?” Journal of
Public Economics 74(2): 235—262.
[28] Datta Gupta, Nabanita and Marianne Simonsen. 2010. ”Non-cognitive child outcomes and universal
high quality child care.” Journal of Public Economics, 94(1-2): 30—43.
[29] Deming, David. 2009. ”Early Childhood Intervention and Life-Cycle Skill Development: Evidence
from Head Start.” American Economic Journal: Applied Economics, 1(3): 111—34.
[30] Deming, David, and Susan Dynarski. 2008. ”The Lengthening of Childhood.” Journal of Economic
Perspectives 22(3): 71—92.
[31] Drange, Nina, and Kjetil Telle. 2010. ”The effect of preschool on the school performance of children
from immigrant families. Results from an introduction of free preschool in two districts in Oslo.”
Discussion Papers 631 - Statistics Norway.
37
[32] Duncan, Greg J., Chantelle J. Dowsett, Amy Claessens, Katherine Magnuson, Aletha C. Huston,
Pamela Klebanov, Linda S. Pagani, Leon Feinstein, Mimi Engel, Jeanne Brooks-Gunn, Holly
Sexton, Kathryn Duckworth and Crista Japel. 2007. ”School readiness and later achievement.”
Developmental Psychology 43(6):1428—1446.
[33] Dustmann, Christian, Steve Machin and Uta Schönberg, 2010. ”Ethnicity and Educational Achievement in Compulsory Schooling.” The Economic Journal 120: F272—F297.
[34] Dustmann, Christian, Tommaso Frattini and Gianandrea Lanzara, 2012. ”Educational achievement
of second-generation immigrants: an international comparison.” Economic Policy 27(69): 143185.
[35] Felfe, Christina, Natalia Nollenberger and Núria Rodrı́guez-Planas, 2012. ”Can’t Buy Mommy’s
Love? Universal Child Care and Children’s Long-Term Cognitive Development.” Universität
St. Gallen, Universitat Autònoma de Barcelona, Universitat Pompeu Fabra: mimeo.
[36] Felfe, Christina, and Rafael Lalive. 2011. ”How Does Early child Care affect Child Development?”
http://cep.lse.ac.uk/seminarpapers/20-05-11-RL.pdf.
[37] Gallahue, David L., and John C. Ozmun. 2005. ”Motor development in early childhood education.”
In Spodek, Bernard, and Olivia N. Saracho. Handbook of Research on the Education of Young
Children. Routledge.
[38] Garces, Eliana, Thomas, Duncan, and Janet Currie. 2002. ”Longer Term Effects of Head Start.”
American Economic Review, 91(4): 999-1012.
[39] Gathmann, Christina and Björn Saß. 2012. ”Taxing Childcare: Effects on Family Labor Supply
and Children.” IZA Discussion Papers 6440, Institute for the Study of Labor (IZA).
[40] Gelbach, Jonah B. 2002. ”Public Schooling for Young Children and Maternal Labor Supply .”
American Economic Review, 92(1): 307—322.
[41] Gelber, Alexander M., and Adam Isen. 2011. ”Children’s Schooling and Parents’ Investment in
Children: Evidence from the Head Start Impact Study.” National Bureau of Economic Research
Working Paper No. 17704.
[42] González, Libertad. 2012. ”The Effect of a Universal Child Benefit on Conceptions, Abortions, and
Early Maternal Labor Supply.” forthcoming in: American Economic Journal: Economic Policy.
[43] Gormley, William T., Jr., and Ted Gayer. 2005. ”Promoting School Readiness in Oklahoma: An
Evaluation of Tulsa’s Pre-K Program.”’ Journal of Human Resources 40(3): 533-558.
[44] Gormley, William T., Jr. 2008. ”The Effects of Oklahoma’s Pre-K Program on Hispanic Children.”
Social Science Quarterly 89(4): 916—936.
[45] Grissmer, David, Kevin J. Grimm, Sophie M. Aiyer, William M. Murrah, and Joel S. Steele. 2010.
”Fine motor skills and early comprehension of the world: Two new school readiness indicators.”
Developmental Psychology 46(5):1008—1017.
[46] Havnes, Tarjei, and Magne Mogstad. 2011a. ”No Child Left Behind: Subsidized Child Care and
Children’s Long-Run Outcomes.” American Economic Journal: Economic Policy, 3(2): 97—129.
38
[47] Havnes, Tarjei, and Magne Mogstad. 2011b. ”Money for nothing? Universal child care and maternal
employment.” Journal of Public Economics 95(11—12): 1455—1465.
[48] Heckman, James, 2007. ”‘The economics, technology, and neuroscience of human capability formation.”’ PNAS 104 (33): 13250—13255.
[49] Heckman, James J., Moon, Seong H., Pinto, Rodrigo, Savelyev, Peter A., and Adam Yavitz. 2010a.
“Analyzing Social Experiments as Implemented: A Reexamination from the Evidence from the
High Scope Perry Preschool Program.” Quantatitive Economics, 1(1): 1-46.
[50] Heckman, James J., Moon, Seong H., Pinto, Rodrigo, Savelyev, Peter A., and Adam Yavitz.
2010b. “The Rate of the Return to the High Scope Perry Preschool Program.” Journal of
Public Economics, 94(1-2): 114-128.
[51] Lenneberg, Eric. 1967. ”Foundations of Language.” New York: John Wiley & Sons, Inc.
[52] Løken, Katrine V., Magne Mogstad, and Matthew Wiswall. 2012. ”What Linear Estimators Miss:
The Effects of Family Income on Child Outcomes.” American Economic Journal: Applied
Economics, 4(2): 1—35.
[53] Loeb, Susanna, Margaret Bridges, Daphna Bassok, Bruce Fuller, and Russell W. Rumberger. 2007.
”How much is too much? The influence of preschool centers on children’s social and cognitive
development.” Economics of Education Review 26: 52—66.
[54] Magnuson, Katherine A., Claudia Lahaie, and Jane Waldfogel. 2006. ”Preschool and School Readiness of Children of Immigrants.” Social Science Quarterly 87(5): 1241—1262.
[55] Magnuson, Katherine A., Christopher Ruhm, and Jane Waldfogel. 2007. ”Does prekindergarten
improve school preparation and performance?” Economics of Education Review 26(1): 33—51.
[56] Moffitt, Robert A. 1996a. Selection Bias Adjustment in Treatment-Effect Models as a Method of
Aggregation. National Bureau of Economic Research Working Papers 187.
[57] Moffitt, Robert A. 1996b. ”Identification of Causal Effects Using Instrumental Variables: Comment.” Journal of the American Statistical Association 91(434):462—465.
[58] Mogstad,
Magne,
and
Matthew
Wiswall.
2011.
”Linearity
strumental
Variables
Estimation:
Problems
and
https://files.nyu.edu/mw109/public/mogstad wiswall iv linearity.pdf.
in
InSolutions.”
[59] National Education Goals Panel. 1995. ”Reconsidering Children’s Early Development and Learning:
Toward Common Views and Vocabulary.” Washington DC: National Education Goals Panel.
http://govinfo.library.unt.edu/negp/reports/child-ea.htm.
[60] Niedersächsisches Landesamt für Statistik. 1994. ”Jugendhilfestatistik 1994” Niedersäsisches Landesamt für Statistik: Hannover.
[61] Niedersäsisches Landesamt für Statistik. 2004. ”Einrichtungen und Personal der Jugendhilfe 2002”
Statistische Berichte Niedersachsen: Hannover.
[62] Niedersächsischer
Städtetag.
2002.
”Kindertageseinrichtungen
http://www.nst.de/media/custom/359 38 1.PDF.
39
im
Umbruch.”
[63] Pagani, Linda S., Caroline Fitzpatrick, Isabelle Archambault, and Michel Janosz. 2010. ”School
readiness and later achievement: A French Canadian replication and extension.” Developmental
Psychology 46(5):984—994.
[64] Pagani, Linda S., and Sylvie Messier. 2012. ”Links between Motor Skills and Indicators of School
Readiness at Kindergarten Entry in Urban Disadvantaged Children.” Journal of Educational
and Developmental Psychology 2(1): 95-107.
[65] Ruhm, Christopher J, and Jane Waldfogel. 2011. ”Long-Term Effects of Early Childhood Care and
Education.” forthcoming in: Nordic Economic Policy Review.
[66] Shonkoff, Jack P., and Deborah A. Philipps. 2000. ”From Neurons to Neighborhoods: The Science
of Early Childhood Development.” National Academy Press: Washington, D.C.
40
Figure 1: Channels through which child care attendance affects child development
Panel A: Child care attendance at age t on child development at the end of period t
investments
by mother
child care
attendance
in period t
maternal
labor supply
investments
by alternative
care givers
child development
at the end of
period t
monetary
investments
Panel B: Dynamic Effects
child care
attendance in
period t
child development
at the end of period
t
returns to
investments/
amounts invested
in period t+1
child
development at
the end of
period t+1
Figure 2: Evolution of Number of Years in Child Care by Year of Birth
Note : The figure plots the average number of years children spend in child care by birth cohort, separately for
children of native ancestry and children of immigrant ancestry.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2004, children born between
January and June.
Figure 3: Evolution of Child Care Attendance Duration Groups by Year of Birth
Panel A: Children of native ancestry
Panel B: Children of immigrant ancestry
Note : The figures plot the share of children who do not attend child care at all, and who attend child care for 1, 2, 3 or more years, separately for children of native ancestry
(Panel A) and children of immigrant ancestry (Panel B).
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2004, children born between January and June.
Table 1: Overview of Outcome Variables
Panel A: Children of native ancestry
language and speech difficulties
fine motor skill problems
gross motor skill problems
deferment recommended
at least one problem
.
Panel B: Children of immigrant ancestry
% former Soviet Union
% Turkish
German Training recommended
fine motor skill problems
gross motor skill problems
deferment recommended
at least one problem
N
78,421
63,960
77,952
78,623
62,001
all
22.2%
17.9%
17.1%
12.7%
38.7%
boys
26.7%
25.9%
23.6%
16.9%
49.0%
girls
17.5%
9.6%
10.3%
8.4%
28.1%
42.0
26.0
N
10,011
8,879
10,243
10,393
8,347
all
34.3%
16.3%
14.1%
22.6%
46.4%
boys
35.3%
21.7%
17.6%
26.4%
51.9%
girls
33.2%
10.4%
10.5%
18.5%
40.6%
Note : The table reports, separately for children of native (Panel A) and children of immigrant
ancestry (Panel B) as well as separately for boys and girls, the incidence that a child experiences
language and speech difficulties (need of German training for immigrant children) the incidence of
fine or gross motor skill problems, as well as the incidence that the pediatrician recommends
deferment from immediate school entry. In the last row the variable is an indicator for having at
least one of the first four problems.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children
born between January and June.
Table 2: Parental Characteristics and Child Care Attendance
Panel A: Children of native ancestry
share attendance
no post-secondary education, mother
tertiary education, mother
no post-secondary education, father
tertiary education, father
child lives with both parents
number of siblings
mother's age
mother working
father working
all
no
attendance
1 year
2 years
3 years or
more
7.4%
15.2%
3.1%
24.5%
81.8%
1.27
36.08
76.0%
91.7%
3%
22.8%
9.9%
9.8%
15.3%
80.0%
1.86
35.07
70.6%
78.4%
6%
16.7%
7.3%
6.0%
12.6%
75.7%
1.72
35.11
74.4%
85.0%
34%
7.9%
9.8%
3.2%
17.3%
85.8%
1.35
35.86
77.2%
92.2%
58%
5.6%
18.7%
2.5%
29.5%
80.5%
1.15
36.30
75.8%
92.6%
all
no
attendance
1 year
2 years
3 years or
more
36.9%
13.5%
24.0%
15.9%
87.5%
1.90
32.69
56.1%
76.5%
9%
45.4%
13.0%
28.3%
15.0%
91.3%
3.23
32.77
52.3%
71.0%
16%
38.3%
12.5%
26.4%
13.9%
86.5%
2.33
32.81
55.7%
76.8%
29%
40.7%
12.1%
27.2%
14.3%
89.3%
1.85
32.67
52.8%
75.0%
46%
32.7%
14.6%
20.5%
17.6%
86.0%
1.49
32.65
58.8%
78.1%
Panel B: Children of immigrant ancestry
share attendance
no post-secondary education, mother
tertiary education, mother
no post-secondary education, father
tertiary education, father
child lives with both parents
number of siblings
mother's age
mother working
father working
Note: The table provides information about mean parental characteristics of children of native (Panel A) and of
immigrant ancestry (Panel B) by child care duration.
Source: Administrative Data from School Entry Examinations for children born January to June, Weser-Ems,
2002-2006, for which parental characteristics are available. Number of observations for native children is
between 16,467 and 24,677 and for immigrant between 1,917 and 3,887.
Table 3: First Stage estimates: The Impact of Aggregate Attendance Rates on Indivdual
Attendance Rates
Panel A: Child care attendance in years
Native children
F-statistic
partial R2
0.954
(0.031)**
1100.993
Immigrant
children
0.92
(0.036)**
402.834
0.014
0.039
Panel B: Child care attendance duration groups
Children of native ancestry
share ≥ 1
1.014
(0.045)**
-0.025
(0.019)
0.017
(0.012)
F-statistic
399.61
share ≥ 2
-0.048
(0.050)
1.136
(0.034)**
-0.012
(0.016)
327.525
share ≥ 3
-0.002
(0.063)
-0.03
(0.054)
1.045
(0.031)**
360.84
0.015
0.012
0.014
F-statistic
share ≥ 1
1.022
(0.037)**
-0.035
(0.035)
-0.038
(0.032)
148.62
share ≥ 2
0.004
(0.051)
1.006
(0.046)**
-0.027
(0.042)
140.061
share ≥ 3
0.038
(0.043)
-0.035
(0.049)
0.949
(0.048)**
121.533
partial R2
0.044
0.041
0.036
partial R2
Children of immigrant ancestry
Note : In Panel A, we report key statistics from our baseline linear first stage regression where
we regress individual child care attendance in years on the native- and immigrant-specific child
care attendance rate in years in the municipality and school examination cohort, and control for
municipality and school examination cohort fixed effects as well as gender, age at examination
in months, month of birth dummies and, for immigrant children, ethnicity dummies. In Panel B,
we display key statistics from our non-linear first stage regressions where we regress individual
indicator variables that the child attends child care for at least 1, 2 or 3 years on the aggregate
share of native and immigrant children who attend child care for at least 1, 2 or 3 years in the
municipality and school examination cohort, and control for municipality and school
examination fixed effects as well as the same individual control variables as in Panel A.
Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002,
children born between January and June. Aggregate attendance rates are computed using all
children in the school examination cohort.
Table 4: Determinants of Intensity in Construction of Child Care Slots, 1994 to 2002
change in aggregate child care attendance rate in years,
2001/02 vs 1994/95:
Children of native
Children of immigrant
ancestry
ancestry
aggregate attendance rate in years for native children,
1994/95
aggregate attendance rate in years for immigrant
children, 1994/95
median wage, 1991/1992
share high-skilled, 1991/1992
share medium-skilled, 1991/1992
share unemployed, 1993
share social welfare recipients, 1995
number of inhabitants in thousands, 1991/92
share immigrants in workforce, 1991/92
0.061
(0.062)
-0.762
(0.142)**
-0.285
(0.106)**
0.003
-0.001
-0.005
(0.013)
0.003
(0.004)
0.033
(0.015)*
-0.019
(0.013)
0.001
(0.000)
0.003
(0.008)
0.581
(0.179)**
0.005
(0.003)
0.011
(0.021)
0.011
(0.009)
0.01
(0.023)
-0.002
(0.026)
0.000
(0.001)
0.016
(0.019)
Note: The table investigates the determinants of the intensity of the construction of new child care slots, by
regressing the change in the aggregate child care attendance rate in years in the municipality between children
who are examined in 2001 or 2002 and children who are examined in 1994 or 1995 on municipality baseline
characteristics, referring to (depending on data availability) the years 1991/92 to 1994/95, respectively. Results
are weighted with the number of native and immigrant children born January to June in the adminstrative data
from school entry examinations, averaged over the years 1994 and 1995. Standard errors are clustered at the
municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Child care attendance rates: Administrative Data from School Entry Examinations, Weser-Ems, 19942002. Median wage, share high- and medium-skilled: Social Security Records, Weser-Ems, 1991/92, 18-65 year
olds. Unemployment rate (from 1993) and share social welfare recipients (from 1995): Statistical Office of
Lower Saxony.
Table 5: The Impact of Child Care Attendance on Child Development: Baseline Results
Panel A: Children of native ancestry
language and speech
difficulties
(i)
OLS
-0.025
(0.003)**
(ii)
IV
0.022
(0.025)
fine motor skill gross motor
problems
skill problems
-0.029
-0.024
(0.004)**
(0.003)**
0.026
0.036
(0.036)
(0.032)
deferment
recommended
-0.051
(0.003)**
0.032
(0.018)
summary
measure
-0.036
(0.005)**
0.064
(0.043)
Panel B: Children of immigrant ancestry
fine motor skill gross motor
German training
problems
skill problems
(i)
OLS
-0.112
-0.022
-0.016
(0.008)**
(0.004)**
(0.004)**
(ii)
IV
-0.141
-0.084
-0.021
(0.030)**
(0.021)**
(0.018)
deferment
recommended
-0.079
(0.007)**
-0.084
(0.022)**
summary
measure
-0.092
(0.007)**
-0.177
(0.030)**
Note : The table reports, separately for children of native (Panel A) and children of immigrant ancestry
(Panel B) OLS (rows (i)) and IV estimates (rows (ii)) based on equation (1) for the impact of attending
one additional year in child care on developmental problems. Our instrument is the native- and immigrantspecific aggegate child care attendance rate in the municipality and school examination cohort. In addition
to municipality and school examination cohort fixed effects, we control for gender, age at examination in
months, month of birth and, for immigrant children, ethnicity. See Table 3, Panel A, for estimates of the
first stage. Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children born
between January and June.
Table 6a: The Impact of Child Care Attendance on Child Development: Robustness Checks
Children of native ancestry
(1)
language and speech difficulties
fine motor skill problems
gross motor skill problems
deferment recommended
summary measure (at least 1 problem)
N
small municipalities with
(3)
composition
municipalities with inflow
municipalities with increase
less than 100 native children
of migrants 2002-1991
in share of immigrants by
above median excluded
0.051
(0.028)
0.002
(0.043)
0.103
(0.037)**
0.039
(0.023)
0.086
(0.061)
46,994
5%-points excluded
0.008
(0.028)
0.056
(0.046)
0.079
(0.043)
0.019
(0.022)
0.088
(0.061)
54,420
excluded
0.016
(0.027)
0.03
(0.038)
0.04
(0.041)
0.028
(0.021)
0.066
(0.049)
70,446
movers excluded
0.023
(0.026)
0.028
(0.037)
0.036
(0.033)
0.036
(0.019)
0.071
(0.046)
75,781
(5)
(6)
plus municipality controls
language and speech difficulties
fine motor skill problems
gross motor skill problems
deferment recommended
summary measure (at least 1 problem)
N
(2)
0.014
(0.025)
0.02
(0.039)
0.025
(0.030)
0.024
(0.018)
0.03
(0.038)
78,133
(7)
random expansion
flexible time trend with
munipicality-specific
respect to baseline aggregate
linear time trend
-0.012
(0.029)
-0.088*
(0.045)
-0.002
(0.026)
0.017
(0.019)
-0.048
(0.042)
78,201
child care attendance rate
0.023
(0.025)
0.044
(0.038)
0.045
(0.031)
0.031
(0.016)*
0.079
(0.047)
77,541
(4)
(8)
flexible time trend with
respect to municipality
baseline characteristics
0.028
(0.032)
-0.057
(0.055)
0.022
(0.031)
0.011
(0.023)
0.011
(0.042)
78,201
Note : The table reports several robustness checks on our baseline specification for the impact of attending one additional year in child care on
developmental problems for native children. In column (1), we exclude small municipalities with less than 100 native children from our sample. In
columns (2) to (4), we test the robustness of our results to compositional changes by excluding from the sample children who moved to the municipality
within the last 2 years, municipalities with an increase in immigration (measured as the percentage point change in the share of immigrants in the
working population) above the median and municipalities with an increase in the share of immigrant children in the School Entry Examinations from one
year to the next of more than 5 percentage points.
In columns (5) to (8), we investigate the assumption that the construction of child care slots was random by first controlling for municipality
characteristics (median wage, share medium- and high-skilled, share foreign workers, number of inhabitants) referring to the years when the child is 2
and 5 years old, respectively. We then allow for a municipality-specific linear time trend (column (6)) as well as for a fully flexible time trend with
respect to the native-specific child care attendance rate in the base year (1994/95) (column (7)) and observable characteristics in the base year (column
(8)). In addition to municipality and school examination cohort effects, all regressions control for gender, age at examination in months, and month of
birth. The reported number of observations is for the outcome "deferment recommended" (and very similar to other outcomes). Standard errors are
clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children of native ancestry born between January and June.
Observable municipality characteristics: Social Security Records, Weser-Ems, 1990-2002, 18-65 year olds and Statistical Office of Lower Saxony.
Table 6b: The Impact of Child Care Attendance on Child Development: Robustness Checks
Children of immigrant ancestry
(1)
German training recommended
fine motor skill problems
gross motor skill problems
deferrment recommended
summary measure (at least 1
problem)
N
German training recommended
fine motor skill problems
gross motor skill problems
deferment recommended
summary measure (at least 1
problem)
N
(2)
small municipalities with
(3)
composition
municipalities with inflow
less than 20 immigrant
of migrants 2002-1991
(4)
children excluded
-0.156
(0.053)**
-0.105
(0.035)**
-0.02
(0.026)
-0.051
(0.032)
-0.168
(0.050)**
movers excluded
-0.134
(0.029)**
-0.098
(0.026)**
-0.022
(0.022)
-0.096
(0.024)**
-0.178
(0.034)**
above median excluded
-0.155
(0.044)**
-0.081
(0.031)**
-0.015
(0.027)
-0.076
(0.034)*
-0.18
(0.044)**
municipalities with
increase in share of
immigrants by 5%-points
excluded
-0.176
(0.040)**
-0.107
(0.030)**
-0.046
(0.026)
-0.119
(0.032)**
-0.172
(0.043)**
8,119
8,985
7,199
6,691
(5)
(6)
(7)
random expansion
flexible time trend with
respect to baseline
plus municipality controls
munipicalityaggregate child care
specific linear time
attendance rate
trend
-0.15
-0.176
-0.129
(0.030)**
(0.026)**
(0.029)**
-0.075
-0.083
-0.083
(0.022)**
(0.022)**
(0.022)**
-0.022
-0.026
-0.022
(0.019)
(0.019)
(0.019)
-0.092
-0.095
-0.081
(0.021)**
(0.021)**
(0.023)**
-0.183
-0.168
-0.151
(0.030)**
(0.032)**
(0.030)**
10,320
10,323
10,285
(8)
flexible time trend with
respect to municipality
baseline characteristics
-0.169
(0.029)**
-0.119
(0.029)**
-0.02
(0.019)
-0.099
(0.022)**
-0.203
(0.034)**
10,323
Note : The table reports several robustness checks on our baseline specification for the impact of attending one additional year in child
care on developmental problems for immigrant children. In column (1), we exclude small municipalities with less than 20 immigrant
children from our sample. In columns (2) to (4), we test the robustness of our results to compositional changes by excluding from the
sample children who moved to the municipality within the last 2 years, municipalities with an increase in immigration (measured as the
percentage point change in the share of immigrants in the working population) above the median and municipalities with an increase in
the share of immigrant children in the School Entry Examinations from one year to the next of more than 5 percentage points.
In columns (5) to (8), we investigate the assumption that the construction of child care slots was random by first controlling for
municipality characteristics (median wage, share medium- and high-skilled, share foreign workers, number of inhabitants) referring to the
years when the child is 2 and 5 years old, respectively. We then allow for a municipality-specific linear time trend (column (6)) as well as
for a fully flexible time trend with respect to the immigrant-specific child care attendance rate in the base year (1994/95) (column (7)) and
observable characteristics in the base year (column (8)). In addition to municipality and school examination cohort effects, all regressions
control for gender, age at examination in months, month of birth, and ethnicity. The reported number of observations is for the outcome
"deferment recommended" (and very similar to other outcomes). Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children of immigrant ancestry born between
January and June. Observable municipality characteristics: Social Security Records, Weser-Ems, 1990-2002, 18-65 year olds and
Statistical Office of Lower Saxony.
Table 7: Changes in Child Care Quality and Spillover Effects from immigrant and younger children
Panel A: Changes in the quality of child care (municipalities with rapid expansion excluded)
language and speech
difficulties/ German fine motor skill gross motor skill
problems
problems
training recommended
Children of native ancestry
0.012
0.03
0.062
(0.032)
(0.058)
(0.041)
Children of immigrant ancestry
-0.111
-0.115
-0.041
(0.039)**
(0.027)**
(0.022)
Panel B: Peer effects: increased exposure of native children to immigrant children
language and speech
difficulties/ German fine motor skill gross motor skill
problems
problems
training recommended
0.022
0.021
0.033
individual child care attendance in
(0.025)
(0.036)
(0.033)
years (IV)
aggregate child care attendance in
0.001
0.015
0.007
years of immigrant children (peer
effect)
(0.007)
(0.009)
Panel C: Peer effects: increased exposure to younger children
language and speech
difficulties/ German fine motor skill
problems
training recommended
Native children
0.009
0.025
individual child care attendance in
(0.029)
(0.039)
years (IV)
0.107
0.114
share of 3-year old children in cohort
(0.070)
(0.071)
t+1 (peer effect)
Immigrant children
-0.161
-0.096
individual child care attendance in
(0.031)**
(0.022)**
years (IV)
0.489
0.009
share of 3-year old children in cohort
(0.162)**
(0.121)
t+1 (peer effect)
deferment
recommended
0.009
(0.018)
-0.081
(0.025)**
summary measure
0.008
(0.027)
-0.196
(0.032)**
deferment
recommended
0.032
(0.019)
summary measure
0.057
(0.044)
0.001
0.019
(0.007)
(0.005)
(0.011)
gross motor skill
problems
deferment
recommended
summary measure
0.018
(0.029)
0.194
(0.049)**
0.029
(0.019)
0.014
(0.033)
0.055
(0.044)
0.125
(0.085)
-0.027
(0.019)
0.07
(0.100)
-0.091
(0.024)**
0.006
(0.111)
-0.195
(0.030)**
0.321
(0.146)*
Note: In Panel A, we investigate the robustness of our results to decreases in child care quality, by excluding municipalities for which the
expansion in child care slots was particularly rapid (i.e. municipalities with an increase in the aggregate child care attendance rate in years
between two school examination cohorts above 90th percentile). In Panel B, we investigate whether the increased attendance rate of
immigrant children had an effect on developmental problems of native children, by including the aggregate child care attendance rate in
years of immigrant children as an additional control variable in our baseline IV regression (see Table 5). In Panel C, we analyze whether
the increased attendance rate of three-year-old children affected the development of native and immigrant children, by including the share
of 3-year olds in child care in the municipality and the cohort that is examined one year later (cohort t+1) as an additional control variable
in our baseline IV regression.
In addition to municipality and school examination cohort fixed effects, all regressions control for gender, age at examination in months,
month of birth, and ethnicity. Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2003, children born between January and June.
Table 8: The Impact of Child Care Attendance on Child Development: Results by Gender and Ethnicity
Panel A: Children of native ancestry
language and speech fine motor skill gross motor skill
difficulties
problems
problems
girls
0.006
0.02
0.021
(0.029)
(0.029)
(0.027)
boys
0.006
0.03
0.05
(0.029)
(0.050)
(0.042)
deferment
recommended
0.009
(0.018)
0.053
(0.027)
summary
measure
0.027
(0.053)
0.096
(0.048)*
Panel B: Children of immigrant ancestry
German training
recommended
girls
-0.101
(0.039)*
boys
-0.163
(0.035)**
former Soviet Union
-0.121
(0.033)**
Turkish
-0.145
(0.068)*
deferment
recommended
-0.048
(0.031)
-0.108
(0.028)**
-0.023
(0.027)
-0.07
(0.059)
summary
measure
-0.126
(0.042)**
-0.203
(0.040)**
-0.186
(0.037)**
-0.155
(0.074)*
fine motor skill gross motor skill
problems
problems
-0.037
-0.014
(0.022)
(0.023)
-0.118
-0.03
(0.032)**
(0.027)
-0.082
-0.041
(0.030)**
(0.021)
-0.112
0.034
(0.064)*
(0.051)
Note: The table first reports, separately for native (Panel A) and immigrant children (Panel B), IV estimates of the
impact of attending one additional year in child care separately by gender, where we estimate our baseline
regression (see equation (1)) separately by gender and instrument individual child care attendance with the nativeand immigrant-specific overall aggregate child care attendance rate. For immigrant children, we also report IV
estimates separately for the two largest immigrant groups: children from the former Soviet Union and from Turkey,
by estimating our baseline regression (equation (1)) separately for these two groups and instrumenting individual
child care attendance with the overall immigrant-specific aggregate child care attendance rate. In addition to
municipality and school entrance examination cohort fixed effects, all regressions control for gender, age at
examination in months, month of birth, and (for immigrant children) ethnicity. Standard errors are clustered at the
municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children born between
January and June.
Table 9: Non-Linear Effects of Child Care Attendance on Child Development
Panel A: Children of native ancestry
(1)
weight given to
marginal return
(2)
language and
(3)
fine motor
(4)
gross motor
speech deficits skill problems skill problems
(5)
(6)
deferment
recommended
summary
measure
OLS
Y -Y
1
0
13.89%
-0.022
(0.010)*
-0.035
(0.009)**
-0.028
(0.012)*
-0.054
(0.011)**
-0.047
(0.013)**
2
1
32.13%
-0.033
(0.007)**
-0.045
(0.010)**
-0.028
(0.006)**
-0.083
(0.007)**
-0.041
(0.010)**
>2
2
53.97%
-0.022
(0.004)**
-0.022
(0.006)**
-0.024
(0.004)**
-0.035
(0.005)**
-0.033
(0.007)**
1
0
16.17%
2
1
-0.141
(0.117)
0.056
(0.073)
0.136
(0.129)
0.027
(0.094)
-0.066
(0.132)
0.057
(0.070)
0.037
(0.115)
-0.016
(0.044)
-0.078
(0.211)
0.116
(0.105)
>2
2
0.05
(0.046)
-0.009
(0.059)
0.059
(0.046)
0.063
(0.026)*
0.053
(0.076)
(2)
German training
(3)
fine motor
(4)
gross motor
(5)
(6)
deferment
recommended
summary
measure
Y -Y
Y -Y
IV
Y -Y
Y -Y
Y -Y
29.93%
53.90%
Panel B: Children of immigrant ancestry
(1)
weight given to
marginal return
recommended skill problems skill problems
OLS
Y -Y
1
0
23.39%
2
1
38.96%
Y -Y
>2
2
37.65%
Y -Y
1
0
2
>2
-0.147
(0.022)**
-0.146
(0.018)**
-0.07
(0.013)**
-0.03
(0.018)
-0.01
(0.010)
-0.031
(0.007)**
-0.009
(0.011)
-0.023
(0.010)*
-0.012
(0.007)
-0.121
(0.017)**
-0.065
(0.010)**
-0.082
(0.012)**
-0.127
(0.021)**
-0.106
(0.019)**
-0.068
(0.012)**
22.56%
-0.204
(0.068)**
0.045
(0.069)
-0.031
(0.066)
-0.14
(0.054)**
-0.247
(0.086)**
1
39.11%
2
38.32%
-0.191
(0.078)*
-0.092
(0.075)
-0.059
(0.065)
-0.165
(0.063)**
-0.015
(0.042)
-0.007
(0.056)
-0.099
(0.057)
-0.037
(0.059)
-0.143
(0.090)
-0.198
(0.080)*
Y -Y
IV
Y -Y
Y -Y
Note : The first column of the table reports for our baseline OLS and IV specification the weights given to each marginal return
to child care attendance in the OLS and IV estimates, separately for children of native (Panel A) and children of immigrant
ancestry (Panel B). See Section 4.2, equation (2) for a computation of the weights for the IV estimation, and footnote 28 for the
OLS estimation.
The table then reports, separately for native and immigrant children, OLS and IV estimates for the marginal return to attending
child care for 1 year versus 0 years, for 2 years versus 1 year, and for three or more years versus 2 years. We obtain the OLS
estimates by estimating equation (5). To obtain IV estimates, we instrument the dummy variables in equation (5) with the
aggregate share of native and immigrant children who attend child care for at least 1, 2 or 3 years in the municipality and school
examination cohort. In addition to municipality and examination cohort fixed effects, we control for gender, age at examination
in months, month of birth and (for immigrant children) ethnicity. Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Source : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002, children born between January and
June.
Figure 4: Evolution of Childcare Attendance and Maternal Labor Force Participation Rates by Age of Child
Note : The figure plots, for native children children and their mothers, the evolution of child care attendance rates and maternal
labor force participation rates in April each year, by the age of the child.
Source : Microcensus, West Germany, 1991, 1993, 1995-2001, native children (and their mothers), who are 3 or 4 years old, and
native mothers whose youngest child is 11 or 12 years old, respectively.
Table 10: Expansions in Child Care Coverage and Maternal Labor Force Participation Rates
Panel A: Child Care Attendance Rates and Mothers' Labor Force Participation
Child care attendance
Change 2001 vs
Level:
1993
1993
i) Native children
Mother working
Level:
1993
Change 2001 vs 1993
"Treatment"
3 and 4 year olds
0.52
0.19
0.38
0.086
.
.
0.63
0.073
0.012
(0.013)
"Control"
11 and 12 year olds
"DinD"
Child care attendance
Change 2000/2001
Level:
vs 1991/93
1991/93
ii) Immigrant children
Mother working
Change 2000/2001 vs
Level
1991/93
1991/93:
"Treatment"
3 and 4 year olds
0.41
0.23
0.26
0.028
.
.
0.49
0.038
-0.008
(0.029)
"Control"
11 and 12 year olds
"DinD"
Panel B: Does Formal Care Crowd Out Maternal or Alternative Care?
i) Native children
3 and 4 year olds, Change 2001 vs
1993
Change in share of women working and using child care
0.13
(replacement of alternative care)
0.06
Change in share of women not working and using child
care (replacement of maternal care)
3 and 4 year olds, Change
ii) Immigrant children
2000/2001 vs 1991/93
Change in share of women working and using child care
0.08
(replacement of alternative care)
0.15
Change in share of women not working and using child
care (replacement of maternal care)
Note: In Panel A, we first report for native children the child care attendance rates and maternal labor force participation
rates in 1993, as well the change in these rates between 2001 and 1993, for two groups of children and their mothers: 3and 4-year-old children for whom child care attendance rates rose sharply ("treatment") and for 11- and 12-year-old
children who have no younger siblings and who were unaffected by the expansion in child care slots ("control"). By
restricting the control group to 11- and 12-year-old children who have no younger siblings we ensure that their mothers
were not affected by the expansion in child care slots through the younger sibling. We also display the "DinD" estimate
which calculates the differential increase in employment rates between the treatment group over the control group. Here,
we condition on mother's age and an indicator for singleparenthood (results are very similar when controls ar omitted).
We then repeat the analysis for immigrant children and their mothers where, due to the small sample size, we combine
the years 1991/93 and 2000/2001.
In Panel B, we investigate whether formal child care replaces maternal or alternative care, by separately reporting the
change in the share of women who are using child care and who are or who are not working between 2001 and 1993
(native mothers) and 2000/2001 and 1991/93 immigrant mothers).
Source: Microcensus, 1991, 1993, 1995-2001, West Germany, children and their mothers, who are 3 or 4 years old and
mothers, whose youngest child is 11 or 12 years old. The sample consists of 22,153 native children and 5,396
immigrant children.
Table A1a: Randomness of Timing and Intensity of Construction of Child Care Slots: Children of native ancestry
aggregate
detailed aggregate attendance rates:
attendance rate in
years
share ≥ 1
share ≥ 2
share ≥ 3
Referring to age 5:
median wage
-0.001
0.000
0.001
-0.001
(0.002)
(0.000)
(0.001)
(0.001)
share high-skilled
-0.002
0.001
-0.009
0.002
(0.011)
(0.002)
(0.004)*
(0.008)
share medium-skilled
-0.002
0.001
-0.001
-0.002
(0.003)
(0.001)
(0.001)
(0.002)
number of inhabitants in thousands
-0.029
-0.002
-0.002
-0.010
(0.012)*
(0.002)
(0.005)
(0.006)
share immigrants in population
0.009
0.001
0.004
0.003
(0.009)
(0.001)
(0.002)
(0.005)
share social welfare recipients
0.009
0.003
0.004
0.003
(0.008)
(0.001)
(0.003)
(0.004)
share unemployed
-0.006
-0.001
-0.008
-0.004
(0.012)
(0.002)
(0.004)*
(0.007)
Referring to age 2:
median wage
0.000
-0.001
0.001
-0.001
(0.003)
(0.000)**
(0.001)
(0.002)
share high-skilled
0.011
-0.002
-0.007
0.013
(0.012)
(0.002)
(0.004)
(0.008)
share medium-skilled
-0.005
0.000
0.000
-0.003
(0.004)
(0.001)
(0.002)
(0.002)
number of inhabitants in thousands
-0.016
-0.001
-0.002
-0.001
(0.011)
(0.001)
(0.004)
(0.006)
share immigrants in population
0.000
0.001
0.001
0.000
(0.010)
(0.001)
(0.003)
(0.006)
share unemployed
0.012
0.005
0.001
-0.004
(0.011)
(0.002)**
(0.004)
(0.006)
Note : The table investigates the randomness of the construction of new child care slots, by regressing the aggregate child
care attendance rate for native children in the municipality and school examination cohort in years, as well as the share of
native children in the municipality and school examination cohort who attend child care for at least 1, 2, or 3 years on a
number of municipality characteristics, controlling for observable child characteristics (gender, age at examination and
month of birth) as well as municipality and school examination cohort fixed effects. Municipality characteristics refer to 1
and 4 years before the school examination when children were 5 and 2 years old. Standard errors are clustered at the
municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Sources : Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002. Median wage, share high- and
medium-skilled: Social Security Records, Weser-Ems, 1990-2002, 18-65 year olds. Unemployment rate (from 1993) and
share social welfare recipients (from 1995): Statistical Office of Lower Saxony.
Table A1b: Randomness of Timing and Intensity of Construction of Child Care Slots: Children of immigrant
ancestry
aggregate
detailed aggregate attendance rates:
attendance rate in
years
share ≥ 1
share ≥ 2
share ≥ 3
Referring to age 5:
median wage
0.005
-0.001
0.002
0.003
(0.003)
(0.001)
(0.001)
(0.002)*
share high-skilled
0.013
-0.001
-0.002
0.016
(0.030)
(0.010)
(0.013)
(0.015)
share medium-skilled
-0.01
-0.002
-0.006
-0.004
(0.008)
(0.003)
(0.004)
(0.004)
number of inhabitants in thousands
-0.032
-0.005
-0.011
-0.009
(0.013)*
(0.005)
(0.009)
(0.005)
share immigrants in workforce
-0.015
-0.002
-0.006
-0.004
(0.026)
(0.008)
(0.009)
(0.011)
share social welfare recipients
0.007
0.006
-0.001
0
(0.022)
(0.007)
(0.009)
(0.007)
share unemployed
0.059
0.001
0.031
0.016
(0.033)
(0.011)
(0.015)*
(0.013)
Referring to age 2:
median wage
0.006
-0.003
0.000
0.007
(0.006)
(0.002)
(0.003)
(0.002)**
share high-skilled
-0.033
-0.023
-0.012
-0.005
(0.043)
(0.014)
(0.014)
(0.020)
share medium-skilled
-0.005
0.004
-0.001
-0.008
(0.014)
(0.003)
(0.005)
(0.006)
number of inhabitants in thousands
-0.023
-0.006
-0.007
-0.003
(0.016)
(0.004)
(0.006)
(0.009)
share immigrants in workforce
-0.018
0.005
-0.005
-0.006
(0.026)
(0.005)
(0.011)
(0.013)
share unemployed
0.022
0.000
0.023
0.004
(0.033)
(0.007)
(0.016)
(0.015)
Note : The table investigates the randomness of the construction of new child care slots, by regressing the aggregate child
care attendance rate for immigrant children in the municipality and school examination cohort in years, as well as the
share of immigrant children in the municipality and school examination cohort who attend child care for at least 1, 2, or 3
years on a number of municipality characteristics, controlling for observable child characteristics (gender, age at
examination, month of birth, and ethnicity) as well as municipality and school examination cohort fixed effects.
Municipality characteristics refer to 1 and 4 years before the school examination when children were 5 and 2 years old.
Standard errors are clustered at the municipality level.
* indicates significance at a 5% level, ** indicates significance at a 1% level.
Sources: Administrative Data from School Entry Examinations, Weser-Ems, 1994-2002. Median wage, share high- and
medium-skilled: Social Security Records, Weser-Ems, 1990-2002, 18-65 year olds. Unemployment rate (from 1993) and
share social welfare recipients (from 1995): Statistical Office of Lower Saxony.