SlideShare a Scribd company logo
1 of 43
Maternal Labor Supply and the Availability of Public Pre-K: Evidence
from the Introduction of Prekindergarten into American Public Schools
Sean P. Sall
Department of Economics
University of Notre Dame
July 22, 2012
Abstract
In the 1980s and 1990s, many states and districts began to provide funding for
prekindergarten programs for the first time. This paper takes advantage of the staggered
timing in program funding to investigate the effect that increased availability of
prekindergarten programs has on the labor supply of mothers with four-year-olds. I find
that mothers with a four-year-old and no younger children were significantly more likely
to be in the labor force and employed once prekindergarten became available. Mothers
with a four-year-old and other younger children were also significantly more likely to be
in the labor force and employed.
All views expressed within this paper are those of the author, and any errors made are my own. I
would like to thank William Evans for a number of helpful suggestions. I would also like to
thank the sponsors of the annual Bernoulli Competition at Notre Dame, for which this paper was
written.
1
I. Introduction
In recent decades, there has been increasing evidence that prekindergarten programs (PK)
are effective at preparing children for formal education. Studies have demonstrated that Head
Start not only prepares students to be successful in school by improving cognitive skills (Currie
and Thomas 1995; Abbott-Shim et al., 2003; Puma et al., 2010), but also increases the likelihood
that participants graduate high school, attend college, and have higher earnings (Oden et al.,
2000; Garces et al., 2002). It is not surprising then that the fraction of three and four-year-olds in
such programs has grown substantially. The top line in Figure 1 reports PK enrollment rates for
three and four-year-olds from the October School Enrollment supplement over the 1970 through
2007 period.1
Note that over this period the fraction of children in this age group in PK
programs has more than doubled, from 20.5 percent in 1970 to 54.5 percent in 2007. Much of
the change in PK enrollment since 1980 is the result of larger enrollment in public school
programs. In Figure 1, I also graph the fraction of PK in private and public schools. Note that
between 1970 and 2007, PK enrollment in public schools has increased from 6.2 percent to 30.3
percent, while private has increased from 14.3 percent to 24.2 percent. The expansion in PK
enrollment is primarily due to an increase in the number of public schools offering PK programs.
In Figure 2, I graph the fraction of public elementary schools offering PK programs over this
time period. This data comes from the Common Core of Data, which contains enrollment
information for the universe of schools in the US. These figures indicate that over this period,
the fraction of public schools offering PK programs has increased by nearly a factor of 4, from
14.7 percent in 1987 to 56.1 percent in 2006.
While these trends in enrollment may have educational benefits for attendees, they may
1
These enrollment numbers include three and four-year-olds participating in Head Start.
2
also affect the labor supply of attendees’ parents. Childcare is potentially a large out-of-pocket
cost of participating in the labor force for parents with young children. Since the majority of
public schools offer PK at no or reduced costs, these programs greatly reduce the direct cost of
childcare for millions of families and hence greater availability of these programs should
increase labor supply among mothers with pre-school children.
The growing number of public prekindergarten programs in recent years coincides with
an increase in labor force participation among mothers with four-year-old children. Figure 3
reports the fraction of women that work whose youngest child is aged 2, 4, and 7. The data for
this graph is taken from the March Current Population survey. Note that from 1987 to 2006, as
the fraction of public schools offering prekindergarten programs increased dramatically, the
fraction of mothers whose child was four that were in the labor force increased by 12 percent,
from 62.8 to 70.3 percent. Note also that over this time period, the difference in labor force
participation between mothers of four-year-olds and two-year-olds grows, while the difference
between mothers of four-year-olds and seven-year-olds declines.
In this paper, I investigate the effect of increased accessibility to public prekindergarten
on the labor supply of mothers of four-year-olds by using variation in the introduction of
prekindergarten programs into American public schools. At the local level, most counties have
multiple school districts, but there are 13 states where schools are county-based. When counties
are identified in these states, we then know the student’s school district. Using data for these 13
states from the 5 percent Public Use Micro Sample (PUMS) and the American Community
Survey (ACS), as well as education data from the Common Core, I estimate a difference-in-
difference model that examines the change in labor supply of mothers with four-year-olds in
counties with an increasing supply of prekindergarten seats over time.
3
These reduced-form models aim to answer several questions: Does increased accessibility
to public prekindergarten programs impact the labor force participation of mothers with a four-
year-old as their youngest child? Is there heterogeneity in the labor supply response to public
prekindergarten? Which mothers are most impacted by the expansion of public prekindergarten
programs?
The results of this study indicate substantial effects of prekindergarten availability on the
labor supply of mothers of four-year-olds, both with and without younger children. For mothers
of four-year-olds in a district with no public PK, full implementation is associated with
statistically significant increases of 3.8 percentage points in labor force participation and 4.4
percentage points in employment. Estimates for married mothers were of larger magnitude and
statistically significant, while estimates for single mothers were not statistically significant.
The largest concern with the difference-in-difference model is that the number of
prekindergarten programs in a given county may be determined by trends in the labor supply of
mothers in that county. For example, if a county has a growing number of working mothers, the
district may respond to the demands of the constituents by providing more public PK programs.
The inclusion of county fixed-effects helps to minimize such a bias, but this will not eliminate
time-specific shocks that may contaminate a fixed-effect model. One way to test for the
presence of bias is to estimate separate labor supply models for mothers of two-year-olds and
seven-year-olds, who should not be impacted by increased provision of prekindergarten
programs. If estimates for mothers of four-year-olds are picking up spurious correlation between
increased PK availability and increased labor supply, estimates for these other mothers should
show similar correlations with pre-school availability. On the other hand, if estimates for
mothers of four-year-olds are not biased and are picking up the true effects of increased
4
provision of programs, then there should be no impact of PK availability on the labor supply of
these other mothers, which is exactly what I find.
This study makes several contributions to the existing literature on childcare costs and
maternal labor supply. Most importantly, this is one of two studies that examines the effect of
public prekindergarten programs on the labor supply of mothers within the United States, the
other performed by Fitzpatrick (2010). A number of authors have examined the impact of free
kindergarten on maternal labor supply, including papers by Gelbach (2002), Cascio (2006;
2009), and Fitzpatrick (2012). Meanwhile, Fitzpatrick (2010) is the only study of pre-
kindergarten aged children, examining the effects of universal PK programs in Georgia and
Oklahoma. This study employs a different econometric analysis than Fitzpatrick and a different
sample population. It can help to assess the costs and benefits of public prekindergarten
programs as well as the value of universal programs versus those targeted towards specific
populations. The results of this study are of particular interest to both state and local
governments that have recently decreased funding for public education.
II. Background and Existing Literature
There are a number of different child-care options for parents of three and four-year-olds.
Some are more academically focused, such as prekindergarten and preschool. The latter is
typically not affiliated with a school system. Nursery schools, which place less emphasis on
education, are also an option. Finally, there is an assortment of other daycare programs that do
not have an educational focus.
Before the introduction of prekindergarten programs into American public schools during
the 1960’s, the majority of three and four-year-olds attended day care programs – only ten
5
percent of three and four-year-olds were enrolled in any type of classroom (2003 State of
Preschool Yearbook). Today, over sixty percent of three and four-year-olds are enrolled in a
prekindergarten program, and more than half of these children are enrolled in a public program.
Children that live in areas where public programs are not available are typically enrolled in
private prekindergarten programs or nursery schools (2009 State of Preschool Yearbook).
The majority of states began funding for public prekindergarten programs as pilot or
experimental programs in the 1980’s or 1990’s, although initiatives varied substantially from
state to state. For example, Georgia’s prekindergarten program began as experimental in 1993
and was expanded to be universally available in 1995, while Arizona’s program began as a pilot
in 1991 and expanded as part of a block grant program in 1996. Additional differences in
programs include the population served. Some states, such as New York or Florida, focused on
universal availability, while states such as North Carolina and Texas focused upon at-risk
populations. Programs vary in terms of operating standards as well, although the majority of
programs are offered either full or part-day for four or five days a week (2009 State of Preschool
Yearbook). In any case, the free or reduced cost of these public programs lowers childcare costs
for mothers, which we would expect to increase maternal labor supply.
Existing studies offer a wide range of price elasticity estimates both for single and
married mothers. In their seminal paper on the subject, Blau and Robins (1988) estimated a
childcare price elasticity of labor supply of -0.38 for married mothers. Many studies soon
followed, all using similar estimation procedures but attaining widely different results. Blau and
Robins (1989) estimated elasticities of -0.47 and -0.77, Hotz and Killburn (1991) 0,
Michalopoulous et al. (1992) an elasticity of 0, Connelly (1992) -0.2, Ribar (1992) -0.74, Averett
et al. (1997) -0.78, and Kimmel (1998) -0.92. The results for single mothers are equally varied:
6
Connelly (1990) estimated an elasticity of 0, Kimmel (1995) -1.36, Kimmel (1995) -0.346 and
-0.345, Kimmel (1998) -0.22, and Connelly and Kimmel (2001) -1.0 and -0.8. Several
explanations for the discrepancies between estimates have been put forward, each identifying
potential sources of bias.
Averett et al. (1997) proposes that selection bias impacts many of these estimates.
Childcare costs are typically measured as childcare expenditures per household, presenting
problems when considering that women who are utilizing childcare are more likely to work.
Blau and Hagy (1998) suggests that the variability in estimates is due to omitted variables bias,
where the authors do not accurately control for the quality of childcare. Although many authors
use controls to minimize these potential sources of bias, the fact is that these models explain a
small fraction of the variation in labor supply, raising concerns that many of these studies suffer
from an inability to adequately control for these factors.
One way of sidestepping the problem of selection bias is to utilize quasi-experimental
variation in childcare availability or prices. Berger and Black (1992) were the first to use a
difference-in-difference model to examine the relationship between childcare costs and maternal
labor supply. Focusing on two subsidy programs in Kentucky, they compared the outcomes of
mothers before and after receipt of childcare subsidies to the outcomes for mothers over the same
time period that were eligible but did not receive a subsidy. Berger and Black found that the
program increased the probability of employment by 12 percent and increased labor force
participation by 25.3 percent. Since parents chose the childcare they purchased, however, there
is still concern whether the study effectively controls for quality.
Examining the labor supply consequences of public schooling programs has the
advantage that many programs have a 100 percent subsidy. These programs also reduce
7
concerns about appropriately controlling for quality of childcare, since each mother within the
same school catchment area is offered a program of the same quality. Despite the advantages of
these programs, there have only been five studies using US data that have examined the impact
of these programs on labor supply: one by Gelbach (2002), two by Cascio (2006; 2009), and two
by Fitzpatrick (2010; 2012).
Gelbach (2002) estimated a two-stage least squares model using quarter of birth as an
instrumental variable for kindergarten enrollment. For single mothers of a five-year-old and no
younger children, Gelbach found that public school enrollment lead to statistically significant
increases of 14.5 percent in weeks of work per year, 9.6 percent in usual hours of work per week,
14.5 percent in hours worked last week, 21.5 percent in wage and salary income, and 9.7 percent
in employment. He also found an 11.2 percent reduced probability of welfare receipt conditional
on public school enrollment. Cascio (2006) estimated a difference-in-difference model using
increased availability to public kindergarten as the treatment. For single mothers with a five-
year-old and no younger children, Cascio found statistically significant increases in maternal
labor supply: 13 percent in the likelihood of employment and 14.7 percent in hours worked per
week. No statistically significant results were found for single women with a five-year-old and
other younger children, or for married women with or without other younger children. In 2009,
Cascio followed the same estimation procedure but employed a different variable of interest.2
The results obtained were similar in magnitude to those in her earlier paper, with statistically
significant increases in labor supply variables for single mothers with no younger children and
statistically insignificant results for other mothers. Fitzpatrick (2012) used public school
enrollment as an instrumental variable in a regression discontinuity framework.3
She found that
2
The independent variable in this model was the fraction of the two years prior to the 2000 Census that kindergarten funding was
available in the state in which the mother lived.
3
See the following paragraph for a more complete discussion of this estimation procedure.
8
enrollment statistically significantly increased the employment of single mothers with no
younger children, with no statistically significant results for other mothers.
Fitzpatrick (2010) was the first to examine how availability of public prekindergarten
programs impacts maternal labor supply within the United States. Fitzpatrick utilized a
regression discontinuity design to exploit the fact that a child’s date of birth determines whether
or not he/she can enroll in PK for a given year. Using restricted access data from the 2000
Decennial Census, her sample was limited to mothers of four-year-olds in Oklahoma and
Georgia. The child’s date of birth was used to measure intention-to-treat effects, where those
children born before September 1st
, 1995 were eligible for public PK in the 1999-2000 school
year and those who were born on or after September 2nd
, 1995 were not eligible. Those mothers
with four-year-olds that were eligible were in the treatment group, while those mothers with
four-year-olds that were not eligible were in the comparison group.4
Samples included a full
sample of mothers of four-year-olds, as well as subsamples based on marital status and the
presence of younger children. The results indicated that the availability of universal PK had
small and statistically insignificant effects on the mother’s likelihood of employment in the
previous year and weeks or hours worked in the previous year, regardless of marital status or
whether the mother had younger children. It is important to note that the standard errors on the
estimates in this work are quite large.5
The use of difference-in-difference models to analyze the labor supply of mothers in
foreign countries also reports large impacts of public preschool programs. Schlosser (2005)
studied free public preschool and its effects on the labor supply of Israeli Arab mothers with
4
Only those mothers with a child born within 100 days before or after the cutoff were considered.
5
The standard errors are larger than the parameter estimates for all but a few of the dependent variables, and in many cases,
several times larger than the parameter estimates.
9
children aged two to four, Berlinkski and Galiani (2007) increased construction of preprimary
schools and its effect on the labor supply of mothers in Argentina, and Baker et al. (2008) the
effect that extension of full-time kindergarten programs had on mothers in Quebec. While the
magnitude and precision of the estimates within these studies is varied, each finds evidence that
increases in the supply of public preschool programs has positive effects on maternal labor
supply, both for married mothers (Schlosser, 2005; Baker et al., 2008) and single mothers
(Berlinkski and Galiani, 2007).
III. Data
A. Sample and Key Variables
Sample data for this analysis is taken from the five-percent samples of the 1990 and 2000
Census, as well as the 2005 and 2006 American Community Survey. In order to accurately
answer the question posed above, it is necessary to correctly assign each mother to the school
district in which she enrolls her child. While school district is not reported within the data, states
that use county-based school districts offer an opportunity to sidestep this limitation. Thirteen
states use county-based school districts, and in these states, the county the mother resides in also
determines the school district in which she enrolls her child. Although county is not reported in
the data, the PUMA number for each mother is reported, and the counties contained within a
given PUMA are also reported. This information allows for restriction of the data that matches
mothers with school districts.
A public use microdata area (PUMA) is defined as an area with a decennial census
population of 100,000 people or more. Thus, the number of counties that are contained within a
PUMA is dependent upon the size of these counties. Small counties that have populations less
10
than 100,000 people must be aggregated with other counties to form a PUMA. Counties that
have populations larger than 100,000 will be uniquely identified by one PUMA, or possibly more
depending upon the size of the county.6
For small counties that are aggregated with others into
one PUMA, it is impossible to match mothers with school districts. PUMAs that consist of only
one county do allow matching of mothers with school districts, since it is known what county the
mother resides in. Therefore, after limiting the data to those thirteen states with county-based
school districts, I then limit it to those PUMAs that are consistent of a single county. The final
step in this process is the aggregation of those counties that are large enough to be designated by
more than one PUMA (population of 200,000 or more). A total of fifty-nine districts in ten
states are used in this study. Table 1 shows the names of these districts (see Appendix 1 for a
more detailed explanation of the restriction process).
Because this is a geographically limited sample, there are concerns that the results of this
study may not be indicative of how pre-kindergarten affects mothers’ labor supply in other parts
of the United States. Table 2 reports the demographic characteristics of mothers of four-year-olds
based on inclusion in the sample. The statistically significant differences in the means for
demographic characteristics calls for caution in the interpretation of how pre-kindergarten affects
mothers’ labor supply in other parts of the United States. For each of the labor supply measures,
however, there are no statistically significant differences in means, suggesting that in terms of
labor supply, mothers in my sample are not different than mothers in the rest of the country.
My primary sample within the county-based school districts includes mothers of four-
year-olds, both with and without younger children. I create similar samples of mothers of two
and seven-year-olds to serve as comparison groups. Subpopulations that are used in the analysis
6
For example, a county with more than 200,000 people but less than 300,000 people may be uniquely identified by two PUMAs,
a county with more than 300,000 people but less than 400,000 uniquely identified by three PUMAs, etc.
11
include both married and single mothers of two, four, and seven-year-olds. All samples are
restricted to include mothers aged 18-45 at the time of the Census.7
The key outcomes in this paper are two measures of employment that are observed in
both the Census and American Community Survey for all years: an indicator for whether the
mother is employed as well as one for whether the mom is in the labor force. In addition to the
variable of interest, my regression includes controls for maternal race and ethnicity (indicators
for black, white, Hispanic, and other race8
), age and number of children (ages zero to one, two,
three, four, five, six, seven, eight to twelve, thirteen to seventeen, and eighteen and older), a
quadratic in maternal age, education level, indicators for veteran and marital status, and an
indicator for whether or not the mom lives in a large city (coded as a city population greater than
500,000).
B. Summary Statistics
Table 3 reports demographic characteristics of mothers of two, four, and seven-year olds with no
younger children. For the four employment outcomes observed in each year – usual weeks
worked, weeks worked last year, employment (an indicator), and labor force participation (an
indicator) – there are systematic differences between mothers of two, four, and seven-year-olds.
Mothers of seven-year-olds consistently provide the highest levels of labor, followed by mothers
of four-year-olds, with mothers of two-year-olds showing the lowest levels of labor supply.
Compared to mothers of four-year-olds with no younger children, mothers of four-year-olds with
younger children are on average younger and report lower levels of labor supply for each of the
employment outcomes, likely because of the presence of younger children.
7
All samples were constructed without using sampling weights.
8
Indicators for black, white, and other race refer only to those mothers that are not of Hispanic origin.
12
IV. Identification Strategy
A. State Prekindergarten Funding Initiatives
The large variation in state and district pre-kindergarten initiatives means that there are
large differences in availability not only between states, but within states over time as well.
Figure 4 presents the distribution of the change in the percentage of public schools within a
district that have prekindergarten from 1990 to 2006 for all districts in the United States and for
only those districts in my sample. For example, for a district with five schools, none of which
offered prekindergarten in 1990 and three of which offered pre-kindergarten in 2006, the
percentage change would be 60 percentage points. This figure shows there is substantial
variation in the growth in availability of prekindergarten, both nationally and in my subsample of
districts. However, while the distribution of the growth for all districts within the United States
is clustered around 0 and 1, it is more evenly spread for my subsample of districts.
My empirical strategy takes advantage of the differences in the growth in availability of
prekindergarten for those districts in my sample. My variable of interest is the percentage of
public schools that offer a prekindergarten program within a given district. I estimate the
following difference-in-difference model:
(1) yidt = θ percent_pkdt + Xidt β + αd + γt + εidt,
where yidt is some measure of employment for mother i in district d in year t; Xidt is the
aforementioned vector of maternal controls; and αd and γt represent district and year fixed-
effects, respectively. The district fixed-effect αd controls for the time-invariant differences in
13
maternal employment across districts, whereas the time effect γt captures changes in maternal
labor supply over time that impact all districts, such as changes in the aggregate economy,
federal programs, etc. The covariate of interest is percent_pkdt, which measures the percentage of
public schools offering prekindergarten programs within a district. For the initial set of
regressions, percent_pkdt is not weighted by the number of students or seats offered. The
parameter of interest, θ, represents the change in some measure of maternal labor supply given a
percentage point increase in the number of public schools offering prekindergarten within a
district.
My goal is to estimate θ for mothers whose youngest child is four-years-old, since this is
the most appropriate age group for prekindergarten.9
One problem that could lead to biased
estimates is differential trends in labor supply between mothers who live in districts that adopted
public prekindergarten programs and those that didn’t. For example, the adoption of a public
prekindergarten program could be correlated with the level of labor supply of mothers in a
district, such that districts with low levels of maternal labor supply are more likely to adopt PK
programs.10
If this were the case, then the estimates would overstate the true effect of increasing
availability to public PK. To control for any bias caused by differential trends, I estimate
difference-in-difference models that include mothers of two and seven-year-olds, who may be
subject to similar trends but unaffected by the introduction of prekindergarten programs.
9
Although it is true that prekindergarten initiatives within some states are inclusive of younger age groups, first priority is given
to four-year-olds. All available program spots are first filled by four-year-old children, and any remaining spots are open to three-
year-old children.
10
To investigate this possibility, separate regressions were run for the change in the percentage of prekindergarten within a
district from 1990 to 2006 on the initial levels of each of the labor supply variables. None of the coefficients were statistically
significant for any of the samples of mothers of four-year-olds used in this study, with the exception of the coefficient on usual
hours worked per week (0.0003) for the sample of any mothers of four-year-olds, which was marginally significant with a p-
value of 0.068. Taken as a whole, these results suggest that PK initiatives are on average not correlated with initial levels of
labor supply of the samples of mothers of four-year-olds used in this study.
14
V. Results
A. Effects on Public/Private Enrollment of Three and Four-Year-Olds
Table 4 presents the coefficients for the difference-in-difference model (1) estimated with
enrollment as the dependent variable.11
These estimates measure the effects of increasing
availability to public PK on the probability of overall enrollment for three and four-year-olds, as
well as public and private enrollment. Column 1 reports that a movement from no availability of
public PK to full implementation is associated with an increase of 1.9 percentage points in the
probability of overall enrollment for three-year-olds, and an increase of 4.8 percentage points for
four-year-olds. The coefficient for three-year-olds is not statistically significant, while the
coefficient for four-year-olds is statistically significant at the 10 percent level. Column 2 shows
that for three-year-olds, a movement from no availability of public PK to full implementation is
associated with a 2.4 percentage point increase in the probability of public enrollment, and for
four-year-olds a 6.0 percentage point increase in the probability of public enrollment. The
estimate for three-year-olds is not statistically significant, while the estimate for four-year-olds is
statistically significant at the 5 percent level. The coefficients for the effects on the probability of
private enrollment, -0.005 for three-year-olds and -0.013 for four-year-olds, are not statistically
significant from zero for both samples.
These estimates imply large effects on the probability of enrollment in public PK for
four-year-olds, with smaller effects for three-year-olds. For the sample of four-year-olds, which
reported a 43 percent enrollment rate in public PK in 1990, the 6.0 percentage point increase in
the probability of public enrollment translates to a 14.0 percent (6.0/43.0) increase in the
probability of public PK enrollment. For three-year-olds, whose reported enrollment in public
11
Enrollment is coded as a dummy variable, which equals one if the child is enrolled in school. The samples of three and four-
year-olds used for these regressions were those from samples of mothers with three and four-year-olds and no younger children.
15
PK was 26.1 percent in 1990, the 2.4 percentage point increase means a 9.2 percent (2.4/26.1)
increase in the probability of public PK enrollment. The smaller, statistically insignificant
effects of public PK on public enrollment of three-year-olds is as expected, since the majority of
state initiatives allowed three-year-olds to enroll only after all four-year-olds were offered seats.
In terms of private PK enrollment, the estimates suggest that increasing availability of public PK
has no statistically significant effect on private enrollment for the sample of three or four-year-
olds. These results suggest that there was little to no substitution between public and private PK
programs, and that the increases in public PK enrollment were primarily the result of new
enrollees.
These findings are similar to those of Cascio (2006; 2009), who found large effects of
increasing preschool availability on public school enrollment rates of five-year-olds, and
Fitzpatrick (2010), who found large effects of increasing availability on overall enrollment rates
of four-year-olds. Cascio (2006) found that a movement from no availability of kindergarten to
full implementation was associated with an increase of 18.9 percentage points in the probability
of public enrollment for five-year-olds of single mothers with no younger children and a 23.3
percentage point increase for five-year-olds of married mothers with no younger children.
Neither of these estimates were statistically significant at traditional levels of significance.
Cascio (2009) found a 15.2 percentage point increase in the probability of public enrollment for
five-year-olds of single mothers with no younger children and an increase of 14.5 percentage
points for five-year-olds of married mothers with no younger children, significant at the 5 and 1
percent levels, respectively. Fitzpatrick (2010) found that a movement from no availability of
public PK to full implementation is associated with an increase of 7.2 percentage points in the
probability of overall preschool enrollment for four-year-olds, significant at the 1 percent level.
16
B. Effects for Mothers of Four-Year-Olds with No Younger Children
Table 5 presents the effects of prekindergarten availability on mothers of two, four, and
seven-year olds with no younger children. The first set of estimates reported are for the full
sample of mothers.12
Regressions were run for two sets of mothers of four-year-olds: those with
no younger children, and those with no younger children and no seven-year-old.13
The results for
both of these samples are reported in Columns 1 and 2 of Table 5, respectively. The results for
these samples are effectively the same, and thus I will hereafter only discuss those for the former
sample, mothers of four-year-olds with no younger children (Column 1). For the full sample of
mothers with a four-year-old and no younger children, the estimates imply that movement from
no availability to full implementation of public PK is associated with a 3.8 percentage point
increase in the probability of being in the labor force and a 4.4 percentage point increase in the
likelihood of employment. Both of these coefficients are statistically significant at the 1 percent
level.
The second set of estimates reported in Table 5 are for samples of married mothers. The
estimates for married mothers are larger in magnitude than the estimates for the full sample of
mothers. The estimates for married mothers of four-year-olds with no younger children imply a
4.3 percentage point increase in the probability of labor force participation and a 5.5 percentage
point increase in the probability of employment associated with movement from no availability
to full implementation of public PK. The coefficient on labor force participation is statistically
significant at the 5 percent level, while the coefficient on employment is statistically significant
12
This sample includes both single and married mothers, as well as those that are in neither group. Single mothers are considered
to be those that are not married and are the head householder, which leaves mothers that are not married but not a head
householder in neither the married nor single group.
13
Both samples are samples of mothers of four-year-olds with no younger children (one, two, or three-year-olds). The latter
sample further excludes mothers that also have a seven-year-old child.
17
at the 1 percent level. The third set of estimates in Table 5 are for single mothers of four-year-
olds with no younger children. These estimates for single moms show large but statistically
insignificant effects for increasing availability of public PK – 4.6 percentage points for the
probability of labor force participation and 2.1 percentage points for the probability of
employment.14
These estimates imply large effects of prekindergarten supply on the labor supply of
mothers of a four-year-old with no younger children. For the full sample of mothers, the
percentage point increases in labor supply translate to increases of 5.4 percent (3.8/71.0) in the
probability of being in the labor force and 6.7 percent (4.4/66.2) in the probability of being
employed. These effects are largest for married mothers, who show increases of 6.3 percent
(4.3/68.6) in the probability of labor force participation and 8.4 percent (5.5/65.8) in the
probability of employment.15
These findings are similar to those of Cascio (2006; 2009), Gelbach (2002), and
Fitzpatrick (2012), who found large effects of kindergarten programs on single and married
mothers of five-year-olds with no younger children. Cascio (2006; 2009) found that the full
sample of mothers showed the smallest response. Although each found that single mothers
showed larger response to kindergarten availability than married mothers, this study found the
opposite with respect to prekindergarten availability. We should caution, however, that because
of the small sample size of single mothers and the subsequent lack of precision with regards to
the estimates, it is difficult to draw firm conclusions about which group is more responsive. In
fact, if we conduct a t-test for the difference in parameter estimates between married and single
14
In a separate set of regressions, the same model was employed with the addition of state-fixed time effects. While the point
estimates from these regressions are similar to those reported in Table 6, the standard errors are considerably larger. A table with
these results is available upon request.
15
As before, these percentage increases were calculated as the percentage point change divided by the initial levels of labor
supply, reported in 1990. For example, the percentage increase in the probability of labor force participation for the full sample of
mothers was calculated as the 3.8 percentage point increase divided by the 71 percent labor force participation rate in 1990.
18
mothers of four-year-olds with no younger (Column 1), the test-statistics for labor force
participation (-0.066) and employment (0.707) are not statistically significant at traditional levels
of significance.16
This suggests that the coefficient estimates for married and single mothers, and
therefore their responses to increasing PK, are not significantly different.
Fitzpatrick (2010), on the other hand, found that universal availability to PK resulted in
statistically insignificant responses in maternal labor supply when she employed restricted-access
data from the 2000 Census for only those mothers in Georgia and Oklahoma. However, since
the standard errors on most coefficients in that work are larger than parameter estimates, there is
a concern that the study was underpowered. Gelbach (2002) used restricted-access data from the
1980 Census for all fifty states, while Cascio (2006; 2009) used data from the 1950 to 1990
Census for twenty-four different states. Finally, Fitzpatrick (2012) used restricted-access data
from the 2000 Census for all fifty-states. In each of these studies, the sample sizes were much
larger than in Fitzpatrick (2010).
For the full sample of mothers of four-year-olds with no younger children, Fitzpatrick
(2010) found that a movement from no availability of public prekindergarten to full
implementation decreased the probability of employment in the previous year by 0.5 percentage
points, with a standard error of 1.1 percentage points.17
Thus, the 95 percent confidence interval
around the estimate is -1.7 to 2.7 percentage points. While my estimate of 4.4 percentage points
does not fall within this confidence interval, the confidence interval constructed around my
estimate, 1.6 to 7.2 percentage points, does intersect hers. Therefore, my results are not
necessarily in conflict with those of Fitzpatrick (2010).
16
These were calculated under the assumption that the covariance between the two estimates is 0. Therefore, the test-statistic is
simply the difference in the coefficients divided by the square root of the sum of the variances of the estimates.
17
It is worth noting that the employment variable employed in this study was employment at the time of the Census.
Employment in the previous year is the variable in Fitzpatrick (2010) that is most comparable to any in this study.
19
Even in the case that the insignificant results of Fitzpatrick (2010) cannot be explained by
the small sample size and subsequently large standard errors, it is not surprising that they are
different from those of this study. Fitzpatrick (2010) examined the effects of universal
prekindergarten availability, focusing upon mothers in Georgia and Oklahoma. This study
focuses upon mothers in ten different states, nine of which do not support universal PK
initiatives. Georgia is the only state within my sample that aims for universal PK, while the rest
focus upon at-risk populations (State of Preschool Yearbook 2003). Thus, in the case that
universal prekindergarten initiatives have different effects than programs aimed towards at-risk
populations, we would expect different results from this study when compared to the results of
Fitzpatrick (2010).18
Finally, it important to note that there is little concern that the responses of mothers of
four-year-olds with no younger children are due to initial differences in levels of maternal labor
supply across counties, since the estimates for all the samples of mothers of two and seven-year-
olds are not significantly different from zero or are significantly negatively different from zero.
If anything, these results suggest that the DD estimates for mothers of four-year-olds and no
younger children understate the true effect of increasing availability of PK on labor supply. If
we conduct t-tests for the difference in parameter estimates between the coefficients for the
sample of mothers of four-year-olds with no younger (Column 1) and the samples of mothers of
two and seven-year-olds with no younger, the majority of the test-statistics are greater than 1.96.
This implies that the parameter estimates for mothers of four-year-olds are significantly larger
than those for mothers of two and seven-year-olds at the 5 percent level19
, and that the labor
supply responses for those mothers of four-year-olds are statistically significantly larger.
18
In fact, in a separate set of regressions, I restricted my sample to only those mothers in Georgia, and found insignificant effects
of universal PK on the full sample of mothers of four-year-olds and each one of the subsamples employed in this study.
19
These were again calculated under the assumption that the covariance between the two estimates is 0.
20
C. Effects for Mothers of Four-Year-Olds with Younger Children
Table 6 reports the estimates of the effects of prekindergarten availability on mothers of
four-year-olds with younger children.20
The first set of estimates in Column 2 for the full sample
of mothers show statistically significant increases of 3.6 percentage points in the probability of
labor force participation and 4.2 percentage points in the probability of employment associated
with a movement from no availability of prekindergarten to full implementation. The coefficient
on labor force participation is statistically significant at the 5 percent level and the coefficient on
employment is statistically significant at the 1 percent level. The second set of estimates for
married mothers are similar in magnitude but less precise. The coefficients in Column 2 imply a
3.6 percentage point increase in the probability of labor force participation and a 4.1 percentage
point increase in the probability of employment. The estimate for labor force participation is
statistically significant at the 10 percent level, while the estimate for employment is statistically
significant at the 5 percent level. The third set of estimates for single mothers are large but
statistically insignificant. Column 2 reports coefficients of 4.1 percentage points for labor force
participation and 3.5 percentage points for employment for single mothers of four-year-olds with
younger children. Again, the test-statistics for the difference in parameter estimates between
married and single moms for labor force participation (-0.089) and employment (0.096) suggest
that the labor supply responses for these mothers are not statistically significantly different.
These estimates imply large effects of prekindergarten programs on the labor supply of
mothers with a four-year-old and other younger children. These results are similar to those of
Cascio (2006; 2009), who found that the samples of mothers of five-year-olds with younger
children exhibited a large response to increased kindergarten availability. These results are
20
Note that the first column is for reference – it is simply the results reported in the first column of table 5.
21
opposite of those of Fitzpatrick (2010) for mothers of four-year-olds with younger children,
possibly for the same reasons discussed above.
Just as with the samples of mothers of four-year-olds and no younger children, there is
little concern that these estimated increases in labor supply are due to differences in the initial
levels of maternal labor supply across counties, since the estimates for mothers of two and seven-
year-olds with younger children are either not statistically significant from zero or are
statistically negatively significant from zero. Again, this suggests that the difference-in-
difference estimates for mothers of four-year-olds with younger children may understate the true
effect of increasing availability of PK on labor supply.
D. Placing these Results in Context
Some of the previous literature surrounding maternal labor supply and public schooling
focuses upon elasticity estimates, calculating the elasticity of a mother’s labor supply with
respect to child enrollment. With a little work, the coefficients in this paper can be transformed
into elasticity estimates. The regressions in Table 4 examine the effects of increasing availability
to public PK on enrollment of four-year-olds, whereas the regression in Table 5 examines the
effects of increasing availability to public PK on the labor supply of mothers. This first set of
regressions is simply a first-stage estimation for the regression of maternal labor supply on child
enrollment, and thus we can find the elasticity estimate by dividing the coefficient for the effect
on employment by the coefficient for the effect on enrollment. Thus, the child enrollment
elasticity of employment for the full sample of mothers of four-year-olds with no younger
children (Column 1 of Table 6) is 0.73 (0.044/0.060). In terms of child enrollment, this implies
that for every ten children that are enrolled in public PK, at least seven mothers of four-year-olds
22
with no younger children enter the workforce. Cascio (2009) found a similar but smaller effect
for kindergarten enrollment – for every ten children enrolled in public kindergarten, at least 3
mothers of five-year-olds entered the workforce.
E. Extensions
Throughout this paper, I have presented estimates using a measure of the percentage of
public schools in a district offering prekindergarten that is unweighted, either by the number of
students within the school or the number of PK seats offered. This is an important distinction,
because what really matters is how many mothers in a school catchment area have public PK
available to them, which is identified by weighting this measure. In an alternative specification,
I weighted each school by the number of kindergarten students in the school, where the number
of kindergarten students served as a proxy for the number of PK seats available. The number of
kindergarten students was used in place of the number of prekindergarten students because in
many schools, prekindergarten programs were offered but no students enrolled. Estimates using
this weighted measure are effectively the same as those using the unweighted measure.21
Throughout this paper, I have also operated under the assumption that family size is
exogenous to the model. While it might be unlikely, it is possible that the opportunity for a
mother to place her child in public PK could be the difference that leads to an additional birth.
To test this hypothesis, I estimated model (1) without controls for the number and age of other
children in the household. Estimates for these regressions are very similar to the model in which
controls were included.22
One of the questions I initially set out to answer was whether the labor supply response to
21
Results are also robust to the specification of a binary variable for any PK or none within a district.
22
Results were also robust to the specification in which the remainder of maternal covariates—race and ethnicity, age, marital
and veteran status, and whether the mom lives in a large city—were removed from the model.
23
increasing PK was the same for all mothers, or whether there was heterogeneity among different
samples of mothers. While we have seen heterogeneity in labor supply response based off
marital status, one could also imagine that a mother’s labor supply response would differ at
differing income levels. Since income is an outcome of the model and is thus endogenous, it is
not appropriate to separate mothers into samples based off income level. However, we can allow
education level to serve as a proxy for income level, and separate mothers based off educational
attainment. My final set of regressions estimates model (1) for samples of low-educated and
high-educated mothers, where low-educated mothers are considered to be those that have less
than a college degree and high-educated mothers are those with a college degree. For all but one
of these samples, the coefficients are not statistically significantly different from zero. The only
exception is for the coefficient on employment for the low-educated sample of mothers of four-
year-olds with no younger and no seven-year-old, which is a 4.5 percentage point increase,
significant at the 10 percent level. In general, the estimates are larger for high-educated mothers,
suggesting that highly educated mothers of four-year-olds show greater response to increasing
PK than low-educated mothers.23
While it might be expected that low-educated mothers would show a greater response to
increasing PK, there are several explanations for why the results suggest that highly educated
mothers show a greater response. On average, highly educated mothers are more likely to be
married. The presence of the spouse as a potential caregiver might leave the highly educated
mother less constrained in her labor supply choice, giving her a greater margin to respond to
increasing PK. Alternatively, it is possible that highly educated mothers are more inclined to
return back to work after giving birth.
23
The results for all regressions discussed within this section are available upon request.
24
VI. Conclusions
The results of this study are the first to report a statistically significant impact of
increasing public PK availability on the labor supply of mothers of four-year-olds. They are
different from the results obtained by Fitzpatrick (2010) in her study of mothers of four-year-
olds, although I have given reasons why this inconsistency might exist. More importantly is that
these results are consistent with the general body of literature surrounding the effects of public
schooling on maternal labor supply. Gelbach (2002) and Cascio (2006; 2009) found large effects
of kindergarten supply on the labor supply of married mothers of five-year-olds, which this study
replicates for prekindergarten and married mothers of four-year-olds. This study finds that
introducing pre-kindergarten into a school district is associated with a 3.8 percentage point
increase in labor force participation and a 4.4 percentage point increase in employment by
mothers of four-year-olds. These studies, along with Fitzpatrick (2012), also found large effects
of kindergarten supply on the labor supply of single mothers of five-year-olds, which this study
replicates for prekindergarten supply and single mothers of four-year-olds.
The results from this study, combined with those of these previous studies, provide key
information for a cost-benefit analysis of public preschool programs. This study, along with
those of Gelbach (2002), Cascio (2006; 2009), and Fitzpatrick (2012), suggest that public
preschool programs benefit both attendees and their mothers. The results from Fitzpatrick
(2010) further imply that programs aimed towards at-risk populations may be more effective
than universal preschool initiatives. The geographic limitations of both this study and Fitzpatrick
(2010) call for caution in extrapolating the results of this study or other studies into broader
contexts. The general body of literature implies that the benefits of public preschool, including
public PK, outweigh the costs. As many public prekindergarten programs are still relatively new
25
and are continuing to expand, however, future research is warranted in order to reaffirm these
positive effects.
Appendix 1
Restriction of Census and ACS Data
Data was taken from the 5 percent samples of the 1990 and 2000 Census as well as the
2005 and 2006 American Community Survey, and was downloaded from IPUMS. The original
data contained roughly thirty-three million observations. The process of constructing samples
began with the dropping of all states that did not have county-based school districts, since school
district is not reported in the Census or ACS. States that have county-based school districts were
kept because mothers can more accurately be matched to school districts, since county is
indicative of school district in these states. While county is not reported directly in the Census or
ACS, both the PUMA number and the names of each county identified within the PUMA are
reported. Thus, the next step included an examination of PUMA boundaries in 1990 as well as
26
2000 (PUMA boundaries were the same in 2000, 2005, and 2006) to identify what counties were
located within each PUMA.
Those PUMAs that consisted of multiple counties in either year were dismissed outright,
since mothers in these PUMAs could only be assigned to a group of counties or school districts
rather than a particular county or school district. Although each of the remaining PUMAs
consisted of only one county, there were a number of PUMAs that identified the same county
within a given state (i.e. for counties that had a population of 200,000 or more). These PUMAs
were aggregated together and given the same county identification number. The final step
included a comparison of the remaining counties with the reported school districts for both years.
The Local Agency Universe Survey provides data on public education agencies, both by
state and county. Thus, data from this source provided information on the number of school
districts located in a given county within a state. The majority of the counties that remained at
this point were inclusive of only one school district. However, there were several that included
multiple school districts. The majority of these were counties that had their own school districts
but contained a larger city that also had its own school district. If that large city was not
aggregated into the same PUMA as the county and could instead be identified by its own PUMA,
this was not a problem. However, if the city was not identified by its own PUMA but was
instead aggregated into the same PUMA as the county it was located in, it became problematic.
For counties like these, it is hard to match mothers to a school district because the mother could
enroll her child in either the county or city school district. Therefore, counties of this type were
dropped.
Examples of these situations can be seen in looking at Baltimore County in Maryland and
Ouachita Parish in Louisiana. Baltimore County in Maryland contains both Baltimore County
27
School District and Baltimore City School District. Baltimore County and Baltimore City can
both be identified by a distinct group of PUMAs, however, and were therefore included in the
sample. Ouachita Parish contains both Ouachita Parish School District and Monroe City School
District. Both Ouachita Parish and Monroe City are both identified only within the same PUMA.
Thus, mothers in this PUMA may enroll their child in either one of the two school districts in
Ouachita Parish, making the matching of mothers to school districts impossible without further
information. For this reason, Ouachita Parish was dropped from the sample.
A total of fifty-nine districts in ten states were used in this study. Their names, as well as
the names of the states they are located in, are given in Table 2. In addition, Appendix Table 1
gives examples of different types of PUMAs discussed above.
Appendix 2
Census and ACS Samples
After limiting the data to the fifty-nine districts in the ten states (see Appendix 1), all children
between the ages of 0 to 17 were matched to their mothers for each year. Additionally, any non-
spousal members of the household that were 18 years or older were matched to mothers.
Samples were then limited to mothers of two and four-year-olds aged 18 to 45 at the time of the
Census. Further limitations resulted in full samples of married and single mothers. Mothers
were considered to be single if reported to be the head of a household where no spouse was
present. Households where more than one family was present were dropped from the samples,
since estimates within these households may be affected by a number of unobservable
characteristics (this resulted in a small number of mothers being excluded from the sample—0.24
28
percent of the sample in the most extreme case). Furthermore, the samples were restricted to
mothers of only one two, four, or seven-year-old in order to maintain comparability (Gelbach
2002).
Appendix 3
Prekindergarten Availability and Construction of percent_pk
Data on grade span was used to construct the variable of interest, percent_pk, the fraction
of public schools within a district that offer prekindergarten programs. Data was taken from the
Public Secondary-Elementary School Universe Survey for the appropriate years and was
downloaded from the National Center for Education Statistics. Grade span is reported as the low
and high grades offered within a school. For each district in a given year, the fraction of public
schools offering prekindergarten programs was calculated as the sum of operating schools
offering prekindergarten (PK-K, PK-1, PK-2, PK-3, PK-4, PK-5, PK-6, PK-7, PK-8, PK-9, PK-
12) over the sum of operating schools offering prekindergarten or kindergarten (PK-K, PK-1,
PK-2, PK-3, PK-4, PK-5, PK-6, PK-7, PK-8, PK-9, PK-12, K-1, K-2, K-3, K-4, K-5, K-6, K-7,
K-8, K-9, K-12).
Any changes in the availability of prekindergarten will not be recognized until the
beginning of the school year, which takes place in the fall. Consequently, changes in
employment that are a response to the availability of prekindergarten will not be recognized until
the following year. For this reason, percent_pk for a given year was taken from the beginning of
the previous year’s school year. This means that percent_pk for 1990 was taken from the 1989-
1990 school year, for 2000 from the 1999-2000 school year, for 2005 from the 2004-2005 school
year, and for 2006 from the 2005-2006 school year. Percent_pk was matched to mothers in
29
district d in year t with a unique identification number that was coded as a combination of the
state FIPS identification number and ICPSR county identification number in each year.
It is important to note that in four of the fifty-nine districts (Broward County, Orange
County, Hillsborough County, and Mecklenburg County) the number for percent_pk was
replaced in the year 2000 because there appeared to be reporting error. For Osceola County
School District, percent_pk was replaced in both 1990 and 2000. This change was made because
the time-path for percent_pk appeared to follow an unnatural path over the time period for these
districts. For example, Orange County School District in Florida reported 42.7 percent in 1990,
0.03 percent in 2000, and 91.0 percent in 2005/2006. For each one of these districts, the number
used as replacement was the average of percent_pk for the year before and after the replaced
year.
References
Abbott-Shim, M., Lambert, R., & McCarty, F. “A comparison of school readiness
outcomes for children randomly assigned to a Head Start program and the program’s wait
list.” Journal of Education for Students Placed at Risk 8.2 (2003): 191–214.
Averett, Susan L., Elizabeth Peters, and Donald M. Waldman. “Tax Credits, Labor Supply, and
Child Care.” The Review of Economics and Statistics 79.1 (1997): 125-135.
Berger, Mark and Dan Black. “Child Care Subsidies, Quality of Care, and the Labor Supply of
Low-Income, Single Mothers.” The Review of Economics and Statistics 74.4 (1992):
635-642.
Baker, Michael et al. “Universal Child Care, Maternal Labor Supply, and Family Well-Being.”
Journal of Political Economy 116.4 (2008): 709-745.
Barnett, W. Steven et al. “The State of Preschool 2009.” National Institute for Early Education
Research.
Barnett, W. Steven et al. “The State of Preschool 2003.” National Institute for Early Education
Research.
30
Berlinski, Samuel, and Sebastian Galiani. “The Effect of a Large Expansion of Pre-Primary
School Facilities on Preschool Attendance and Maternal Employment.” Labour
Economics 14.3 (2007): 665-680.
Blau, David M. and A.P. Hagy. “The Demand for Quality in Child Care.” Journal of
Political Economy 106.1 (1998): 104-146.
Blau, David M. and Philip K. Robins. “Child-care Costs and Family Labor Supply.”
The Review of Economics and Statistics 70.3 (1988): 374-381.
Blau, David M. and Philip K. Robins. “Fertility, Employment, and Child-Care Costs.”
Demography 26.2 (1989): 287-299.
Cascio, Elizabeth. “Maternal Labor Supply and the Introduction of Kindergartens into American
Public Schools.” The Journal of Human Resources (2009).
Cascio, Elizabeth. “Public Preschool and Maternal Labor Supply: Evidence from the
Introduction of Kindergartens into American Public Schools.” National Bureau of
Economic Research (2006).
Connelly, Rachel and Jean Kimmell. “The Effect of Child Care Costs on the Labor Force
Participation and Welfare Recipiency of Single Mothers: Implications for Welfare
Reform.” Upjohn Institute Working Paper No 01-69 (2001). Kalamazoo, MI: W.E.
Upjohn Institute for Employment Research.
Connelly, Rachel. “The Effect of Child Care Costs on Married Women’s Labor Force
Participation.” The Review of Economics and Statistics 74.1 (1992): 83-90.
Connelly, Rachel. “The Effect of Child Care Costs on the Labor Force Participation and AFDC
Recipiency of Single Mothers.” Institute for Research on Poverty Discussion Paper no.
920-90 (1990).
Currie, Janet and Thomas, Duncan. “Does Head Start Make a Difference?” The American
Economic Review 85.3 (1995): 341-364.
Fitzpatrick, Maria Donovan. “Preschoolers Enrolled and Mothers at Work? The Effects of
Universal Prekindergarten.” Journal of Labor Economics 28.1 (2010): 51-85.
Fitzpatrick, Maria Donovan. “Revising Our Thinking about the Relationship between Maternal
Labor Supply and Preschool.” Forthcoming, Journal of Human Resources (2012).
Garces, Eliana, Duncan Thomas, and Janet Currie. “Longer Term Effects of Head
Start.” American Economic Review 92.4 (2002): 999-1012.
Gelbach, Jonah B. “Public Schooling for Young Children and Maternal Labor
Supply.” American Economic Association 92.1 (2002): 307-322.
31
Hotz, V. Joseph, and M. Rebecca Kilburn. “The Demand for Child Care and Child
Care Costs: Should We Ignore Families with Non-Working Mothers?” Working Paper
Series 92-1 (1991), Harris School, University of Chicago.
Kimmel, Jean. “Child Care Costs as a Barrier to Employment for Single and Married
Mothers.” The Review of Economics and Statistics 80.2 (1998): 287-299.
Kimmel, Jean. “The Effectiveness of Child-Care Subsidies in Encouraging the
Welfare-to-Work Transition of Low-Income Single Mothers.” The American
Economic Review 85.2 (1995): 271-275.
Micalopoulos, Charles et al. “A Structural Model of Labor Supply and Child Care Demand.”
The Journal of Human Resources 27.1 (1992): 166-203.
Oden, S., Schweinhart et al. (2000). Into adulthood: A study of the effects of Head Start.
Ypsilanti, MI: High/Scope Press.
Puma et al. “Head Start Impact Study: Final Report.” U.S. Department of Health and Human
Services, Office of Planning, Research and Evaluation (2010).
Ribar, David C. “Child Care and the Labor Supply of Married Women Reduced Form
Evidence.” The Journal of Human Resources 27.1 (1992): 134-165.
Schlosser, Analia. “Public Preschool and the Labor Supply of Arab Mothers: Evidence from a
Natural Experiment.” The Hebrew University of Jerusalem,
Department of Economics (2005).
32
Figure 1
Fraction of Three and Four-Year-Olds Enrolled in Prekindergarten Programs
Source: October Supplement of the Community Population Survey
33
Figure 2
Fraction of Public Schools in the United States Offering Prekindergarten Programs
Sources: Public Secondary-Elementary School Universe Survey
Note: See Appendix 2 for a description of how this fraction was calculated.
34
Figure 3
Labor force participation rate of mothers by age of their youngest child
Sources: March Supplement of the Community Population Survey
35
Figure 4
Change in the percentage of schools within a school district offering public prekindergarten
Source: Public Secondary-Elementary School Universe Survey
Notes: Negative changes in percent_pk are included in the 0 to 0.1 category.
36
Table 1 Names of States and School Districts Used
State District State District
Alabama: Mobile County Maryland: Anne Arundel County
Montgomery County Baltimore City
Baltimore County
Florida: Alachua County Carroll County
Broward County Charles County
Charlotte County Frederick County
Duval County Harford County
Hernando County Howard County
Hillsborough County Montgomery County
Lake County Prince Georges County
Manatee County Washington County
Marion County
Martin County North Carolina: Pitt County
Orange County Forsyth County
Osceola County Mecklenburg County
Pasco County Onslow County
Pinellas County Wake County
Polk County
St. Lucie County South Carolina: Aiken County
Volusia County Charleston County
Greenville County
Georgia: Bibb County Horry County
Chatham County
Clayton County Tennessee: Knox County
Fulton County Montgomery County
Sumner County
Kentucky: Jefferson County
Virginia Alexandria City
Louisiana: Jefferson Parish Arlington County
Lafayette Parish Chesapeake City
Rapides Parish Chesterfield
Hampton City
Henrico County
Newport News City
Norfolk City
Richmond City
37
Table 2
Demographic Characteristics of Mothers of Four-Year-Olds, by inclusion in sample, 2005/06
ACS
Variable In Sample
Not In
Sample P-Value
Demographic characteristics
Age 33.1 32.3 0
White 0.58 0.65 0
Black 0.23 0.09 0
Hispanic 0.12 0.18 0
Other Race 0.07 0.08 0.49
Married 0.75 0.65 0
Education
In School 0.10 0.13 0
Less Than High School 0.09 0.12 0.02
High School 0.30 0.35 0.06
Some College 0.24 0.25 0.48
College Graduate 0.37 0.29 0.03
Labor Supply
Usual Hours Worked per Week 25.0 24.8 0.54
Weeks Worked Last Year 29.7 29.7 0.996
Employed 0.59 0.60 0.58
In Labor Force 0.64 0.65 0.35
In Poverty 0.15 0.18 0.12
On Welfare 0.03 0.04 0.23
Observations 4773 85816
Notes: P-Values reported are for the difference in means.
Table 3
38
Demographic Characteristics of Mothers of Two, Four, and Seven-Year Olds, by presence and
age of younger children, 1990/2000 Census 5% PUMS, 2005/06 ACS
Demographic Characteristics
Variable
4-Year Olds
(No Younger)
4-Year Olds
(No Younger,
No 7-Year Olds)
2-Year Olds
(No Younger No
4 or 7-Year Olds)
7-Year Olds
(No Younger)
Age 32.9 32.8 30.5 35.6
White 0.63 0.62 0.65 0.63
Black 0.25 0.26 0.23 0.26
Hispanic 0.08 0.07 0.07 0.07
Other Race 0.04 0.05 0.05 0.04
In Large City 0.04 0.04 0.04 0.04
Education:
In School 0.10 0.10 0.10 0.09
Less Than High School 0.09 0.09 0.09 0.09
High School 0.36 0.37 0.35 0.37
Some College 0.29 0.29 0.28 0.29
College Graduate 0.26 0.25 0.28 0.25
Outcomes:
Usual Hours Worked/Week 27.8 28.4 26.6 30.0
Weeks Worked Last Year 32.9 33.6 31.2 35.9
Employed 0.66 0.67 0.63 0.72
In Labor Force 0.70 0.72 0.68 0.76
In Poverty 0.14 0.13 0.13 0.12
On Welfare 0.05 0.05 0.05 0.04
Number & Age of Children:
1-Year Old 0 0 0 0
2-Year Old 0 0 1 0
3-Year Old 0 0 0.08 0
4-Year Old 1 1 0 0
5-Year Old 0.07 0.07 0.15 0
6-Year Old 0.16 0.17 0.12 0
7-Year Old 0.15 0 0 1
8-12 Year Old 0.45 0.46 0.26 0.67
13-17 Year Old 0.17 0.18 0.12 0.32
18 or older Non-Spouse 0.43 0.45 0.40 0.63
Preschool Availability:
% Schools in District With
Prekindergarten 0.43 0.43 0.43 0.44
Sample Size 15945 13562 16512 13104
Table 4
39
Coefficients on the percentage of public schools offering prekindergarten for enrollment of three
and four-year olds, 1990/2000 Census 5% PUMS, 2005/06 ACS
Difference-in-Difference Estimates
Sample
Total
Enrollment
Public
Enrollment
Private
Enrollment
3-Year Olds
Percent_pk
0.019
(0.023)
0.024
(0.017)
-0.005
(0.020)
Sample Mean 0.41 0.15 0.26
Sample Size 18562
4-Year Olds
Percent_pk
0.048*
(0.026)
0.060**
(0.025)
-0.013
(0.023)
Sample Mean 0.61 0.27 0.34
Sample Size 15945
Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **,
and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels,
respectively. In addition to the variable of interest (percentage of public schools offering
prekindergarten), each regression includes controls for year and district effects as well as a
vector of maternal controls including one for race and ethnicity, age and number of children in
the household, a quadratic in maternal age, education level, indicators for veteran and marital
status, and an indicator for whether or not the mom lives in a large-sized city.
Table 5
40
Coefficients on the percentage of public schools offering prekindergarten for Mothers Two,
Four, and Seven-Year Olds by presence and age of younger children, 1990/2000 Census 5%
PUMS, 2005/06 ACS
Difference-in-Difference Estimates
4-Year Olds
(No Younger)
4-Year Olds
(No Younger, No 7-
year Olds)
2-Year Olds
(No Younger, No 4
or 7-Year Olds)
7-Year Olds
(No Younger)
Sample
In labor
force Employed
In labor
force Employed
In labor
force Employed
In labor
force Employed
Full Sample
Percent_pk 0.038***
(0.014)
0.044***
(0.014)
0.037***
(0.014)
0.048***
(0.016)
-0.016
(0.019)
-0.023
(0.022)
-0.016
(0.020)
-0.027
(0.018)
Sample Mean 0.70 0.66 0.72 0.67 0.68 0.63 0.76 0.72
Sample Size 15945 13562 16512 13104
Married
Percent_pk 0.043**
(0.020)
0.055***
(0.017)
0.043**
(0.020)
0.058***
(0.027)
-0.0002
(0.022)
-0.0002
(0.024)
-0.004
(0.026)
-0.019
(0.027)
Sample Mean 0.67 0.65 0.68 0.66 0.66 0.63 0.73 0.71
Sample Size 11319 9446 12205 8877
Single
Percent_pk 0.046
(0.041)
0.021
(0.045)
0.040
(0.037)
0.019
(0.044)
-0.080*
(0.046)
-0.151***
(0.046)
-0.055*
(0.031)
-0.059
(0.040)
Sample Mean 0.80 0.72 0.81 0.73 0.75 0.66 0.83 0.76
Sample Size 3582 3165 2983 3381
Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **,
and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels,
respectively. In addition to the variable of interest (percentage of public schools offering
prekindergarten), each regression includes controls for year and district effects as well as a
vector of maternal controls including one for race and ethnicity, age and number of children in
the household, a quadratic in maternal age, education level, indicators for veteran and marital
status, and an indicator for whether or not the mom lives in a large-sized city.
Table 6
41
Coefficients on the percentage of public schools offering prekindergarten for Mothers of Four-
Year Olds by presence of younger children, 1990/2000 Census 5% PUMS, 2005/06 ACS
Difference-In-Difference Estimates
4-year olds
(No Younger)
4-Year Olds
(Any Aged Children)
Sample
In labor
force Employed
In labor
force Employed
Full Sample
Percent_pk
0.038***
(0.014)
0.044***
(0.014)
0.036**
(0.014)
0.042***
(0.014)
Sample Mean 0.77 0.66 0.64 0.59
Sample Size 15945 26938
Married
Percent_pk
0.043**
(0.020)
0.055***
(0.017)
0.036*
(0.019)
0.041**
(0.017)
Sample Mean 0.67 0.65 0.61 0.58
Sample Size 11319 19764
Single
Percent_pk
0.046
(0.041)
0.021
(0.045)
0.041
(0.053)
0.035
(0.049)
Sample Mean 0.80 0.72 0.74 0.64
Sample Size 3582 5485
Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **,
and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels,
respectively. In addition to the variable of interest (percentage of public schools offering
prekindergarten), each regression includes controls for year and district effects as well as a
vector of maternal controls including one for race and ethnicity, age and number of children in
the household, a quadratic in maternal age, education level, indicators for veteran and marital
status, and an indicator for whether or not the mom lives in a large-sized city.
Appendix Table 1
Examples of PUMAs encountered during restriction process and action taken
42
County Puma Treatment
Baltimore 00501 Aggregate
00502 Aggregate
00503 Aggregate
00504 Aggregate
00505 Aggregate
00506 Aggregate
00507 Aggregate
Baltimore City 00801 Aggregate
00802 Aggregate
00803 Aggregate
00804 Aggregate
00805 Aggregate
00806 Aggregate
Carroll 00400
Keep but don't
aggregate
Cecil 00700 Delete
Kent 00700 Delete
Queen Anne's 01300 Delete
Caroline 01300 Delete
Talbot 01300 Delete
Dorchester 01300 Delete
Wicomico 01400 Delete
Somerset 01400 Delete
Worcester 01400 Delete
Notes: This figure provides examples of the different types of counties that were encountered in
the restriction process and the treatment that was given to each one. If a county was large
enough to be uniquely identified by more than one PUMA, all of the PUMAs that identified that
county were aggregated together and given the same identification number. If a county was
large enough to be uniquely identified by only one PUMA, the PUMA identifying that county
was kept, with no further action necessary. Those small counties that were in the same PUMA
as another distinct county were dropped from the sample.
43

More Related Content

What's hot

Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...
Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...
Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...IFPRIMaSSP
 
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...Moawia Alshiek
 
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...ijtsrd
 
Early Childhood Education is playing a role in market development
Early Childhood Education is playing a role in market development Early Childhood Education is playing a role in market development
Early Childhood Education is playing a role in market development UTAR
 
PPL Young Fathers Historical Data
PPL Young Fathers Historical DataPPL Young Fathers Historical Data
PPL Young Fathers Historical DataPatrick Morley
 

What's hot (12)

Family report card
Family report cardFamily report card
Family report card
 
Exploring the Landscape
Exploring the LandscapeExploring the Landscape
Exploring the Landscape
 
Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...
Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...
Child labor and schooling in Malawi: Does mother’s employment matter ? by Mic...
 
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...
Mohieldin et al. - Unknown - The impact of feeding practices on prevalence of...
 
Original article
Original articleOriginal article
Original article
 
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...
A Quasi Experimental Study to Assess the Effectiveness of Structured Teaching...
 
Ghel
GhelGhel
Ghel
 
Nclb artifact
Nclb artifactNclb artifact
Nclb artifact
 
Early Childhood Education is playing a role in market development
Early Childhood Education is playing a role in market development Early Childhood Education is playing a role in market development
Early Childhood Education is playing a role in market development
 
Miracles
MiraclesMiracles
Miracles
 
PPL Young Fathers Historical Data
PPL Young Fathers Historical DataPPL Young Fathers Historical Data
PPL Young Fathers Historical Data
 
Case To Reopen Schools
Case To Reopen SchoolsCase To Reopen Schools
Case To Reopen Schools
 

Similar to Maternal_Labor_07222012

BROOKINGS December 2011 `
BROOKINGS  December 2011        ` BROOKINGS  December 2011        `
BROOKINGS December 2011 ` VannaSchrader3
 
An Analysis Of US Newspaper Coverage. Of Early Childhood Education
An Analysis Of US Newspaper Coverage. Of Early Childhood EducationAn Analysis Of US Newspaper Coverage. Of Early Childhood Education
An Analysis Of US Newspaper Coverage. Of Early Childhood EducationEmily Smith
 
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptx
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptxThe Impact of the Pantawid Pamilyang Pilipino Program ppt.pptx
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptxYmil1
 
ABSTRACT OF THE THESIS.docx
ABSTRACT OF THE THESIS.docxABSTRACT OF THE THESIS.docx
ABSTRACT OF THE THESIS.docxlhye park
 
Final Draft of Seminar Course
Final Draft of Seminar CourseFinal Draft of Seminar Course
Final Draft of Seminar CourseBethany Watson
 
Son preference and fertility behavior evidence from Viet Nam - Project statement
Son preference and fertility behavior evidence from Viet Nam - Project statementSon preference and fertility behavior evidence from Viet Nam - Project statement
Son preference and fertility behavior evidence from Viet Nam - Project statementHanh To
 
Reproductive Health and Economic Development: What Connections Should We Focu...
Reproductive Health and Economic Development: What Connections Should We Focu...Reproductive Health and Economic Development: What Connections Should We Focu...
Reproductive Health and Economic Development: What Connections Should We Focu...The Population and Poverty Research Network
 
Webb_Dicaire_774Final
Webb_Dicaire_774FinalWebb_Dicaire_774Final
Webb_Dicaire_774FinalSimon Webb
 
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...The Influence of Parental Level of Income in Pre-School Preference in Nyamira...
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...paperpublications3
 
Capstone Research Paper. Fall 2015. Huang, Nguyen & Zhang
Capstone Research Paper. Fall 2015. Huang, Nguyen & ZhangCapstone Research Paper. Fall 2015. Huang, Nguyen & Zhang
Capstone Research Paper. Fall 2015. Huang, Nguyen & ZhangZijian Huang
 
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docx
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docxUNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docx
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docxwillcoxjanay
 
New Trends in Parent Involvement and Student Achievement
New Trends in Parent Involvement and Student AchievementNew Trends in Parent Involvement and Student Achievement
New Trends in Parent Involvement and Student Achievementnoblex1
 
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...Driessen Research
 

Similar to Maternal_Labor_07222012 (20)

BROOKINGS December 2011 `
BROOKINGS  December 2011        ` BROOKINGS  December 2011        `
BROOKINGS December 2011 `
 
An Analysis Of US Newspaper Coverage. Of Early Childhood Education
An Analysis Of US Newspaper Coverage. Of Early Childhood EducationAn Analysis Of US Newspaper Coverage. Of Early Childhood Education
An Analysis Of US Newspaper Coverage. Of Early Childhood Education
 
33.pdf
33.pdf33.pdf
33.pdf
 
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptx
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptxThe Impact of the Pantawid Pamilyang Pilipino Program ppt.pptx
The Impact of the Pantawid Pamilyang Pilipino Program ppt.pptx
 
ABSTRACT OF THE THESIS.docx
ABSTRACT OF THE THESIS.docxABSTRACT OF THE THESIS.docx
ABSTRACT OF THE THESIS.docx
 
Final Draft of Seminar Course
Final Draft of Seminar CourseFinal Draft of Seminar Course
Final Draft of Seminar Course
 
Son preference and fertility behavior evidence from Viet Nam - Project statement
Son preference and fertility behavior evidence from Viet Nam - Project statementSon preference and fertility behavior evidence from Viet Nam - Project statement
Son preference and fertility behavior evidence from Viet Nam - Project statement
 
Maternial education
Maternial educationMaternial education
Maternial education
 
Reproductive Health and Economic Development: What Connections Should We Focu...
Reproductive Health and Economic Development: What Connections Should We Focu...Reproductive Health and Economic Development: What Connections Should We Focu...
Reproductive Health and Economic Development: What Connections Should We Focu...
 
FINAL THESIS
FINAL THESISFINAL THESIS
FINAL THESIS
 
J4163070.pdf
J4163070.pdfJ4163070.pdf
J4163070.pdf
 
Webb_Dicaire_774Final
Webb_Dicaire_774FinalWebb_Dicaire_774Final
Webb_Dicaire_774Final
 
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...The Influence of Parental Level of Income in Pre-School Preference in Nyamira...
The Influence of Parental Level of Income in Pre-School Preference in Nyamira...
 
Capstone Research Paper. Fall 2015. Huang, Nguyen & Zhang
Capstone Research Paper. Fall 2015. Huang, Nguyen & ZhangCapstone Research Paper. Fall 2015. Huang, Nguyen & Zhang
Capstone Research Paper. Fall 2015. Huang, Nguyen & Zhang
 
The Impact of a Food for Education Program on Schooling in Cambodia
The Impact of a Food for Education Program on Schooling in CambodiaThe Impact of a Food for Education Program on Schooling in Cambodia
The Impact of a Food for Education Program on Schooling in Cambodia
 
Taxing Families: The Impact of Child-related Transfers on Maternal Labor Supply
Taxing Families: The Impact of Child-related Transfers on Maternal Labor SupplyTaxing Families: The Impact of Child-related Transfers on Maternal Labor Supply
Taxing Families: The Impact of Child-related Transfers on Maternal Labor Supply
 
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docx
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docxUNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docx
UNIT 13 Evaluation Paper (1)UNIT 13 Evaluation Paper (1)Criter.docx
 
New Trends in Parent Involvement and Student Achievement
New Trends in Parent Involvement and Student AchievementNew Trends in Parent Involvement and Student Achievement
New Trends in Parent Involvement and Student Achievement
 
TWSB-Paper
TWSB-PaperTWSB-Paper
TWSB-Paper
 
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...
Geert Driessen (2023) Encyclopedia The Perry HighScope Preschool Program A Cr...
 

Maternal_Labor_07222012

  • 1. Maternal Labor Supply and the Availability of Public Pre-K: Evidence from the Introduction of Prekindergarten into American Public Schools Sean P. Sall Department of Economics University of Notre Dame July 22, 2012 Abstract In the 1980s and 1990s, many states and districts began to provide funding for prekindergarten programs for the first time. This paper takes advantage of the staggered timing in program funding to investigate the effect that increased availability of prekindergarten programs has on the labor supply of mothers with four-year-olds. I find that mothers with a four-year-old and no younger children were significantly more likely to be in the labor force and employed once prekindergarten became available. Mothers with a four-year-old and other younger children were also significantly more likely to be in the labor force and employed. All views expressed within this paper are those of the author, and any errors made are my own. I would like to thank William Evans for a number of helpful suggestions. I would also like to thank the sponsors of the annual Bernoulli Competition at Notre Dame, for which this paper was written. 1
  • 2. I. Introduction In recent decades, there has been increasing evidence that prekindergarten programs (PK) are effective at preparing children for formal education. Studies have demonstrated that Head Start not only prepares students to be successful in school by improving cognitive skills (Currie and Thomas 1995; Abbott-Shim et al., 2003; Puma et al., 2010), but also increases the likelihood that participants graduate high school, attend college, and have higher earnings (Oden et al., 2000; Garces et al., 2002). It is not surprising then that the fraction of three and four-year-olds in such programs has grown substantially. The top line in Figure 1 reports PK enrollment rates for three and four-year-olds from the October School Enrollment supplement over the 1970 through 2007 period.1 Note that over this period the fraction of children in this age group in PK programs has more than doubled, from 20.5 percent in 1970 to 54.5 percent in 2007. Much of the change in PK enrollment since 1980 is the result of larger enrollment in public school programs. In Figure 1, I also graph the fraction of PK in private and public schools. Note that between 1970 and 2007, PK enrollment in public schools has increased from 6.2 percent to 30.3 percent, while private has increased from 14.3 percent to 24.2 percent. The expansion in PK enrollment is primarily due to an increase in the number of public schools offering PK programs. In Figure 2, I graph the fraction of public elementary schools offering PK programs over this time period. This data comes from the Common Core of Data, which contains enrollment information for the universe of schools in the US. These figures indicate that over this period, the fraction of public schools offering PK programs has increased by nearly a factor of 4, from 14.7 percent in 1987 to 56.1 percent in 2006. While these trends in enrollment may have educational benefits for attendees, they may 1 These enrollment numbers include three and four-year-olds participating in Head Start. 2
  • 3. also affect the labor supply of attendees’ parents. Childcare is potentially a large out-of-pocket cost of participating in the labor force for parents with young children. Since the majority of public schools offer PK at no or reduced costs, these programs greatly reduce the direct cost of childcare for millions of families and hence greater availability of these programs should increase labor supply among mothers with pre-school children. The growing number of public prekindergarten programs in recent years coincides with an increase in labor force participation among mothers with four-year-old children. Figure 3 reports the fraction of women that work whose youngest child is aged 2, 4, and 7. The data for this graph is taken from the March Current Population survey. Note that from 1987 to 2006, as the fraction of public schools offering prekindergarten programs increased dramatically, the fraction of mothers whose child was four that were in the labor force increased by 12 percent, from 62.8 to 70.3 percent. Note also that over this time period, the difference in labor force participation between mothers of four-year-olds and two-year-olds grows, while the difference between mothers of four-year-olds and seven-year-olds declines. In this paper, I investigate the effect of increased accessibility to public prekindergarten on the labor supply of mothers of four-year-olds by using variation in the introduction of prekindergarten programs into American public schools. At the local level, most counties have multiple school districts, but there are 13 states where schools are county-based. When counties are identified in these states, we then know the student’s school district. Using data for these 13 states from the 5 percent Public Use Micro Sample (PUMS) and the American Community Survey (ACS), as well as education data from the Common Core, I estimate a difference-in- difference model that examines the change in labor supply of mothers with four-year-olds in counties with an increasing supply of prekindergarten seats over time. 3
  • 4. These reduced-form models aim to answer several questions: Does increased accessibility to public prekindergarten programs impact the labor force participation of mothers with a four- year-old as their youngest child? Is there heterogeneity in the labor supply response to public prekindergarten? Which mothers are most impacted by the expansion of public prekindergarten programs? The results of this study indicate substantial effects of prekindergarten availability on the labor supply of mothers of four-year-olds, both with and without younger children. For mothers of four-year-olds in a district with no public PK, full implementation is associated with statistically significant increases of 3.8 percentage points in labor force participation and 4.4 percentage points in employment. Estimates for married mothers were of larger magnitude and statistically significant, while estimates for single mothers were not statistically significant. The largest concern with the difference-in-difference model is that the number of prekindergarten programs in a given county may be determined by trends in the labor supply of mothers in that county. For example, if a county has a growing number of working mothers, the district may respond to the demands of the constituents by providing more public PK programs. The inclusion of county fixed-effects helps to minimize such a bias, but this will not eliminate time-specific shocks that may contaminate a fixed-effect model. One way to test for the presence of bias is to estimate separate labor supply models for mothers of two-year-olds and seven-year-olds, who should not be impacted by increased provision of prekindergarten programs. If estimates for mothers of four-year-olds are picking up spurious correlation between increased PK availability and increased labor supply, estimates for these other mothers should show similar correlations with pre-school availability. On the other hand, if estimates for mothers of four-year-olds are not biased and are picking up the true effects of increased 4
  • 5. provision of programs, then there should be no impact of PK availability on the labor supply of these other mothers, which is exactly what I find. This study makes several contributions to the existing literature on childcare costs and maternal labor supply. Most importantly, this is one of two studies that examines the effect of public prekindergarten programs on the labor supply of mothers within the United States, the other performed by Fitzpatrick (2010). A number of authors have examined the impact of free kindergarten on maternal labor supply, including papers by Gelbach (2002), Cascio (2006; 2009), and Fitzpatrick (2012). Meanwhile, Fitzpatrick (2010) is the only study of pre- kindergarten aged children, examining the effects of universal PK programs in Georgia and Oklahoma. This study employs a different econometric analysis than Fitzpatrick and a different sample population. It can help to assess the costs and benefits of public prekindergarten programs as well as the value of universal programs versus those targeted towards specific populations. The results of this study are of particular interest to both state and local governments that have recently decreased funding for public education. II. Background and Existing Literature There are a number of different child-care options for parents of three and four-year-olds. Some are more academically focused, such as prekindergarten and preschool. The latter is typically not affiliated with a school system. Nursery schools, which place less emphasis on education, are also an option. Finally, there is an assortment of other daycare programs that do not have an educational focus. Before the introduction of prekindergarten programs into American public schools during the 1960’s, the majority of three and four-year-olds attended day care programs – only ten 5
  • 6. percent of three and four-year-olds were enrolled in any type of classroom (2003 State of Preschool Yearbook). Today, over sixty percent of three and four-year-olds are enrolled in a prekindergarten program, and more than half of these children are enrolled in a public program. Children that live in areas where public programs are not available are typically enrolled in private prekindergarten programs or nursery schools (2009 State of Preschool Yearbook). The majority of states began funding for public prekindergarten programs as pilot or experimental programs in the 1980’s or 1990’s, although initiatives varied substantially from state to state. For example, Georgia’s prekindergarten program began as experimental in 1993 and was expanded to be universally available in 1995, while Arizona’s program began as a pilot in 1991 and expanded as part of a block grant program in 1996. Additional differences in programs include the population served. Some states, such as New York or Florida, focused on universal availability, while states such as North Carolina and Texas focused upon at-risk populations. Programs vary in terms of operating standards as well, although the majority of programs are offered either full or part-day for four or five days a week (2009 State of Preschool Yearbook). In any case, the free or reduced cost of these public programs lowers childcare costs for mothers, which we would expect to increase maternal labor supply. Existing studies offer a wide range of price elasticity estimates both for single and married mothers. In their seminal paper on the subject, Blau and Robins (1988) estimated a childcare price elasticity of labor supply of -0.38 for married mothers. Many studies soon followed, all using similar estimation procedures but attaining widely different results. Blau and Robins (1989) estimated elasticities of -0.47 and -0.77, Hotz and Killburn (1991) 0, Michalopoulous et al. (1992) an elasticity of 0, Connelly (1992) -0.2, Ribar (1992) -0.74, Averett et al. (1997) -0.78, and Kimmel (1998) -0.92. The results for single mothers are equally varied: 6
  • 7. Connelly (1990) estimated an elasticity of 0, Kimmel (1995) -1.36, Kimmel (1995) -0.346 and -0.345, Kimmel (1998) -0.22, and Connelly and Kimmel (2001) -1.0 and -0.8. Several explanations for the discrepancies between estimates have been put forward, each identifying potential sources of bias. Averett et al. (1997) proposes that selection bias impacts many of these estimates. Childcare costs are typically measured as childcare expenditures per household, presenting problems when considering that women who are utilizing childcare are more likely to work. Blau and Hagy (1998) suggests that the variability in estimates is due to omitted variables bias, where the authors do not accurately control for the quality of childcare. Although many authors use controls to minimize these potential sources of bias, the fact is that these models explain a small fraction of the variation in labor supply, raising concerns that many of these studies suffer from an inability to adequately control for these factors. One way of sidestepping the problem of selection bias is to utilize quasi-experimental variation in childcare availability or prices. Berger and Black (1992) were the first to use a difference-in-difference model to examine the relationship between childcare costs and maternal labor supply. Focusing on two subsidy programs in Kentucky, they compared the outcomes of mothers before and after receipt of childcare subsidies to the outcomes for mothers over the same time period that were eligible but did not receive a subsidy. Berger and Black found that the program increased the probability of employment by 12 percent and increased labor force participation by 25.3 percent. Since parents chose the childcare they purchased, however, there is still concern whether the study effectively controls for quality. Examining the labor supply consequences of public schooling programs has the advantage that many programs have a 100 percent subsidy. These programs also reduce 7
  • 8. concerns about appropriately controlling for quality of childcare, since each mother within the same school catchment area is offered a program of the same quality. Despite the advantages of these programs, there have only been five studies using US data that have examined the impact of these programs on labor supply: one by Gelbach (2002), two by Cascio (2006; 2009), and two by Fitzpatrick (2010; 2012). Gelbach (2002) estimated a two-stage least squares model using quarter of birth as an instrumental variable for kindergarten enrollment. For single mothers of a five-year-old and no younger children, Gelbach found that public school enrollment lead to statistically significant increases of 14.5 percent in weeks of work per year, 9.6 percent in usual hours of work per week, 14.5 percent in hours worked last week, 21.5 percent in wage and salary income, and 9.7 percent in employment. He also found an 11.2 percent reduced probability of welfare receipt conditional on public school enrollment. Cascio (2006) estimated a difference-in-difference model using increased availability to public kindergarten as the treatment. For single mothers with a five- year-old and no younger children, Cascio found statistically significant increases in maternal labor supply: 13 percent in the likelihood of employment and 14.7 percent in hours worked per week. No statistically significant results were found for single women with a five-year-old and other younger children, or for married women with or without other younger children. In 2009, Cascio followed the same estimation procedure but employed a different variable of interest.2 The results obtained were similar in magnitude to those in her earlier paper, with statistically significant increases in labor supply variables for single mothers with no younger children and statistically insignificant results for other mothers. Fitzpatrick (2012) used public school enrollment as an instrumental variable in a regression discontinuity framework.3 She found that 2 The independent variable in this model was the fraction of the two years prior to the 2000 Census that kindergarten funding was available in the state in which the mother lived. 3 See the following paragraph for a more complete discussion of this estimation procedure. 8
  • 9. enrollment statistically significantly increased the employment of single mothers with no younger children, with no statistically significant results for other mothers. Fitzpatrick (2010) was the first to examine how availability of public prekindergarten programs impacts maternal labor supply within the United States. Fitzpatrick utilized a regression discontinuity design to exploit the fact that a child’s date of birth determines whether or not he/she can enroll in PK for a given year. Using restricted access data from the 2000 Decennial Census, her sample was limited to mothers of four-year-olds in Oklahoma and Georgia. The child’s date of birth was used to measure intention-to-treat effects, where those children born before September 1st , 1995 were eligible for public PK in the 1999-2000 school year and those who were born on or after September 2nd , 1995 were not eligible. Those mothers with four-year-olds that were eligible were in the treatment group, while those mothers with four-year-olds that were not eligible were in the comparison group.4 Samples included a full sample of mothers of four-year-olds, as well as subsamples based on marital status and the presence of younger children. The results indicated that the availability of universal PK had small and statistically insignificant effects on the mother’s likelihood of employment in the previous year and weeks or hours worked in the previous year, regardless of marital status or whether the mother had younger children. It is important to note that the standard errors on the estimates in this work are quite large.5 The use of difference-in-difference models to analyze the labor supply of mothers in foreign countries also reports large impacts of public preschool programs. Schlosser (2005) studied free public preschool and its effects on the labor supply of Israeli Arab mothers with 4 Only those mothers with a child born within 100 days before or after the cutoff were considered. 5 The standard errors are larger than the parameter estimates for all but a few of the dependent variables, and in many cases, several times larger than the parameter estimates. 9
  • 10. children aged two to four, Berlinkski and Galiani (2007) increased construction of preprimary schools and its effect on the labor supply of mothers in Argentina, and Baker et al. (2008) the effect that extension of full-time kindergarten programs had on mothers in Quebec. While the magnitude and precision of the estimates within these studies is varied, each finds evidence that increases in the supply of public preschool programs has positive effects on maternal labor supply, both for married mothers (Schlosser, 2005; Baker et al., 2008) and single mothers (Berlinkski and Galiani, 2007). III. Data A. Sample and Key Variables Sample data for this analysis is taken from the five-percent samples of the 1990 and 2000 Census, as well as the 2005 and 2006 American Community Survey. In order to accurately answer the question posed above, it is necessary to correctly assign each mother to the school district in which she enrolls her child. While school district is not reported within the data, states that use county-based school districts offer an opportunity to sidestep this limitation. Thirteen states use county-based school districts, and in these states, the county the mother resides in also determines the school district in which she enrolls her child. Although county is not reported in the data, the PUMA number for each mother is reported, and the counties contained within a given PUMA are also reported. This information allows for restriction of the data that matches mothers with school districts. A public use microdata area (PUMA) is defined as an area with a decennial census population of 100,000 people or more. Thus, the number of counties that are contained within a PUMA is dependent upon the size of these counties. Small counties that have populations less 10
  • 11. than 100,000 people must be aggregated with other counties to form a PUMA. Counties that have populations larger than 100,000 will be uniquely identified by one PUMA, or possibly more depending upon the size of the county.6 For small counties that are aggregated with others into one PUMA, it is impossible to match mothers with school districts. PUMAs that consist of only one county do allow matching of mothers with school districts, since it is known what county the mother resides in. Therefore, after limiting the data to those thirteen states with county-based school districts, I then limit it to those PUMAs that are consistent of a single county. The final step in this process is the aggregation of those counties that are large enough to be designated by more than one PUMA (population of 200,000 or more). A total of fifty-nine districts in ten states are used in this study. Table 1 shows the names of these districts (see Appendix 1 for a more detailed explanation of the restriction process). Because this is a geographically limited sample, there are concerns that the results of this study may not be indicative of how pre-kindergarten affects mothers’ labor supply in other parts of the United States. Table 2 reports the demographic characteristics of mothers of four-year-olds based on inclusion in the sample. The statistically significant differences in the means for demographic characteristics calls for caution in the interpretation of how pre-kindergarten affects mothers’ labor supply in other parts of the United States. For each of the labor supply measures, however, there are no statistically significant differences in means, suggesting that in terms of labor supply, mothers in my sample are not different than mothers in the rest of the country. My primary sample within the county-based school districts includes mothers of four- year-olds, both with and without younger children. I create similar samples of mothers of two and seven-year-olds to serve as comparison groups. Subpopulations that are used in the analysis 6 For example, a county with more than 200,000 people but less than 300,000 people may be uniquely identified by two PUMAs, a county with more than 300,000 people but less than 400,000 uniquely identified by three PUMAs, etc. 11
  • 12. include both married and single mothers of two, four, and seven-year-olds. All samples are restricted to include mothers aged 18-45 at the time of the Census.7 The key outcomes in this paper are two measures of employment that are observed in both the Census and American Community Survey for all years: an indicator for whether the mother is employed as well as one for whether the mom is in the labor force. In addition to the variable of interest, my regression includes controls for maternal race and ethnicity (indicators for black, white, Hispanic, and other race8 ), age and number of children (ages zero to one, two, three, four, five, six, seven, eight to twelve, thirteen to seventeen, and eighteen and older), a quadratic in maternal age, education level, indicators for veteran and marital status, and an indicator for whether or not the mom lives in a large city (coded as a city population greater than 500,000). B. Summary Statistics Table 3 reports demographic characteristics of mothers of two, four, and seven-year olds with no younger children. For the four employment outcomes observed in each year – usual weeks worked, weeks worked last year, employment (an indicator), and labor force participation (an indicator) – there are systematic differences between mothers of two, four, and seven-year-olds. Mothers of seven-year-olds consistently provide the highest levels of labor, followed by mothers of four-year-olds, with mothers of two-year-olds showing the lowest levels of labor supply. Compared to mothers of four-year-olds with no younger children, mothers of four-year-olds with younger children are on average younger and report lower levels of labor supply for each of the employment outcomes, likely because of the presence of younger children. 7 All samples were constructed without using sampling weights. 8 Indicators for black, white, and other race refer only to those mothers that are not of Hispanic origin. 12
  • 13. IV. Identification Strategy A. State Prekindergarten Funding Initiatives The large variation in state and district pre-kindergarten initiatives means that there are large differences in availability not only between states, but within states over time as well. Figure 4 presents the distribution of the change in the percentage of public schools within a district that have prekindergarten from 1990 to 2006 for all districts in the United States and for only those districts in my sample. For example, for a district with five schools, none of which offered prekindergarten in 1990 and three of which offered pre-kindergarten in 2006, the percentage change would be 60 percentage points. This figure shows there is substantial variation in the growth in availability of prekindergarten, both nationally and in my subsample of districts. However, while the distribution of the growth for all districts within the United States is clustered around 0 and 1, it is more evenly spread for my subsample of districts. My empirical strategy takes advantage of the differences in the growth in availability of prekindergarten for those districts in my sample. My variable of interest is the percentage of public schools that offer a prekindergarten program within a given district. I estimate the following difference-in-difference model: (1) yidt = θ percent_pkdt + Xidt β + αd + γt + εidt, where yidt is some measure of employment for mother i in district d in year t; Xidt is the aforementioned vector of maternal controls; and αd and γt represent district and year fixed- effects, respectively. The district fixed-effect αd controls for the time-invariant differences in 13
  • 14. maternal employment across districts, whereas the time effect γt captures changes in maternal labor supply over time that impact all districts, such as changes in the aggregate economy, federal programs, etc. The covariate of interest is percent_pkdt, which measures the percentage of public schools offering prekindergarten programs within a district. For the initial set of regressions, percent_pkdt is not weighted by the number of students or seats offered. The parameter of interest, θ, represents the change in some measure of maternal labor supply given a percentage point increase in the number of public schools offering prekindergarten within a district. My goal is to estimate θ for mothers whose youngest child is four-years-old, since this is the most appropriate age group for prekindergarten.9 One problem that could lead to biased estimates is differential trends in labor supply between mothers who live in districts that adopted public prekindergarten programs and those that didn’t. For example, the adoption of a public prekindergarten program could be correlated with the level of labor supply of mothers in a district, such that districts with low levels of maternal labor supply are more likely to adopt PK programs.10 If this were the case, then the estimates would overstate the true effect of increasing availability to public PK. To control for any bias caused by differential trends, I estimate difference-in-difference models that include mothers of two and seven-year-olds, who may be subject to similar trends but unaffected by the introduction of prekindergarten programs. 9 Although it is true that prekindergarten initiatives within some states are inclusive of younger age groups, first priority is given to four-year-olds. All available program spots are first filled by four-year-old children, and any remaining spots are open to three- year-old children. 10 To investigate this possibility, separate regressions were run for the change in the percentage of prekindergarten within a district from 1990 to 2006 on the initial levels of each of the labor supply variables. None of the coefficients were statistically significant for any of the samples of mothers of four-year-olds used in this study, with the exception of the coefficient on usual hours worked per week (0.0003) for the sample of any mothers of four-year-olds, which was marginally significant with a p- value of 0.068. Taken as a whole, these results suggest that PK initiatives are on average not correlated with initial levels of labor supply of the samples of mothers of four-year-olds used in this study. 14
  • 15. V. Results A. Effects on Public/Private Enrollment of Three and Four-Year-Olds Table 4 presents the coefficients for the difference-in-difference model (1) estimated with enrollment as the dependent variable.11 These estimates measure the effects of increasing availability to public PK on the probability of overall enrollment for three and four-year-olds, as well as public and private enrollment. Column 1 reports that a movement from no availability of public PK to full implementation is associated with an increase of 1.9 percentage points in the probability of overall enrollment for three-year-olds, and an increase of 4.8 percentage points for four-year-olds. The coefficient for three-year-olds is not statistically significant, while the coefficient for four-year-olds is statistically significant at the 10 percent level. Column 2 shows that for three-year-olds, a movement from no availability of public PK to full implementation is associated with a 2.4 percentage point increase in the probability of public enrollment, and for four-year-olds a 6.0 percentage point increase in the probability of public enrollment. The estimate for three-year-olds is not statistically significant, while the estimate for four-year-olds is statistically significant at the 5 percent level. The coefficients for the effects on the probability of private enrollment, -0.005 for three-year-olds and -0.013 for four-year-olds, are not statistically significant from zero for both samples. These estimates imply large effects on the probability of enrollment in public PK for four-year-olds, with smaller effects for three-year-olds. For the sample of four-year-olds, which reported a 43 percent enrollment rate in public PK in 1990, the 6.0 percentage point increase in the probability of public enrollment translates to a 14.0 percent (6.0/43.0) increase in the probability of public PK enrollment. For three-year-olds, whose reported enrollment in public 11 Enrollment is coded as a dummy variable, which equals one if the child is enrolled in school. The samples of three and four- year-olds used for these regressions were those from samples of mothers with three and four-year-olds and no younger children. 15
  • 16. PK was 26.1 percent in 1990, the 2.4 percentage point increase means a 9.2 percent (2.4/26.1) increase in the probability of public PK enrollment. The smaller, statistically insignificant effects of public PK on public enrollment of three-year-olds is as expected, since the majority of state initiatives allowed three-year-olds to enroll only after all four-year-olds were offered seats. In terms of private PK enrollment, the estimates suggest that increasing availability of public PK has no statistically significant effect on private enrollment for the sample of three or four-year- olds. These results suggest that there was little to no substitution between public and private PK programs, and that the increases in public PK enrollment were primarily the result of new enrollees. These findings are similar to those of Cascio (2006; 2009), who found large effects of increasing preschool availability on public school enrollment rates of five-year-olds, and Fitzpatrick (2010), who found large effects of increasing availability on overall enrollment rates of four-year-olds. Cascio (2006) found that a movement from no availability of kindergarten to full implementation was associated with an increase of 18.9 percentage points in the probability of public enrollment for five-year-olds of single mothers with no younger children and a 23.3 percentage point increase for five-year-olds of married mothers with no younger children. Neither of these estimates were statistically significant at traditional levels of significance. Cascio (2009) found a 15.2 percentage point increase in the probability of public enrollment for five-year-olds of single mothers with no younger children and an increase of 14.5 percentage points for five-year-olds of married mothers with no younger children, significant at the 5 and 1 percent levels, respectively. Fitzpatrick (2010) found that a movement from no availability of public PK to full implementation is associated with an increase of 7.2 percentage points in the probability of overall preschool enrollment for four-year-olds, significant at the 1 percent level. 16
  • 17. B. Effects for Mothers of Four-Year-Olds with No Younger Children Table 5 presents the effects of prekindergarten availability on mothers of two, four, and seven-year olds with no younger children. The first set of estimates reported are for the full sample of mothers.12 Regressions were run for two sets of mothers of four-year-olds: those with no younger children, and those with no younger children and no seven-year-old.13 The results for both of these samples are reported in Columns 1 and 2 of Table 5, respectively. The results for these samples are effectively the same, and thus I will hereafter only discuss those for the former sample, mothers of four-year-olds with no younger children (Column 1). For the full sample of mothers with a four-year-old and no younger children, the estimates imply that movement from no availability to full implementation of public PK is associated with a 3.8 percentage point increase in the probability of being in the labor force and a 4.4 percentage point increase in the likelihood of employment. Both of these coefficients are statistically significant at the 1 percent level. The second set of estimates reported in Table 5 are for samples of married mothers. The estimates for married mothers are larger in magnitude than the estimates for the full sample of mothers. The estimates for married mothers of four-year-olds with no younger children imply a 4.3 percentage point increase in the probability of labor force participation and a 5.5 percentage point increase in the probability of employment associated with movement from no availability to full implementation of public PK. The coefficient on labor force participation is statistically significant at the 5 percent level, while the coefficient on employment is statistically significant 12 This sample includes both single and married mothers, as well as those that are in neither group. Single mothers are considered to be those that are not married and are the head householder, which leaves mothers that are not married but not a head householder in neither the married nor single group. 13 Both samples are samples of mothers of four-year-olds with no younger children (one, two, or three-year-olds). The latter sample further excludes mothers that also have a seven-year-old child. 17
  • 18. at the 1 percent level. The third set of estimates in Table 5 are for single mothers of four-year- olds with no younger children. These estimates for single moms show large but statistically insignificant effects for increasing availability of public PK – 4.6 percentage points for the probability of labor force participation and 2.1 percentage points for the probability of employment.14 These estimates imply large effects of prekindergarten supply on the labor supply of mothers of a four-year-old with no younger children. For the full sample of mothers, the percentage point increases in labor supply translate to increases of 5.4 percent (3.8/71.0) in the probability of being in the labor force and 6.7 percent (4.4/66.2) in the probability of being employed. These effects are largest for married mothers, who show increases of 6.3 percent (4.3/68.6) in the probability of labor force participation and 8.4 percent (5.5/65.8) in the probability of employment.15 These findings are similar to those of Cascio (2006; 2009), Gelbach (2002), and Fitzpatrick (2012), who found large effects of kindergarten programs on single and married mothers of five-year-olds with no younger children. Cascio (2006; 2009) found that the full sample of mothers showed the smallest response. Although each found that single mothers showed larger response to kindergarten availability than married mothers, this study found the opposite with respect to prekindergarten availability. We should caution, however, that because of the small sample size of single mothers and the subsequent lack of precision with regards to the estimates, it is difficult to draw firm conclusions about which group is more responsive. In fact, if we conduct a t-test for the difference in parameter estimates between married and single 14 In a separate set of regressions, the same model was employed with the addition of state-fixed time effects. While the point estimates from these regressions are similar to those reported in Table 6, the standard errors are considerably larger. A table with these results is available upon request. 15 As before, these percentage increases were calculated as the percentage point change divided by the initial levels of labor supply, reported in 1990. For example, the percentage increase in the probability of labor force participation for the full sample of mothers was calculated as the 3.8 percentage point increase divided by the 71 percent labor force participation rate in 1990. 18
  • 19. mothers of four-year-olds with no younger (Column 1), the test-statistics for labor force participation (-0.066) and employment (0.707) are not statistically significant at traditional levels of significance.16 This suggests that the coefficient estimates for married and single mothers, and therefore their responses to increasing PK, are not significantly different. Fitzpatrick (2010), on the other hand, found that universal availability to PK resulted in statistically insignificant responses in maternal labor supply when she employed restricted-access data from the 2000 Census for only those mothers in Georgia and Oklahoma. However, since the standard errors on most coefficients in that work are larger than parameter estimates, there is a concern that the study was underpowered. Gelbach (2002) used restricted-access data from the 1980 Census for all fifty states, while Cascio (2006; 2009) used data from the 1950 to 1990 Census for twenty-four different states. Finally, Fitzpatrick (2012) used restricted-access data from the 2000 Census for all fifty-states. In each of these studies, the sample sizes were much larger than in Fitzpatrick (2010). For the full sample of mothers of four-year-olds with no younger children, Fitzpatrick (2010) found that a movement from no availability of public prekindergarten to full implementation decreased the probability of employment in the previous year by 0.5 percentage points, with a standard error of 1.1 percentage points.17 Thus, the 95 percent confidence interval around the estimate is -1.7 to 2.7 percentage points. While my estimate of 4.4 percentage points does not fall within this confidence interval, the confidence interval constructed around my estimate, 1.6 to 7.2 percentage points, does intersect hers. Therefore, my results are not necessarily in conflict with those of Fitzpatrick (2010). 16 These were calculated under the assumption that the covariance between the two estimates is 0. Therefore, the test-statistic is simply the difference in the coefficients divided by the square root of the sum of the variances of the estimates. 17 It is worth noting that the employment variable employed in this study was employment at the time of the Census. Employment in the previous year is the variable in Fitzpatrick (2010) that is most comparable to any in this study. 19
  • 20. Even in the case that the insignificant results of Fitzpatrick (2010) cannot be explained by the small sample size and subsequently large standard errors, it is not surprising that they are different from those of this study. Fitzpatrick (2010) examined the effects of universal prekindergarten availability, focusing upon mothers in Georgia and Oklahoma. This study focuses upon mothers in ten different states, nine of which do not support universal PK initiatives. Georgia is the only state within my sample that aims for universal PK, while the rest focus upon at-risk populations (State of Preschool Yearbook 2003). Thus, in the case that universal prekindergarten initiatives have different effects than programs aimed towards at-risk populations, we would expect different results from this study when compared to the results of Fitzpatrick (2010).18 Finally, it important to note that there is little concern that the responses of mothers of four-year-olds with no younger children are due to initial differences in levels of maternal labor supply across counties, since the estimates for all the samples of mothers of two and seven-year- olds are not significantly different from zero or are significantly negatively different from zero. If anything, these results suggest that the DD estimates for mothers of four-year-olds and no younger children understate the true effect of increasing availability of PK on labor supply. If we conduct t-tests for the difference in parameter estimates between the coefficients for the sample of mothers of four-year-olds with no younger (Column 1) and the samples of mothers of two and seven-year-olds with no younger, the majority of the test-statistics are greater than 1.96. This implies that the parameter estimates for mothers of four-year-olds are significantly larger than those for mothers of two and seven-year-olds at the 5 percent level19 , and that the labor supply responses for those mothers of four-year-olds are statistically significantly larger. 18 In fact, in a separate set of regressions, I restricted my sample to only those mothers in Georgia, and found insignificant effects of universal PK on the full sample of mothers of four-year-olds and each one of the subsamples employed in this study. 19 These were again calculated under the assumption that the covariance between the two estimates is 0. 20
  • 21. C. Effects for Mothers of Four-Year-Olds with Younger Children Table 6 reports the estimates of the effects of prekindergarten availability on mothers of four-year-olds with younger children.20 The first set of estimates in Column 2 for the full sample of mothers show statistically significant increases of 3.6 percentage points in the probability of labor force participation and 4.2 percentage points in the probability of employment associated with a movement from no availability of prekindergarten to full implementation. The coefficient on labor force participation is statistically significant at the 5 percent level and the coefficient on employment is statistically significant at the 1 percent level. The second set of estimates for married mothers are similar in magnitude but less precise. The coefficients in Column 2 imply a 3.6 percentage point increase in the probability of labor force participation and a 4.1 percentage point increase in the probability of employment. The estimate for labor force participation is statistically significant at the 10 percent level, while the estimate for employment is statistically significant at the 5 percent level. The third set of estimates for single mothers are large but statistically insignificant. Column 2 reports coefficients of 4.1 percentage points for labor force participation and 3.5 percentage points for employment for single mothers of four-year-olds with younger children. Again, the test-statistics for the difference in parameter estimates between married and single moms for labor force participation (-0.089) and employment (0.096) suggest that the labor supply responses for these mothers are not statistically significantly different. These estimates imply large effects of prekindergarten programs on the labor supply of mothers with a four-year-old and other younger children. These results are similar to those of Cascio (2006; 2009), who found that the samples of mothers of five-year-olds with younger children exhibited a large response to increased kindergarten availability. These results are 20 Note that the first column is for reference – it is simply the results reported in the first column of table 5. 21
  • 22. opposite of those of Fitzpatrick (2010) for mothers of four-year-olds with younger children, possibly for the same reasons discussed above. Just as with the samples of mothers of four-year-olds and no younger children, there is little concern that these estimated increases in labor supply are due to differences in the initial levels of maternal labor supply across counties, since the estimates for mothers of two and seven- year-olds with younger children are either not statistically significant from zero or are statistically negatively significant from zero. Again, this suggests that the difference-in- difference estimates for mothers of four-year-olds with younger children may understate the true effect of increasing availability of PK on labor supply. D. Placing these Results in Context Some of the previous literature surrounding maternal labor supply and public schooling focuses upon elasticity estimates, calculating the elasticity of a mother’s labor supply with respect to child enrollment. With a little work, the coefficients in this paper can be transformed into elasticity estimates. The regressions in Table 4 examine the effects of increasing availability to public PK on enrollment of four-year-olds, whereas the regression in Table 5 examines the effects of increasing availability to public PK on the labor supply of mothers. This first set of regressions is simply a first-stage estimation for the regression of maternal labor supply on child enrollment, and thus we can find the elasticity estimate by dividing the coefficient for the effect on employment by the coefficient for the effect on enrollment. Thus, the child enrollment elasticity of employment for the full sample of mothers of four-year-olds with no younger children (Column 1 of Table 6) is 0.73 (0.044/0.060). In terms of child enrollment, this implies that for every ten children that are enrolled in public PK, at least seven mothers of four-year-olds 22
  • 23. with no younger children enter the workforce. Cascio (2009) found a similar but smaller effect for kindergarten enrollment – for every ten children enrolled in public kindergarten, at least 3 mothers of five-year-olds entered the workforce. E. Extensions Throughout this paper, I have presented estimates using a measure of the percentage of public schools in a district offering prekindergarten that is unweighted, either by the number of students within the school or the number of PK seats offered. This is an important distinction, because what really matters is how many mothers in a school catchment area have public PK available to them, which is identified by weighting this measure. In an alternative specification, I weighted each school by the number of kindergarten students in the school, where the number of kindergarten students served as a proxy for the number of PK seats available. The number of kindergarten students was used in place of the number of prekindergarten students because in many schools, prekindergarten programs were offered but no students enrolled. Estimates using this weighted measure are effectively the same as those using the unweighted measure.21 Throughout this paper, I have also operated under the assumption that family size is exogenous to the model. While it might be unlikely, it is possible that the opportunity for a mother to place her child in public PK could be the difference that leads to an additional birth. To test this hypothesis, I estimated model (1) without controls for the number and age of other children in the household. Estimates for these regressions are very similar to the model in which controls were included.22 One of the questions I initially set out to answer was whether the labor supply response to 21 Results are also robust to the specification of a binary variable for any PK or none within a district. 22 Results were also robust to the specification in which the remainder of maternal covariates—race and ethnicity, age, marital and veteran status, and whether the mom lives in a large city—were removed from the model. 23
  • 24. increasing PK was the same for all mothers, or whether there was heterogeneity among different samples of mothers. While we have seen heterogeneity in labor supply response based off marital status, one could also imagine that a mother’s labor supply response would differ at differing income levels. Since income is an outcome of the model and is thus endogenous, it is not appropriate to separate mothers into samples based off income level. However, we can allow education level to serve as a proxy for income level, and separate mothers based off educational attainment. My final set of regressions estimates model (1) for samples of low-educated and high-educated mothers, where low-educated mothers are considered to be those that have less than a college degree and high-educated mothers are those with a college degree. For all but one of these samples, the coefficients are not statistically significantly different from zero. The only exception is for the coefficient on employment for the low-educated sample of mothers of four- year-olds with no younger and no seven-year-old, which is a 4.5 percentage point increase, significant at the 10 percent level. In general, the estimates are larger for high-educated mothers, suggesting that highly educated mothers of four-year-olds show greater response to increasing PK than low-educated mothers.23 While it might be expected that low-educated mothers would show a greater response to increasing PK, there are several explanations for why the results suggest that highly educated mothers show a greater response. On average, highly educated mothers are more likely to be married. The presence of the spouse as a potential caregiver might leave the highly educated mother less constrained in her labor supply choice, giving her a greater margin to respond to increasing PK. Alternatively, it is possible that highly educated mothers are more inclined to return back to work after giving birth. 23 The results for all regressions discussed within this section are available upon request. 24
  • 25. VI. Conclusions The results of this study are the first to report a statistically significant impact of increasing public PK availability on the labor supply of mothers of four-year-olds. They are different from the results obtained by Fitzpatrick (2010) in her study of mothers of four-year- olds, although I have given reasons why this inconsistency might exist. More importantly is that these results are consistent with the general body of literature surrounding the effects of public schooling on maternal labor supply. Gelbach (2002) and Cascio (2006; 2009) found large effects of kindergarten supply on the labor supply of married mothers of five-year-olds, which this study replicates for prekindergarten and married mothers of four-year-olds. This study finds that introducing pre-kindergarten into a school district is associated with a 3.8 percentage point increase in labor force participation and a 4.4 percentage point increase in employment by mothers of four-year-olds. These studies, along with Fitzpatrick (2012), also found large effects of kindergarten supply on the labor supply of single mothers of five-year-olds, which this study replicates for prekindergarten supply and single mothers of four-year-olds. The results from this study, combined with those of these previous studies, provide key information for a cost-benefit analysis of public preschool programs. This study, along with those of Gelbach (2002), Cascio (2006; 2009), and Fitzpatrick (2012), suggest that public preschool programs benefit both attendees and their mothers. The results from Fitzpatrick (2010) further imply that programs aimed towards at-risk populations may be more effective than universal preschool initiatives. The geographic limitations of both this study and Fitzpatrick (2010) call for caution in extrapolating the results of this study or other studies into broader contexts. The general body of literature implies that the benefits of public preschool, including public PK, outweigh the costs. As many public prekindergarten programs are still relatively new 25
  • 26. and are continuing to expand, however, future research is warranted in order to reaffirm these positive effects. Appendix 1 Restriction of Census and ACS Data Data was taken from the 5 percent samples of the 1990 and 2000 Census as well as the 2005 and 2006 American Community Survey, and was downloaded from IPUMS. The original data contained roughly thirty-three million observations. The process of constructing samples began with the dropping of all states that did not have county-based school districts, since school district is not reported in the Census or ACS. States that have county-based school districts were kept because mothers can more accurately be matched to school districts, since county is indicative of school district in these states. While county is not reported directly in the Census or ACS, both the PUMA number and the names of each county identified within the PUMA are reported. Thus, the next step included an examination of PUMA boundaries in 1990 as well as 26
  • 27. 2000 (PUMA boundaries were the same in 2000, 2005, and 2006) to identify what counties were located within each PUMA. Those PUMAs that consisted of multiple counties in either year were dismissed outright, since mothers in these PUMAs could only be assigned to a group of counties or school districts rather than a particular county or school district. Although each of the remaining PUMAs consisted of only one county, there were a number of PUMAs that identified the same county within a given state (i.e. for counties that had a population of 200,000 or more). These PUMAs were aggregated together and given the same county identification number. The final step included a comparison of the remaining counties with the reported school districts for both years. The Local Agency Universe Survey provides data on public education agencies, both by state and county. Thus, data from this source provided information on the number of school districts located in a given county within a state. The majority of the counties that remained at this point were inclusive of only one school district. However, there were several that included multiple school districts. The majority of these were counties that had their own school districts but contained a larger city that also had its own school district. If that large city was not aggregated into the same PUMA as the county and could instead be identified by its own PUMA, this was not a problem. However, if the city was not identified by its own PUMA but was instead aggregated into the same PUMA as the county it was located in, it became problematic. For counties like these, it is hard to match mothers to a school district because the mother could enroll her child in either the county or city school district. Therefore, counties of this type were dropped. Examples of these situations can be seen in looking at Baltimore County in Maryland and Ouachita Parish in Louisiana. Baltimore County in Maryland contains both Baltimore County 27
  • 28. School District and Baltimore City School District. Baltimore County and Baltimore City can both be identified by a distinct group of PUMAs, however, and were therefore included in the sample. Ouachita Parish contains both Ouachita Parish School District and Monroe City School District. Both Ouachita Parish and Monroe City are both identified only within the same PUMA. Thus, mothers in this PUMA may enroll their child in either one of the two school districts in Ouachita Parish, making the matching of mothers to school districts impossible without further information. For this reason, Ouachita Parish was dropped from the sample. A total of fifty-nine districts in ten states were used in this study. Their names, as well as the names of the states they are located in, are given in Table 2. In addition, Appendix Table 1 gives examples of different types of PUMAs discussed above. Appendix 2 Census and ACS Samples After limiting the data to the fifty-nine districts in the ten states (see Appendix 1), all children between the ages of 0 to 17 were matched to their mothers for each year. Additionally, any non- spousal members of the household that were 18 years or older were matched to mothers. Samples were then limited to mothers of two and four-year-olds aged 18 to 45 at the time of the Census. Further limitations resulted in full samples of married and single mothers. Mothers were considered to be single if reported to be the head of a household where no spouse was present. Households where more than one family was present were dropped from the samples, since estimates within these households may be affected by a number of unobservable characteristics (this resulted in a small number of mothers being excluded from the sample—0.24 28
  • 29. percent of the sample in the most extreme case). Furthermore, the samples were restricted to mothers of only one two, four, or seven-year-old in order to maintain comparability (Gelbach 2002). Appendix 3 Prekindergarten Availability and Construction of percent_pk Data on grade span was used to construct the variable of interest, percent_pk, the fraction of public schools within a district that offer prekindergarten programs. Data was taken from the Public Secondary-Elementary School Universe Survey for the appropriate years and was downloaded from the National Center for Education Statistics. Grade span is reported as the low and high grades offered within a school. For each district in a given year, the fraction of public schools offering prekindergarten programs was calculated as the sum of operating schools offering prekindergarten (PK-K, PK-1, PK-2, PK-3, PK-4, PK-5, PK-6, PK-7, PK-8, PK-9, PK- 12) over the sum of operating schools offering prekindergarten or kindergarten (PK-K, PK-1, PK-2, PK-3, PK-4, PK-5, PK-6, PK-7, PK-8, PK-9, PK-12, K-1, K-2, K-3, K-4, K-5, K-6, K-7, K-8, K-9, K-12). Any changes in the availability of prekindergarten will not be recognized until the beginning of the school year, which takes place in the fall. Consequently, changes in employment that are a response to the availability of prekindergarten will not be recognized until the following year. For this reason, percent_pk for a given year was taken from the beginning of the previous year’s school year. This means that percent_pk for 1990 was taken from the 1989- 1990 school year, for 2000 from the 1999-2000 school year, for 2005 from the 2004-2005 school year, and for 2006 from the 2005-2006 school year. Percent_pk was matched to mothers in 29
  • 30. district d in year t with a unique identification number that was coded as a combination of the state FIPS identification number and ICPSR county identification number in each year. It is important to note that in four of the fifty-nine districts (Broward County, Orange County, Hillsborough County, and Mecklenburg County) the number for percent_pk was replaced in the year 2000 because there appeared to be reporting error. For Osceola County School District, percent_pk was replaced in both 1990 and 2000. This change was made because the time-path for percent_pk appeared to follow an unnatural path over the time period for these districts. For example, Orange County School District in Florida reported 42.7 percent in 1990, 0.03 percent in 2000, and 91.0 percent in 2005/2006. For each one of these districts, the number used as replacement was the average of percent_pk for the year before and after the replaced year. References Abbott-Shim, M., Lambert, R., & McCarty, F. “A comparison of school readiness outcomes for children randomly assigned to a Head Start program and the program’s wait list.” Journal of Education for Students Placed at Risk 8.2 (2003): 191–214. Averett, Susan L., Elizabeth Peters, and Donald M. Waldman. “Tax Credits, Labor Supply, and Child Care.” The Review of Economics and Statistics 79.1 (1997): 125-135. Berger, Mark and Dan Black. “Child Care Subsidies, Quality of Care, and the Labor Supply of Low-Income, Single Mothers.” The Review of Economics and Statistics 74.4 (1992): 635-642. Baker, Michael et al. “Universal Child Care, Maternal Labor Supply, and Family Well-Being.” Journal of Political Economy 116.4 (2008): 709-745. Barnett, W. Steven et al. “The State of Preschool 2009.” National Institute for Early Education Research. Barnett, W. Steven et al. “The State of Preschool 2003.” National Institute for Early Education Research. 30
  • 31. Berlinski, Samuel, and Sebastian Galiani. “The Effect of a Large Expansion of Pre-Primary School Facilities on Preschool Attendance and Maternal Employment.” Labour Economics 14.3 (2007): 665-680. Blau, David M. and A.P. Hagy. “The Demand for Quality in Child Care.” Journal of Political Economy 106.1 (1998): 104-146. Blau, David M. and Philip K. Robins. “Child-care Costs and Family Labor Supply.” The Review of Economics and Statistics 70.3 (1988): 374-381. Blau, David M. and Philip K. Robins. “Fertility, Employment, and Child-Care Costs.” Demography 26.2 (1989): 287-299. Cascio, Elizabeth. “Maternal Labor Supply and the Introduction of Kindergartens into American Public Schools.” The Journal of Human Resources (2009). Cascio, Elizabeth. “Public Preschool and Maternal Labor Supply: Evidence from the Introduction of Kindergartens into American Public Schools.” National Bureau of Economic Research (2006). Connelly, Rachel and Jean Kimmell. “The Effect of Child Care Costs on the Labor Force Participation and Welfare Recipiency of Single Mothers: Implications for Welfare Reform.” Upjohn Institute Working Paper No 01-69 (2001). Kalamazoo, MI: W.E. Upjohn Institute for Employment Research. Connelly, Rachel. “The Effect of Child Care Costs on Married Women’s Labor Force Participation.” The Review of Economics and Statistics 74.1 (1992): 83-90. Connelly, Rachel. “The Effect of Child Care Costs on the Labor Force Participation and AFDC Recipiency of Single Mothers.” Institute for Research on Poverty Discussion Paper no. 920-90 (1990). Currie, Janet and Thomas, Duncan. “Does Head Start Make a Difference?” The American Economic Review 85.3 (1995): 341-364. Fitzpatrick, Maria Donovan. “Preschoolers Enrolled and Mothers at Work? The Effects of Universal Prekindergarten.” Journal of Labor Economics 28.1 (2010): 51-85. Fitzpatrick, Maria Donovan. “Revising Our Thinking about the Relationship between Maternal Labor Supply and Preschool.” Forthcoming, Journal of Human Resources (2012). Garces, Eliana, Duncan Thomas, and Janet Currie. “Longer Term Effects of Head Start.” American Economic Review 92.4 (2002): 999-1012. Gelbach, Jonah B. “Public Schooling for Young Children and Maternal Labor Supply.” American Economic Association 92.1 (2002): 307-322. 31
  • 32. Hotz, V. Joseph, and M. Rebecca Kilburn. “The Demand for Child Care and Child Care Costs: Should We Ignore Families with Non-Working Mothers?” Working Paper Series 92-1 (1991), Harris School, University of Chicago. Kimmel, Jean. “Child Care Costs as a Barrier to Employment for Single and Married Mothers.” The Review of Economics and Statistics 80.2 (1998): 287-299. Kimmel, Jean. “The Effectiveness of Child-Care Subsidies in Encouraging the Welfare-to-Work Transition of Low-Income Single Mothers.” The American Economic Review 85.2 (1995): 271-275. Micalopoulos, Charles et al. “A Structural Model of Labor Supply and Child Care Demand.” The Journal of Human Resources 27.1 (1992): 166-203. Oden, S., Schweinhart et al. (2000). Into adulthood: A study of the effects of Head Start. Ypsilanti, MI: High/Scope Press. Puma et al. “Head Start Impact Study: Final Report.” U.S. Department of Health and Human Services, Office of Planning, Research and Evaluation (2010). Ribar, David C. “Child Care and the Labor Supply of Married Women Reduced Form Evidence.” The Journal of Human Resources 27.1 (1992): 134-165. Schlosser, Analia. “Public Preschool and the Labor Supply of Arab Mothers: Evidence from a Natural Experiment.” The Hebrew University of Jerusalem, Department of Economics (2005). 32
  • 33. Figure 1 Fraction of Three and Four-Year-Olds Enrolled in Prekindergarten Programs Source: October Supplement of the Community Population Survey 33
  • 34. Figure 2 Fraction of Public Schools in the United States Offering Prekindergarten Programs Sources: Public Secondary-Elementary School Universe Survey Note: See Appendix 2 for a description of how this fraction was calculated. 34
  • 35. Figure 3 Labor force participation rate of mothers by age of their youngest child Sources: March Supplement of the Community Population Survey 35
  • 36. Figure 4 Change in the percentage of schools within a school district offering public prekindergarten Source: Public Secondary-Elementary School Universe Survey Notes: Negative changes in percent_pk are included in the 0 to 0.1 category. 36
  • 37. Table 1 Names of States and School Districts Used State District State District Alabama: Mobile County Maryland: Anne Arundel County Montgomery County Baltimore City Baltimore County Florida: Alachua County Carroll County Broward County Charles County Charlotte County Frederick County Duval County Harford County Hernando County Howard County Hillsborough County Montgomery County Lake County Prince Georges County Manatee County Washington County Marion County Martin County North Carolina: Pitt County Orange County Forsyth County Osceola County Mecklenburg County Pasco County Onslow County Pinellas County Wake County Polk County St. Lucie County South Carolina: Aiken County Volusia County Charleston County Greenville County Georgia: Bibb County Horry County Chatham County Clayton County Tennessee: Knox County Fulton County Montgomery County Sumner County Kentucky: Jefferson County Virginia Alexandria City Louisiana: Jefferson Parish Arlington County Lafayette Parish Chesapeake City Rapides Parish Chesterfield Hampton City Henrico County Newport News City Norfolk City Richmond City 37
  • 38. Table 2 Demographic Characteristics of Mothers of Four-Year-Olds, by inclusion in sample, 2005/06 ACS Variable In Sample Not In Sample P-Value Demographic characteristics Age 33.1 32.3 0 White 0.58 0.65 0 Black 0.23 0.09 0 Hispanic 0.12 0.18 0 Other Race 0.07 0.08 0.49 Married 0.75 0.65 0 Education In School 0.10 0.13 0 Less Than High School 0.09 0.12 0.02 High School 0.30 0.35 0.06 Some College 0.24 0.25 0.48 College Graduate 0.37 0.29 0.03 Labor Supply Usual Hours Worked per Week 25.0 24.8 0.54 Weeks Worked Last Year 29.7 29.7 0.996 Employed 0.59 0.60 0.58 In Labor Force 0.64 0.65 0.35 In Poverty 0.15 0.18 0.12 On Welfare 0.03 0.04 0.23 Observations 4773 85816 Notes: P-Values reported are for the difference in means. Table 3 38
  • 39. Demographic Characteristics of Mothers of Two, Four, and Seven-Year Olds, by presence and age of younger children, 1990/2000 Census 5% PUMS, 2005/06 ACS Demographic Characteristics Variable 4-Year Olds (No Younger) 4-Year Olds (No Younger, No 7-Year Olds) 2-Year Olds (No Younger No 4 or 7-Year Olds) 7-Year Olds (No Younger) Age 32.9 32.8 30.5 35.6 White 0.63 0.62 0.65 0.63 Black 0.25 0.26 0.23 0.26 Hispanic 0.08 0.07 0.07 0.07 Other Race 0.04 0.05 0.05 0.04 In Large City 0.04 0.04 0.04 0.04 Education: In School 0.10 0.10 0.10 0.09 Less Than High School 0.09 0.09 0.09 0.09 High School 0.36 0.37 0.35 0.37 Some College 0.29 0.29 0.28 0.29 College Graduate 0.26 0.25 0.28 0.25 Outcomes: Usual Hours Worked/Week 27.8 28.4 26.6 30.0 Weeks Worked Last Year 32.9 33.6 31.2 35.9 Employed 0.66 0.67 0.63 0.72 In Labor Force 0.70 0.72 0.68 0.76 In Poverty 0.14 0.13 0.13 0.12 On Welfare 0.05 0.05 0.05 0.04 Number & Age of Children: 1-Year Old 0 0 0 0 2-Year Old 0 0 1 0 3-Year Old 0 0 0.08 0 4-Year Old 1 1 0 0 5-Year Old 0.07 0.07 0.15 0 6-Year Old 0.16 0.17 0.12 0 7-Year Old 0.15 0 0 1 8-12 Year Old 0.45 0.46 0.26 0.67 13-17 Year Old 0.17 0.18 0.12 0.32 18 or older Non-Spouse 0.43 0.45 0.40 0.63 Preschool Availability: % Schools in District With Prekindergarten 0.43 0.43 0.43 0.44 Sample Size 15945 13562 16512 13104 Table 4 39
  • 40. Coefficients on the percentage of public schools offering prekindergarten for enrollment of three and four-year olds, 1990/2000 Census 5% PUMS, 2005/06 ACS Difference-in-Difference Estimates Sample Total Enrollment Public Enrollment Private Enrollment 3-Year Olds Percent_pk 0.019 (0.023) 0.024 (0.017) -0.005 (0.020) Sample Mean 0.41 0.15 0.26 Sample Size 18562 4-Year Olds Percent_pk 0.048* (0.026) 0.060** (0.025) -0.013 (0.023) Sample Mean 0.61 0.27 0.34 Sample Size 15945 Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **, and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively. In addition to the variable of interest (percentage of public schools offering prekindergarten), each regression includes controls for year and district effects as well as a vector of maternal controls including one for race and ethnicity, age and number of children in the household, a quadratic in maternal age, education level, indicators for veteran and marital status, and an indicator for whether or not the mom lives in a large-sized city. Table 5 40
  • 41. Coefficients on the percentage of public schools offering prekindergarten for Mothers Two, Four, and Seven-Year Olds by presence and age of younger children, 1990/2000 Census 5% PUMS, 2005/06 ACS Difference-in-Difference Estimates 4-Year Olds (No Younger) 4-Year Olds (No Younger, No 7- year Olds) 2-Year Olds (No Younger, No 4 or 7-Year Olds) 7-Year Olds (No Younger) Sample In labor force Employed In labor force Employed In labor force Employed In labor force Employed Full Sample Percent_pk 0.038*** (0.014) 0.044*** (0.014) 0.037*** (0.014) 0.048*** (0.016) -0.016 (0.019) -0.023 (0.022) -0.016 (0.020) -0.027 (0.018) Sample Mean 0.70 0.66 0.72 0.67 0.68 0.63 0.76 0.72 Sample Size 15945 13562 16512 13104 Married Percent_pk 0.043** (0.020) 0.055*** (0.017) 0.043** (0.020) 0.058*** (0.027) -0.0002 (0.022) -0.0002 (0.024) -0.004 (0.026) -0.019 (0.027) Sample Mean 0.67 0.65 0.68 0.66 0.66 0.63 0.73 0.71 Sample Size 11319 9446 12205 8877 Single Percent_pk 0.046 (0.041) 0.021 (0.045) 0.040 (0.037) 0.019 (0.044) -0.080* (0.046) -0.151*** (0.046) -0.055* (0.031) -0.059 (0.040) Sample Mean 0.80 0.72 0.81 0.73 0.75 0.66 0.83 0.76 Sample Size 3582 3165 2983 3381 Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **, and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively. In addition to the variable of interest (percentage of public schools offering prekindergarten), each regression includes controls for year and district effects as well as a vector of maternal controls including one for race and ethnicity, age and number of children in the household, a quadratic in maternal age, education level, indicators for veteran and marital status, and an indicator for whether or not the mom lives in a large-sized city. Table 6 41
  • 42. Coefficients on the percentage of public schools offering prekindergarten for Mothers of Four- Year Olds by presence of younger children, 1990/2000 Census 5% PUMS, 2005/06 ACS Difference-In-Difference Estimates 4-year olds (No Younger) 4-Year Olds (Any Aged Children) Sample In labor force Employed In labor force Employed Full Sample Percent_pk 0.038*** (0.014) 0.044*** (0.014) 0.036** (0.014) 0.042*** (0.014) Sample Mean 0.77 0.66 0.64 0.59 Sample Size 15945 26938 Married Percent_pk 0.043** (0.020) 0.055*** (0.017) 0.036* (0.019) 0.041** (0.017) Sample Mean 0.67 0.65 0.61 0.58 Sample Size 11319 19764 Single Percent_pk 0.046 (0.041) 0.021 (0.045) 0.041 (0.053) 0.035 (0.049) Sample Mean 0.80 0.72 0.74 0.64 Sample Size 3582 5485 Notes: Standard errors are reported in parenthesis and are clustered at the county level. *, **, and *** denote statistical significance at the 10 percent, 5 percent, and 1 percent levels, respectively. In addition to the variable of interest (percentage of public schools offering prekindergarten), each regression includes controls for year and district effects as well as a vector of maternal controls including one for race and ethnicity, age and number of children in the household, a quadratic in maternal age, education level, indicators for veteran and marital status, and an indicator for whether or not the mom lives in a large-sized city. Appendix Table 1 Examples of PUMAs encountered during restriction process and action taken 42
  • 43. County Puma Treatment Baltimore 00501 Aggregate 00502 Aggregate 00503 Aggregate 00504 Aggregate 00505 Aggregate 00506 Aggregate 00507 Aggregate Baltimore City 00801 Aggregate 00802 Aggregate 00803 Aggregate 00804 Aggregate 00805 Aggregate 00806 Aggregate Carroll 00400 Keep but don't aggregate Cecil 00700 Delete Kent 00700 Delete Queen Anne's 01300 Delete Caroline 01300 Delete Talbot 01300 Delete Dorchester 01300 Delete Wicomico 01400 Delete Somerset 01400 Delete Worcester 01400 Delete Notes: This figure provides examples of the different types of counties that were encountered in the restriction process and the treatment that was given to each one. If a county was large enough to be uniquely identified by more than one PUMA, all of the PUMAs that identified that county were aggregated together and given the same identification number. If a county was large enough to be uniquely identified by only one PUMA, the PUMA identifying that county was kept, with no further action necessary. Those small counties that were in the same PUMA as another distinct county were dropped from the sample. 43