How much is too much?

The Effects of Duration and Intensity of Child Care Experiences on

Childrens Social and Cognitive Development

Susanna Loeb
Margaret Bridges
Daphna Bassok
Bruce Fuller
Russ Rumberger

May 2005
Introduction

Much debate surrounds the issue of how best to raise childrens school achievement.

Research on cognitive development indicates that one of the most effective interventions may be

formal center-based care or preschool participation, before formal schooling has begun (Shonkoff &

Phillips, 2000; Heckman, 2000). In fact, approximately two-thirds of four-year-olds attend center

programs before starting kindergarten (U.S. Department of Education, 2003), although the extent of

exposure to these programs and the quality of the care varies dramatically. This paper extends

previous research conducted by Magnuson, Ruhm, & Waldfogel (2004) to consider the effects of

different child-care arrangements on childrens social and cognitive skills at the start of kindergarten,

focusing on how the intensity of childrens participation is related to program effects. It also

addresses differences in program effects across income groups. These issues are germane to recent

debates over implementing universal preschool, as policy makers consider age-eligibility and whether

programs should be half-day or full-day.

In this paper we address the following questions:

Does center care and/or Head Start improve childrens cognitive and social-

behavioral outcomes at kindergarten entry, after considering their family background

and neighborhood characteristics?

Does age at entry or intensity of exposure to center care affect childrens cognitive

and social-behavioral outcomes?

Achievement Gaps: School readiness is of particular concern for those children at risk for

low school achievement. Disparities between groups in a wide range of domains are already evident

at kindergarten entry, across social class and ethnic groups (Jencks & Phillips, 1998; Reardon, 2003;

Rumberger and Arellano, 2003; Wertheimer, Croan, Moore, & Hair, 2003). For example, the

difference between African-American and White children in their early language and cognitive
development is equal to the approximate amount that children learn during two to three months of

kindergarten (Jencks & Phillips, 1998; Reardon, 2003). English-proficient Latino five-year-olds in

California were also estimated to be about 0.38 of a standard deviation (SD)or two to three

monthsbehind White youngsters in pre-reading and math skills (Rumberger and Arellano, 2003).

Similar achievement gaps exist between children from poor and affluent families. Research

indicates that children in the lowest socioeconomic group are several months behind their middle-

class peers in pre-reading and pre-math skills at kindergarten entry. This chasm almost triples when

poor children are compared to the most affluent fifth (Bridges, Fuller, Rumberger, & Tran, 2004).

These disparities between groups often grow even larger over the course of their schooling (Jencks

& Phillips, 1998; Fryer & Levitt, 2004).

Child-Care Arrangements that Contribute to Outcomes: Research indicates that attending

high-quality child-care boosts childrens developmental trajectories, leading to speculation about the

possibility of its closing achievement gaps (Barnett, 1995; Bridges et al., 2004; Brooks-Gunn, 2000;

Currie, 2001; Waldfogel, 2002). Researchers have compared various care arrangements to determine

which might hold the most promise for improving childrens cognitive and social-behavioral

outcomes; center-based programs appear to offer the most benefits (Loeb, Fuller, Kagan, & Carrol,

2004). Participating in center-based care has been associated with immediate and long-term benefits

for children (Barnett, 1995; Campbell & Ramey, 1995). Head Start, which is differentiated by having

particular regulations and serving a distinct population of the most disadvantaged children, also has

been shown to improve childrens outcomes in both the short and long term, though the evidence is

mixed (Currie & Thomas, 2000; Garces, Thomas, & Currie, 2002).

Children from socioeconomically disadvantaged homes often exhibit the greatest gains from

participating in center programs (Burchinal, Campbell, Bryant, Wasik, & Ramey, 1997; Campbell &

Ramey, 1994). Center-based care may also benefit English-language learners; given that research has

2
indicated these children are less likely than others to experience home literacy practices found to

facilitate early learning (August & Hakuta, 1999; Loeb, Fuller, & Kagan, 2004; Snow, Burns, &

Griffin, 1998).

Some studies have also found that although children experience cognitive gains in centers,

other aspects of their school readinesspositive social behaviors and impulse controlsuffer

(Belsky, 2001; NICHD ECCRN, 2003). While some evidence suggests that increases in hours spent

in center care increases behavioral problems (NICHD ECCRN, 2003), it is less clear what the

tradeoff is between social problems and cognitive gains as hours increase. In addition, it is not clear

how these effects differ for children with different background characteristics.

Differences in Types of Care: Children participate in a wide variety of care arrangements,

which are differentially associated with their outcomes. In this paper, we consider four mutually

exclusive arrangements: center care, Head Start care, parental care, and non-parental care (which

includes non-parent relatives or non-relatives, such as babysitters). Though parents in the data we

use were asked to specify the type of center care their child attendeda day-care center, a

preschool, a nursery school, or a Pre-K programdifferences between the four types are often

difficult for parents to distinguish so we do not rely on these self-reports to distinguish center types

(Kisker, Hofferth, Phillips, & Farquhar, 1991).

Differences in Access: In spite the range of child-care arrangements available, childrens

access varies markedly in predictable ways. Participation in center-based programs increases with

social class: children from affluent families are much more likely to receive center care than children

from other SES groups (OBrien-Strain, Moye, & Sonenstein, 2003; Reardon, 2004). For example,

more than 70 percent of upper middle-class children attend center-based programs before starting

kindergarten, compared to 45 percent of those from low-income families (Hofferth, Shauman,

Heake & West, 1995).

3
 However, with the expansion of Head Start and state preschool programs, children from

low-income families are increasingly participating in center-based programs before school entry

(Smith, 2003). In fact, it may be that working-class families have less access to preschool than their

poor counterparts because they are just above the income threshold for child care subsidies, yet

cannot afford the high monthly fees (Fuller & Huang, 2003; Fuller, Livas, & Bridges, 2005).

Ethnic disparities in preschool access remain dramatic: Latino parents enroll their children in

preschool centers at a rate 23 percentage points below the enrollment rate of African-American

children, and 11 percent below Whites even after taking into account maternal employment status

(Liang, Fuller & Singer, 2000). Asian American children, too, participate in preschool at substantially

lower rates than do Whites or African Americans (NCES, 1998).

Differences in Exposure: The effect of different non-parental care arrangements on

children is likely based in part on the amount of time children are exposed to them. This might be

thought of as a dosage effect and can be conceptualized as the age at entry and intensity of

attendance. Children entering at younger ages or attending for more hours per week may exhibit

greater benefitsor greater detrimentsthan those with later or less exposure. Little is known

about the amount of exposure that maximizes cognitive gains for children. Research on

interventions to improve cognitive outcomes suggests early intervention is best (Shonkoff &

Phillips, 2000). Some research, however, has indicated that children spending long hours or more

months of the year in center care exhibit increases in problem behavior (Bates, Marvinney, Kelly,

Dodge, Bennett, & Pettit, 1994; Belsky, 2001; NICHD ECCRN, 2003).

Additions to the Literature: In addition to using a representative sample of English-

proficient children, this study extends previous research on child care in the following ways. First, it

assesses the duration effect of center-care experiences to ask, what is the optimal age at which to

enter center programs? Second, it looks at the intensity effect, asking whether there is some

4
number of hours per week of attendance that has a more positive effect than either greater or lesser

exposure. Third, we look at both these effects by the income of the childs family, asking whether

center care experiences are more or less important for children from different groups. All three of

these questions are central to the current discussion of universal preschool. In addition, we run

numerous specification checks, including an instrumental variables analysis and propensity score

matching to assess selection bias.

Data

Data for this study come from the Early Childhood Longitudinal Study, Kindergarten Class

of 1998-99 (ECLS-K). ECLS-K follows a nationally representative cohort of children from

kindergarten entry through fifth grade. Our study includes 14,162 children who entered kindergarten

for the first time in 1998. We exclude children who were missing scores on any of the assessments,

as well as those children for whom we have no child care information.

Child Care Measures: In the fall of the kindergarten year, parents were asked a series of

questions regarding their childs care arrangement in the year prior to kindergarten entrance. Based

on their responses, we created four mutually exclusive child-care arrangement groups; center care,

Head Start care, parental care and non-parental care (which includes care by non-parent relatives or

non-relatives such as a babysitter). Though parents were asked to specify whether their child went to

a day care center, a preschool, a nursery school or a Pre-K program, we were concerned that the

differences between the four center types were difficult for parents to distinguish. We therefore

created a single center care group that includes children who went to any type of child care center

with the exception of Head Start. If parents indicated that their child received care at multiple

settings, we coded them as follows: if a child attended center care in combination with parental or

non-parental care, they were placed in the center care group. Similarly, if a child attended Head Start

5
in combination with parental or non-parental care, he was placed in the Head Start group. Finally, if

a child participated in center care as well as a Head Start program, we placed her in the group in

which she spent more hours per week.1

In order to test whether the amount of center exposure made a difference in childrens

outcomes, we created a series of age of entrance and intensity variables. To get at the impact of early

entrance, we created indicator variables for the childs age when he first entered center care (age 0-1,

1-2, 2-3, 3-4, 4-5, greater than 5, and unknown). We also created two intensity variables: the first

dummy variable indicates whether the child attended center care for 15 to 30 hours per week and

nine months out of the year; the second raises the weekly hours to at least 30 hours per week and at

least nine months. We created these distinctions in keeping with the typical half-day program and

the typical full-day program.

Outcomes: In the fall of kindergarten, one-on-one child assessments were conducted to

assess reading and mathematics ability. The reading assessment measures a variety of skills including

print familiarity, letter and word recognition, beginning and ending sounds, rhyming sounds,

vocabulary, and comprehension. The math test evaluates each childs knowledge of numbers as well

as their spatial sense and problem solving abilities2. The assessments in each subject area were

administered in two stages, the first involved a routing test, and the second involved items at the

appropriate difficulty level. In our analysis we use standardized T-Scores. These scores are

transformations of ability scores that have been rescaled with a mean of 50 and a standard deviation

of ten. T-Scores are norm-referenced measures of achievement and provide an indicator of how

each individual child is doing relative to the national average.

1
We were also concerned that parents might misreport Head Start attendance. However, ECLA-K independently
confirms Head Start attendance. As a specification check we reassigned to the center care group those who reported
attending Head Start but were not confirmed as attending Head Start.
2
Early Childhood Longitudinal Study, Kindergarten Class of [Link]-K Base Year Data Files and Electronic
Code Book.

6
 In addition to assessing cognitive outcomes, our study considers childrens social skills as

reported by kindergarten teachers through a self-administered questionnaire. Teachers were asked

to evaluate the social skills of the sampled children in their classroom on a scale from one to four

with respect to their approaches to learning, self-control and a variety of interpersonal skills. Using

factor analysis we created a composite score that combines measures of self control, interpersonal

skills, and externalizing behavior (Chronbachs alpha of 0.87). The behavior score is standardized

with mean of zero and a standard deviation of 10.

Other Variables: In order for the results to be comparable with Magnuson, Ruhm, &

Waldfogel (2004), we use the same set of ECLS-K variables as controls in the analyses. Appendix

Table 1 lists these numerous measures. In addition, in order to capture neighborhood effects, we

include a set of zip code level variables, aggregated from the Long Form of the 2000 Decennial

Census. Appendix Table 2 lists these variables, capturing community characteristics.

Table 1 gives the means and standard deviations of a subset of the variables for the full

sample and separately for each child care type. We see that about 64 percent of the children

attended center care, compared with 17 percent in parental care, eight percent in Head Start and 12

percent in other non-parental care. Children in Head Start were somewhat more likely to come

from the southern region of the country and were substantially more likely to be black. Center care

is more evenly distributed across the country, but Hispanic children are less likely to attend. Large

differences in socio-economic status are evident across child care types, as well. Children in Head

Start have lower weight at birth, are more likely live in a single-parent family, are more likely to have

parents without high school degrees, are more likely to have participated in WIC, and come from

lower income families. Parents who provide the sole care for their children are, not surprisingly, less

likely to be single parents and less likely to work full time. Children in center care are more likely to

7
speak English only at home and have more highly educated parents. Children with other non-

parental care are more likely to come from families in which the mother works full time.

In addition to social status difference, children vary in home life by child care setting. For

example, children in Head Start have the least number of childrens books in their home, while those

in other centers have the most. Children in Head Start watch the most television, while those in

centers watch the least. Children in Head Start are most likely to be spanked, while those in center

care are least likely. Parents of children in Head Start are least likely to attend a parent-teacher

conference during kindergarten or volunteer at school; those in center care are most likely.

Methodology

Our goal is to assess the effects of child care on childrens social and cognitive development

at the start of kindergarten. Unfortunately, as noted above, there are substantial differences in the

characteristics of children across settings. The heart of our analysis relies on the rich measures of

children and families available in the ECLS-K to adjust for differences across child care settings

using a regression framework. Equation 1 summarizes this approach:

Yizs = 0 + C izs 1 + X izx 2 + Z zs 3 + s + izs (1)

The outcome (Y) of child (i) in zipcode (z) and state (s) is a function of child care type (C), child and

family characteristics (X), zipcode characteristics (Z), state fixed effects (), and a random and

normally distributed error term (). Child care type in the base model is a series of three dummy

variables for center care, Head Start and other non-parental care in comparison to parental care. In

the models assessing duration, child care type is expanded to include the duration of center care.

The center care dummy variable, in this case, is replaced by seven dummy variables measuring

starting center care at age zero to one year, one to two years, two to three years, three to four years,

four to five years, greater than five years, and start date unknown. In the model assessing intensity

8
the center care dummy is supplemented by mutually exclusive dummy variables for attendance of 15

to 30 hours per week for at least nine months per year and for attendance of at least 30 hours per

week for at least nine months per year.

There are some limitations of this approach. For example, regression often assumes a linear

relationship among variables when the relationship is meaningful but non-linear. The bias created

by this misspecification can be larger when there is less overlap across treatments, as is the case for

child care type. For example, we may estimate the effect of income on child outcomes using data

that is mostly over one income range; the group of children in this income range will be most

important for determining the estimate. If children in a particular care type (for example Head Start)

have much different income then we may apply estimates that are inaccurate for this group. We

mitigate the potential bias from misspecification by using multiple dummy variables instead of

continuous variables for measures such as education and income (see Appendix Table 1). In

addition we check whether we have adequately adjusted for these non-linearities and lack of overlap

by using statistical matching of children across child care type. We use kernel matching, a non-

parametric matching approach that creates matches for the treatment using (biweight) kernel

weighted averages of those not in the treatment (Heckman, Ichimura and Todd 1998).3

Analyses that assess treatment effects with non-experimental data are often subject to

omitted variables bias. As discussed above, we reduce this possibility in these analyses by using an

unusually rich set of control variables. In addition, we gathered a variety of zip code and state level

data with the goal of using instrumental variable (IV) models to adjust for selection. In particular,

we obtained counts of child care establishments and community organizations at the zip code level.

Given the extensive set of zip code level controls from the census, we posited that these measures

3
This gives very similar results to the simple procedure of predicting treatment probabilities, P, as a function of all
the other right-hand-side variables in Equation 1 using a probit and then creating a weight that is one for those in the
treatment and P/(1-P) for those not in the treatment.

9
would predict center use but not child outcomes. We also obtained a number of state-level

measures to capture state intervention. We created three variables that measure state level spending

on child care: each states 1999 spending on pre-Kindergarten, Head Start, and the child care

component of their Temporary Assistance to Needy Families (TANF) is divided by the number of

children under 5 years of age living in poverty4. Finally, we included measures for the income cutoff

for state child care assistance both as a level and as a percentage of the state median5. These

variables were meant to capture each states commitment to child care provision, particularly for

poor children. Unfortunately, while the difference between the IV estimates and the OLS estimates

are not statistically significant and the IV estimates are in the same direction as the OLS estimates,

the standard errors are too large to draw any meaningful conclusions from the IV analyses.

Results

Table 2 presents four models for each of the three student outcomes. Model 1 includes only

the child care variables; Model 2 introduces the family and child controls for ECLS-K; Model 3 adds

in zip code demographics; and Model 4 includes state fixed effects. Across outcomes, we see that

the inclusion of the ECLS-K controls dramatically changes the estimates of the Head Start

coefficient across all outcomes, as well as the estimates of the center care coefficients for reading

4
Head Start and Pre-Kindergarten spending figures are taken from Blank, Helen. Karen Schulman and Denielle
Ewan. Seeds of Success, Childrens Defense Fund, September 1999.
[Link]
TANF spending figures for 1999 are available at the US Department of Health and Human Services Website. See:
[Link]
Percentage of the population under five in poverty is taken from census data available from Geolytics: 2000 Long
Form DVD.
5
Data on 2001 income cutoffs for state child care assistance comes from
Schulman, Karen, Helen Blank. Child Care Assistance Policies 2001-2004: Families Struggling to Move Forward,
States Going Backward. National Womens Law Center. September 2004.
[Link]
For states that have a range of cutoffs throughout the state we used the lower end of the range, so that any estimates
of a relationship between cutoffs and child care choice is conservative.

10
and math. The addition of the zip code controls and the state fixed effects do not have as strong an

effect.

The results provide evidence that center care improves childrens reading and math skills but

also increases behavioral problems relative to parental care. The center care coefficient for reading

suggests that center care attendance increases reading skills by 1.1 points. The standard deviation in

reading is 10.0; thus the point estimate implies a .11 standard deviation increase. The results are very

similar for math. Center care is associated with a 1.2 point (.12 standard deviation) increase in math

performance. The estimatation of social development show negative effects of center care. Center

attendance in Model 4 is associated with a 0.095 point (measured in standard deviation units)

decrease in the behavioral index.

The Head Start effects estimated in Table 2 are not statistically different from zero for either

reading or math, though they do suggest a negative effect of Head Start on behavior. Children in

Head Start exhibit behavior that is about 0.12 standard deviations below children in parental care.

One caution about the Head Start effects is that additional controls, moving from Model 1 to Model

4 continue to increase the association between Head Start and all three child outcomes. This may be

an indication of continued omitted variables bias and that additional controls are needed to

adequately estimate Head Start effects.

As discussed above, we use both instrumental variables and propensity score matching to

test the specification of the models. Table 3 shows that our instruments were not successful in

accurately estimating child care effects. However, the statistical matching results are quite similar to

those discussed above. Relative, in this case, to both parental and other non-parental, non-Head

Start care center attendance is associated with a .13 standard deviation increase in reading and math

performance and decrease in the behavioral measure.

11
 Many studies have postulated that the effects of center care will differ across families. For

example, children in families with many resources for early math and reading may not benefit as

much from center care as those in families without such resources. Table 4 presents the results of

the Model 4 estimates from Table 2, but separately for children from families in the lowest income-

to-needs ratio quartile, the middle half and the top quartile. There is surprising little difference

across income groups. With few exceptions, center care appears to benefit reading and math

learning and harm behavior. We also ran these results using weights from the propensity score

matching described above (see Appendix Table 3). The results are very similar to the OLS results

indicating that the limited overlap in our sample across child care types is unlikely to be biasing our

results.6

Next we move onto the focus of our study, the effects of duration and intensity of center

care. Table 5 gives the results of the duration measure. The first column provides the results of

estimates using the full sample of children. The greatest benefit of center care for reading and math

skills accrues to children who start center-based programs between the ages of two and three.

Interestingly, those who start both before and after that time appear to gain less, though except for

those who entered after age five, attending a center is still associated with higher scores than parental

care. The results for behavior are different, though perhaps predictable. The negative behavioral

effects are greater the earlier a child enters. The difference in the relationship between center

duration and achievement in comparison to center and behavior suggests that simple selection bias

is unlikely to be biasing the results. If particularly strong families put their kids in center care at a

given age and we were not able to adjust for that with the many controls, we would expect to see the

6
The results are slightly different for high income reading results and the middle income behavior results but this is
solely do to differences between parental and other non-parental care. Parental care is the comparison group for the
OLS results, while the combination of parental and non-parental care is the comparison for the statistical matching
results.

12
highest scores across all three measures for children associated with the duration of center

attendance. This is not the case; the behavioral effects differ from the cognitive effects.

The final three columns of Table 5 give the results separately for the lowest quartile, middle

half and highest quartile of the income-to-needs ratio. Again the results are similar across the

income groups. For all three groups, the strongest reading and math effects occur for those who

enter centers between the ages of two and three, and the worst behavioral effects occur for those

who enter the earliest.

Duration or starting date of center care is one dimension for measuring the amount of

center care a child receives. However, children differ meaningfully in how much time they spend in

center care during a given year. Table 6 looks at the effect of intensity of care using dummy

variables to capture whether a child attended a center for between 15 and 30 hours per week for at

least nine months per year or for greater than 30 hours per week for at least nine months per year.

In this case, the coefficient on the center care dummy variable picks up the effect of any center

attendance and should be added to the coefficients on the measures of intensity to get the total

effect of intense attendance relative to parental care. The first column presents the results for the

full sample. These estimates suggest that some intensity, at least 15 hours per week, substantially

increases the contribution of center care to student learning but that there is little evident difference

between the scores of those who attend for 15 to 30 hours per week and those that attend for

greater than 30 hours. However, differences are evident in the behavioral index. Not only do the

negative behavioral effects appear for those with at least 15 hours of care per week but additional

care -- as measured by at least 30 hours of center care -- approximately doubles this negative effect.

The estimates for the full sample obscure interesting differences across income groups. In

particular for the lowest income group, reading and math achievement do increase with additional

care, not only going from less than 15 hours to more than 15 hours, but also going from 15 to 30

13
hours to more than 30 hours. In addition, the negative effect of center care on behavioral outcomes

does not increase in a qualitatively important or statistically significant way with additional hours of

care. On the other extreme, children from higher income families score lower in reading and math

when they attend centers for more than 30 hours per week than they do if they attend for 15 to 30

hours. In addition, unlike the lower-income children, these children from higher income families

demonstrate increasingly bad behavior the longer they are in center care each week.

Conclusion

This paper asks, how much is too much center-based child care? Evidence from ECLS-K

suggests that center care is associated with gains of about .1 standard deviations in math and reading,

though it is also associated with approximately the same size negative effect on a teacher-reported

behavioral measure that captures approaches to learning, self-control and a variety of interpersonal

skills. These differential effects suggest further exploration of the characteristics of center

attendance that effect cognitive and social outcomes. Is there a type of center care that maximizes

the positive cognitive effects while minimizing the negative social effects? In this paper we have

looked at the amount of center care, both duration and intensity, that a child experiences. Certainly

the quality of the center care is another aspect worth addressing (see for example, Loeb, Fuller,

Kagan, & Carrol, 2004). Unfortunately, while the ECLS-K data provides a national view of child

care use and kindergarten preparedness not available from other sources, it does not have

information on the quality of care across child care types.

We found consistent effects of the duration of center care attendance across income groups.

The data indicate that the greatest benefits of center care are for those who enter center-based

programs between the ages of two and three years old. Children who start earlier do not display

greater math and reading skills, and, in fact, in the sample the predicted averages are somewhat

14
lower than for those who start between two and three. In addition, the negative behavioral effects

associated with center attendance, relative to parental care, are much greater for those who start care

at less than two years of age and are particularly large for those who start at less than one year of

age. For both low and high income children, starting center care before the age of two is not

particularly beneficial for cognitive development and appears to be detrimental for behavior.

The results for the intensity of center care are more subtle, differing by income group. For

children from low-income families, additional hours per week are associated with gains in reading

and math and very little behavioral detriment. However, while high income children see reading and

math gains associated with moderate levels of care (15 to 30 hours per week), they see no cognitive

gains and substantially greater behavioral problems associated with additional hours of center care.

While these results do not provide cost-benefit analyses of universal preschool or center-

based child care programs that target children with particular background characteristics, they do

provide evidence of the benefit of center care for children from families at all income levels. They

also suggest that greater benefits can accrue from interventions that start earlier than the age of four.

Generally, children learn more when they start between two and three years of age. In addition,

while half day programs may be beneficial for children from higher income families, full day

programs better serve children from lower income families, allowing them to gain reading and math

skills without detriment to social behavior.

15
References

August, D. & Hakuta, K., Eds. (1997). Improving Schooling for Language Minority Children: A Research
Agenda. Washington, D.C.: National Academy Press.

Barnett, S. (1995). Long-term effects of early childhood programs on cognitive and school
outcomes. The Future of Children, 5, 25-50.

Bates, J. E., Marvinney, D., Kelly, T., Dodge, K. A., Bennett, D. S., & Pettit, G. S. (1994). Child care
history and kindergarten adjustment. Developmental Psychology, 30, 690-700.

Belsky, J. (2001). Developmental risks (still) associated with early child care. Journal of Child Psychology
& Psychiatry & Allied Disciplines, 42, 845-859.

Burchinal, M. R., Campbell, F. A., Bryant, D. B., Wasik, B. H., Ramey, C. T. (1997). Early
intervention and mediating processes in cognitive performance of children of low-income
African-American families. Child Development, 68 (5), 935-954.

Bridges, M., Fuller, B. C., Rumberger, R., & Tran, L. (2004). Preschool for Californias children:
Promising benefits, unequal access. PACE Policy Brief, 04-3, Berkeley, CA: Policy Analysis for
California Education (PACE).

Brooks-Gunn, J. (2003). Do you believe in magic? What we can expect from early childhood
intervention programs. SRCD Policy Brief, 17.

Campbell, F.A. & Ramey, C.T. (1994). Effects of early intervention on intellectual and academic
achievement: A follow-up study of children from low-income families. Child Development 65,
684-698.

Campbell, F. A., & Ramey, C. T. (1995). Cognitive and school outcomes for high-risk African-
American student at middle adolescence: Positive effects of early intervention. American
Educational Research Journal, 32, 743-772.

Currie, J. (2001). Early childhood intervention programs: What do we know? Journal of Economic
Perspectives, 15, 213-238.

Currie, J. & Thomas, D. (2000). School quality and the longer-term effects of Head Start. Journal of
Human Resources, 35, 755-774.

Currie, J. & Thomas, D. (2001). Early test scores, socioeconomic status, school quality and future
outcomes. Research in Labor Economics, 20, 103-132.

Fryer, R. & Levitt, S (2004). Understanding the Black-White Test Score Gap in the First Two Years
of School, The Review of Economics and Statistics, May.

16
Fuller, B. C. & Huang, D. (2003). Targeting investments for universal preschool: Which families to
serve first? Who will respond? PACE Working Paper, 03-1, Berkeley, California: Policy
Analysis for California Education (PACE).

Fuller, B. C., Livas, A., & Bridges, M. (2005). How to expand and improve preschool in California:
Ideals, evidence, and policy options. PACE Working Paper 05-1. Berkeley, California: Policy
Analysis for California Education (PACE).

Garces, E., Thomas, D., & Currie, J. (2002). Longer term effects of Head Start. The American
Economic Review, 92, 999-1012.

Heckman, James (2000). Policies to Foster Human Capital, Research in Economics, 54(1), 3-56

Heckman, J., Ichimura, H. & Todd, P. (1998), Matching as an econometric evaluation estimator,
Review of Economic Studies 65, 261294.

Hofferth, S., Shauman, K., Henke, R., & West, J. (1995). Characteristics of Childrens Early Care and
Education Programs: Data from the 1995 National Household Education Survey. Washington, D.C.:
National Center for Education Statistics.

Jencks, C. & Phillips, M. (1998). The Black-White Test Score Gap. Washington, D.C.: Brookings
Institution.

Kisker, E., Hofferth, S., Phillips, D., & Farquhar, E. (1991). Profile of Child Care Settings: Early
Education and Care in 1990. Washington, D.C.: United States Department of Education.

Liang, X., Fuller, B., & Singer, J. (2000). Ethnic differences in child care selection: The influence of
family structure, parenting practices, and home language. Early Childhood Research Quarterly,
15, 357-384.

Loeb, S., Fuller, B., Kagan, S. L., & Carrol, B. (2004). Child care in poor communities: Early learning
effects of type, quality, and stability. Child Development, 75, 47-65.

Magnuson, K. A., Ruhm, C. J., & Waldfogel, J. (2004). Does pre-kindergarten improve school
preparation and performance? National Bureau of Economic Research Working Paper 10452.
Cambridge, MA: NBER.

National Center for Educational Statistics (1998). Users Manual for the ECLS-K Base Year Public-Use
Data. Washington, D. C.: US Department of Education, Office of Educational Research and
Improvement.

NICHD Early Child Care Research Network. (2003). Does the amount of time spent in child care
predict socioemotional adjustment during the transition to kindergarten? Child Development,
74, 976-1005.

OBrien-Strain, M., Moye, L., & Sonenstein, F. L. (2003). Arranging and Paying for Child Care. San
Francisco, California: Public Policy Institute of California.

17
Reardon, S. (2003). Sources of educational inequality: The growth of racial/ethnic and
socioeconomic test score gaps in kindergarten and first grade. State College, PA: Penn State
University (Manuscript).

Rumberger, R. W., & Arellano, B. (2003). Understanding and Addressing the Latino Achievement Gap in
California. Draft Report submitted to the Latino Policy Institute, California Policy Research
Center, University of California.

Shonkoff, J. P. & Phillips, D. A., Eds. (2000). From Neurons to Neighborhoods: The Science of Early Child
Development. Washington, D.C.: National Academy Press.

Smith, T., Kleiner, A., Parsad, B., Farris, E., & Green, B. (2003). Prekindergarten in the U.S. Public
Schools. Washington, DC: U. S. Department of Education, National Center for Education
Statistics.

Snow, C. E., Burns, M. S., & Griffin, P. (Eds.). (1998). Preventing Reading Difficulties in Young Children.
Washington, DC : National Academy Press.

U. S. Department of Education, National Center for Education Statistics. (2003). The Condition of
Education. Washington, D.C.: U.S. Government Printing Office.

Waldfogel, J. (2002). Child care, womens employment, and child outcomes. Journal of Population
Economics, 15, 527-548.

Wertheimer, R., Croan, T., Moore, K. A., & Hair, E. C. (2003). Attending kindergarten already
behind: A statistical portrait of vulnerable young children. Child Trends Research Brief, 2003-20.
Washington, D.C.: Child Trends.

18
Table 1: Descriptives of Selected Variables by Child Care Type

All Parental Center Head Start Other

Sample Size 14162 2363 9015 1093 1691
South 0.37 0.38 0.36 0.44 0.34
Birth Weight 118(21) 118(22) 119(21) 114(23) 117(21)
Race/Ethnicity Black 0.16 0.12 0.14 0.39 0.13
Hispanic 0.13 0.17 0.11 0.19 0.18
English Only at Home 0.84 0.80 0.86 0.80 0.81
Single parent family 0.22 0.16 0.20 0.43 0.28
Mother's Education <HS 0.10 0.19 0.06 0.24 0.12
HS 0.31 0.36 0.27 0.44 0.37
Vocational 0.06 0.06 0.05 0.06 0.06
BA 0.16 0.10 0.20 0.02 0.10
Some Grad 0.02 0.01 0.02 0.01 0.01
MA 0.05 0.02 0.06 0.00 0.03
PhD 0.01 0.00 0.02 0.00 0.01
Mother Employed Full Time 0.46 0.24 0.48 0.44 0.65
Part Time 0.22 0.21 0.23 0.18 0.17
Father Employed Full Time 0.70 0.73 0.74 0.44 0.67
Part Time 0.03 0.03 0.03 0.03 0.03
WIC Participation 0.45 0.49 0.37 0.89 0.52
Income to Needs < .5 0.07 0.08 0.04 0.22 0.06
.5-1.0 0.11 0.14 0.08 0.29 0.13
Home: # of children's books 78(59) 72(58) 85(60) 49(48) 69(56)
computer 0.59 0.53 0.65 0.30 0.52
tv hours 1.84(1.20) 1.96(1.25) 1.73(1.12) 2.22(1.42) 1.98(1.28)
visited zoo 0.40 0.39 0.41 0.37 0.39
visited library 0.55 0.53 0.57 0.46 0.49
dance lessons 0.18 0.12 0.22 0.10 0.14
sports lessons 0.49 0.39 0.56 0.28 0.41
spanked 0.20 0.22 0.18 0.27 0.23
build things 2.36(0.92) 2.43(0.94) 2.34(0.90) 2.43(1.00) 2.31(0.92)
art 2.67(0.87) 2.69(0.89) 2.67(0.85) 2.63(0.95) 2.67(0.86)
games 2.81(0.82) 2.83(0.85) 2.81(0.80) 2.81(0.89) 2.81(0.81)
chores 3.31(0.86) 3.33(0.90) 3.30(0.84) 3.40(0.90) 3.31(0.85)
sports 2.69(0.91) 2.76(0.95) 2.66(0.88) 2.78(1.00) 2.68(0.93)
nature 2.22(0.88) 2.23(0.89) 2.25(0.86) 2.09(0.92) 2.18(0.88)
Parent at School:
PTA meeting 0.34 0.36 0.35 0.30 0.27
Parent-Tchr conference 0.86 0.83 0.88 0.79 0.84
Volunteered 0.50 0.50 0.55 0.31 0.42

19
 Table 2: OLS Estimates of the Effects of Child Care Settings on Cognitive and Behavioral
Outcomes

(1) (2) (3) (4)

Reading
Center Care 3.853*** 1.275*** 1.197*** 1.102***
(0.217) (0.217) (0.218) (0.219)
Head Start Care -4.054*** -0.625* -0.629* -0.494
(0.332) (0.335) (0.339) (0.343)
Other Non-Parent Care -0.489 -0.424 -0.427 -0.442
(0.299) (0.293) (0.295) (0.296)

Child & Family Characteristics Yes Yes Yes

Zip Code Demographics Yes Yes
State Fixed Effects Yes

Observations 14162 11834 11710 11577

R-squared 0.07 0.34 0.35 0.36

Math
Center Care 3.495*** 1.320*** 1.257*** 1.207***
(0.211) (0.207) (0.208) (0.210)
Head Start Care -3.884*** 0.097 0.127 0.253
(0.322) (0.320) (0.325) (0.329)
Other Non-Parent Care -0.196 0.136 0.116 0.165
(0.291) (0.280) (0.282) (0.284)

Child & Family Characteristics Yes Yes Yes

Zip Code Demographics Yes Yes
State Fixed Effects Yes

Observations 14162 11834 11710 11577

R-squared 0.06 0.36 0.36 0.37

Behavior
Center Care -0.096*** -0.091*** -0.092*** -0.095***
(0.023) (0.025) (0.025) (0.026)
Head Start Care -0.414*** -0.125*** -0.128*** -0.120***
(0.035) (0.039) (0.040) (0.040)
Other Non-Parent Care -0.011 0.098*** 0.100*** 0.101***
(0.031) (0.034) (0.034) (0.035)

Child & Family Characteristics Yes Yes Yes

Zip Code Demographics Yes Yes
State Fixed Effects Yes

Observations 14162 11834 11710 11577

R-squared 0.01 0.13 0.13 0.14
Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%

20
Table 3: Specification Checks of the Effects of Child Care Type on Cognitive and Social
Outcomes

Instrumental Variables Results

Reading Math Behavior

Center Care 10.407** 6.613 -0.816

(5.082) (4.610) (0.568)
Head Start Care -2.095 -2.789 -1.481**
(6.220) (5.642) (0.695)
Other Non-Parent Care 6.150 3.784 -0.544
(4.137) (3.752) (0.462)
R-squared (n=9490) 0.22 0.31 0.02
Statistical Matching Results
Head Start Children Excluded
Reading Math Behavior
Center Care (n=10763) 1.289*** 1.255*** -0.134***
(0.157) (0.148) (0.018)
R-squared 0.34 0.34 0.13
Center Care Children Excluded
Reading Math Behavior
Head Start (n=3992) -.453* 0.257 -0.155***
(0.234) (0.234) (0.029)
R-squared 0.34 0.34 0.13
Models include all child/family controls and zip controls as well as dummy variables for Head Start participation and
other non-parental care. Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%

21
Table 4: Estimates of the Effects of Child Care Settings on Cognitive and Behavioral
Outcomes by Income Group (lowest quartile, middle half, upper quartile)

Low Middle High

Reading
Center Care 0.620 1.272*** 0.806
(0.432) (0.304) (0.586)
Head Start Care -0.821 0.120
(0.504) (0.589)
Other Non-Parent Care -0.316 -0.280 -0.882
(0.585) (0.407) (0.765)

Observations 2670 5891 3016

R-squared 0.28 0.30 0.33
Math
Center Care 1.188*** 1.182*** 1.011*
(0.442) (0.289) (0.544)
Head Start Care 0.514 0.097
(0.515) (0.560)
Other Non-Parent Care 0.231 0.410 -0.495
(0.598) (0.387) (0.710)

Observations 2670 5891 3016

R-squared 0.29 0.30 0.32
Behavior
Center Care -0.158*** -0.014 -0.176**
(0.054) (0.035) (0.068)
Head Start Care -0.103 -0.141**
(0.062) (0.067)
Other Non-Parent Care -0.027 0.166*** 0.080
(0.072) (0.047) (0.089)

Observations 2670 5891 3016

R-squared 0.18 0.14 0.15
Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%
Models similar to fourth model in Table 2 plus including a continuous measure of income-to-needs.

22
Table 5: OLS Estimates of the Effects of Age at Center Entry on Cognitive and Social
Outcomes By Income Group
All Low Middle High
(11710) (2670) (5891) (3061)
Reading
Started Center Age 0-1 1.079*** 0.029 1.278** 0.235
(0.366) (0.887) (0.529) (0.751)
Started Center Age 1-2 1.207*** 0.228 1.265** 0.400
(0.404) (0.927) (0.584) (0.817)
Started Center Age 2-3 1.912*** 1.608** 1.918*** 1.301*
(0.323) (0.781) (0.484) (0.669)
Started Center Age 3-4 1.412*** -0.119 1.720*** 0.987
(0.263) (0.608) (0.362) (0.619)
Started Center Age 4-5 0.994*** 1.188** 0.923** 0.317
(0.268) (0.564) (0.358) (0.684)
Started Center Age >5 0.266 -0.492 0.836 -0.653
(0.550) (1.136) (0.729) (1.309)
Math
Started Center Age 0-1 1.503*** 0.212 1.529*** 1.240*
(0.350) (0.906) (0.503) (0.696)
Started Center Age 1-2 0.940** 0.492 1.035* 0.402
(0.386) (0.947) (0.555) (0.758)
Started Center Age 2-3 1.777*** 2.388*** 1.565*** 1.251**
(0.310) (0.798) (0.461) (0.621)
Started Center Age 3-4 1.371*** 0.71 1.360*** 1.400**
(0.251) (0.622) (0.345) (0.575)
Started Center Age 4-5 1.059*** 1.568*** 0.971*** 0.197
(0.257) (0.576) (0.341) (0.635)
Started Center Age >5 0.870* 0.973 0.884 0.416
(0.526) (1.161) (0.693) (1.214)
Behavior
Started Center Age 0-1 -0.289*** -0.345*** -0.161*** -0.387***
(0.043) (0.110) (0.060) (0.087)
Started Center Age 1-2 -0.205*** -0.209* -0.149** -0.292***
(0.047) (0.115) (0.067) (0.095)
Started Center Age 2-3 -0.156*** -0.252*** -0.075 -0.229***
(0.038) (0.097) (0.055) (0.078)
Started Center Age 3-4 -0.103*** -0.190** -0.02 -0.164**
(0.031) (0.075) (0.041) (0.072)
Started Center Age 4-5 -0.032 -0.09 0.029 -0.085
(0.031) (0.070) (0.041) (0.080)
Started Center Age >5 -0.081 -0.279** 0.03 -0.054
(0.064) (0.141) (0.083) (0.152)

Models include all child/family controls and zip controls as well as dummy variables for Head Start participation,
unknown center start date, and other non-parental care. Standard errors in parentheses. * significant at 10%; **
significant at 5%; *** significant at 1%. Models 2, 4, and 6 include state fixed effects and the income to needs
ratio.

23
Table 6: OLS Estimates of the Effects of Intense Center Care on Cognitive and
Behavioral Outcomes

All Low Middle High

Reading
Center Care 0.895*** 0.311 0.934*** 0.835
(0.244) (0.535) (0.325) (0.594)
15-30 hours/week, 9 months 0.812*** 0.338 0.654* 1.072**
(0.264) (0.692) (0.390) (0.428)
At least 30 hours/week, 9 months 0.898*** 1.562*** 1.373*** -0.192
(0.244) (0.590) (0.352) (0.430)

Observations 11293 2439 5784 3070

R-squared 0.35 0.27 0.29 0.31
Math
Center Care 0.938*** 0.970* 0.859*** 0.894
(0.232) (0.534) (0.309) (0.549)
15-30 hours/week, 9 months 0.745*** -0.201 0.542 1.265***
(0.251) (0.691) (0.370) (0.395)
At least 30 hours/week, 9 months 0.807*** 1.152* 1.055*** 0.319
(0.232) (0.590) (0.335) (0.398)

Observations 11293 2439 5784 3070

R-squared 0.36 0.30 0.29 0.30
Behavior
Center Care -0.009 -0.128* 0.050 -0.118*
(0.028) (0.066) (0.037) (0.069)
15-30 hours/week, 9 months -0.109*** -0.022 -0.112** -0.128***
(0.031) (0.085) (0.044) (0.049)
At least 30 hours/week, 9 months -0.252*** -0.045 -0.265*** -0.312***
(0.028) (0.073) (0.040) (0.050)

Observations 11293 2439 5784 3070

R-squared 0.14 0.17 0.14 0.15

Standard errors in parentheses. * significant at 10%; ** significant at 5%; *** significant at 1%

Models include child/family and zip code level controls, as well as Head Start dummy and non-relative care dummy
(excluded group is parental care). Poor, middle, and high income models include a continuous income to needs
variable.

24
Appendix Table 1: Control Variables from ECLS-K

Variable Description
Child age Childs Age at Assessment
Child gender Dummy Variable
Birth weight In ounces
Child weight Average of two interviewer assessed measurements in lbs.
Child height Average of two interviewer assessed measurements in inches.
Race/ethnicity 4 dummy variables for black, Hispanic, Native American, and Asian
# of children Dummy variables ranging from 1 to 11 for the number of children in household
Family structure 3 dummy variables: Single parent (one biological parent), blended family (one
biological and one non-biological parent), adopted or foster parents
Urbanicity Locality is city or town (2 dummy variables).
Region North, South, Midwest (3 dummy variables).
Mothers Dummy variable for whether the mother was ever employed between childs birth
employment and entry into kindergarten?
Parents education Six dummy variables for fathers and mothers education: Less than high school
degree through advanced post-graduate degree
English Dummy variable for whether English is the only language spoken in home .
Parents current Full-time (35 or more hours per week), part-time work (fewer than 35 hours per
employment week), or no work (2 dummy variables for each parent).
WIC Dummy for whether mother or child ever participated in Women, Infants and
Children nutritional supplement program.
income-to-needs Household income vs. federal poverty level ratio (9 dummy variables).
Expectations Parental Expectations for Childs education (4 dummy variables)
Importance of skills Importance of skills: counting, sharing, communication, drawing, knowledge of letters
on scale of 1-5.
Choice of location Parents chose home location for current school (dummy variable).
Home learning Frequency of 7 activities: building things, teaching about nature, playing sports, doing
activities art, doing chores, singing songs, playing games. 7 variables scaled from 1-4.
# of childrens books Ordinal variable. Ranges from 0 to 200.
in home
# of music tapes, Ordinal variable. Ranges from 0 to 100.
CDs, or records in
home
Reading Frequency of child looking at picture books or reading outside of school (2 variables).
School activities Attendance since beginning of school year at PTA meetings, open houses, parent
groups, parent advisory meetings; volunteered at school, participated in school
fundraiser (6 dummy variables).
Parenting stress and Two continuous variables (averages of 8 and 12 items). Higher score composites
depression indicate more stress, depression.
Spanking Dummy, 1 if parent spanked child in last week.
Eating Habits. Days per week family usually eats meals together, at regular time (4 dummy variables)
Computer. Dummy variable if family has a computer.
T.V Number of hours child watches TV on weekdays.
Visiting In the past month, have you visited a zoo, library, museum, concert (4 variables)
Other non-school Child has ever taken lessons or participated in performing arts or organized clubs
activities outside of school (7 dummy variables).
Neighborhood Mean of 6 items asking about neighborhood problems.

25
Appendix Table 2: Zip Code Level Data from the 2000 Long Form of the Decennial
Census

Variable Variable
Total Population % Black
% Urban % Native
% 5 and under % Pacific
% of children under 5 in poverty % Other
% of population over 16 in the % Mixed
labor force who are unemployed
% of children 0-6 living with single mothers % Asian
% of children 0-6 living with single fathers % Hispanic
% of women with children 0-6, in % of households in which Spanish is
the labor force, unemployed sometimes or always spoken
% of women over 25 with less % of households in which a language
than a hs diploma other than English is sometimes or always
% of women over 25 with a hs % of households that are linguistically isoloated
diploma or equivilancy (no one over 14 speaks English
% of women over 25 with a BA or more % of family households with 6 or more members
estimated pct of 3 and 4 year olds in % of population who is non-citizen
preschool/nursery school
estimated pct of 3 and 4 year olds in
PUB preschool/nursery school

Appendix Table 3: Statistical Matching Estimates of the Effects of Child Care Settings on
Cognitive and Behavioral Outcomes

Statistical Matching Results - Head Start Children Excluded

Low Income
Reading Math Behavior
Center Care (n=2121) 0.915** 1.450*** -.104**
(0.358) (0.358) (0.043)
R-squared 0.31 0.34 0.20
Middle Income
Reading Math Behavior
Center Care (n=5642) 1.283*** 1.070*** -0.068***
(0.219) (0.208) (0.018)
R-squared 0.30 0.29 0.14
High Income
Reading Math Behavior
Center Care (n=3000) 1.159*** 1.194*** -0.222***
(0.306) (0.279) (0.034)
R-squared 0.34 0.32 0.17

26