School-Based Early Childhood Education and Age-28 Well-Being: Effects by Timing,
Dosage, and Subgroups
Arthur J. Reynolds,1* Judy A. Temple,2 Suh-Ruu Ou,1 Irma A. Arteaga,3 Barry A. B. White1
1

Institute of Child Development, University of Minnesota, 51 East River Road, Minneapolis, MN 55455, USA. 2Humphrey
School of Public Affairs, Department of Applied Economics, University of Minnesota, Minneapolis, MN 55455, USA. 3Harry S
Truman School of Public Affairs, University of Missouri, Columbia, MO 65211, USA.
*To whom correspondence should be addressed. E-mail: ajr@umn.edu
Advances in understanding the effects of early education
have benefited public policy and developmental science.
Although preschool has demonstrated positive effects on
life-course outcomes, limitations in knowledge on
program scale, subgroup differences, and dosage levels
have hindered progress. We report the effects of the
Child-Parent Center Education Program on indicators of
well-being up to 25 years later for more than 1400
participants. This established, publicly funded
intervention begins in preschool and provides up to 6
years of service in inner-city Chicago schools. Relative to
the comparison group receiving the usual services,
program participation was independently linked to higher
educational attainment, income, socioeconomic status
(SES), and health insurance coverage, as well as lower
rates of justice-system involvement and substance abuse.
Evidence of enduring effects was strongest for preschool,
especially for males and children of high school dropouts.
The positive influence of 4 years or more of service was
limited primarily to education and SES. Dosage within
program components was mostly unrelated to outcomes.
Findings demonstrate support for the enduring effects of
sustained school-based early education to the end of the
third decade of life.
The effects of educational enrichment in the early years of
life are a central focus of developmental science and are
increasingly used to prioritize social programs and policies.
In the past two decades, evidence has grown that preschool or
“prekindergarten” programs enhance well-being in many
domains, and can promote economic benefits to society (1–3).
Although the most enduring effects on school success and
crime prevention are found among economically
disadvantaged children (4), preschool programs can promote
well-being across the entire socioeconomic spectrum (5, 6).
The magnitude, breadth, and duration of impacts for
preschool have been found to be more consistent and stronger
than many prevention strategies (7). This pattern is likely due

to the greater dosage, intensity, and scope of services.
Preschools typically provide 500 hours per year. These
enrichment experiences appear to initiate a pattern of
cumulative advantages (7–9) that can translate to enduring
life-course effects (10). Recent evidence on Head Start (11),
however, suggests that enduring effects are not inevitable,
and may depend on later social contexts (12).
Although evidence is strong that programs of relatively
high quality can promote well-being, four major weaknesses
reduce the strength and generalizability of evidence (13). The
most widely documented limitation is that evidence on longterm effects is primarily from small-sample efficacy trials
rather than effectiveness trials or studies of large-scale
sustained programs (2, 4). Studies of sustained and routinely
implemented programs are essential to translational research
yet long-term evidence is meager (1, 7) and none have
continued past age 25, which is most predictive of later
development (14).
Three other less recognized limitations also have hindered
progress. One is inadequate attention to program dosage, a
prominent and modifiable characteristic. Although some
studies show that the length of participation is positively
associated with short-term outcomes (7, 15), longer-term
effects have been rarely investigated as have the added or
synergistic benefits of later intervention. The second
limitation is that variations in effects by child, family, and
social context are under-investigated. Their identification
provides valuable information for tailoring or strengthening
services. Differences by gender vary by study and outcome,
and long-term effects on high-risk samples warrant greater
investigation. Finally, attrition is rarely taken into account in
estimating effects. Studies frequently lose up to 50% of their
original samples in follow-up (16, 17). The power and
precision of subgroup effects can be especially compromised.
Bias reduction methods to account for attrition and other
selection processes have become more integrated in
estimation (18).

/ www.sciencexpress.org / 9 June 2011 / Page 1 / 10.1126/science.1203618

To assess the effects of a large-scale sustained early
education program in public schools, the Chicago
Longitudinal Study (CLS) (19) has prospectively documented
the life-course development of 1,539 families (93% African
American), the majority of whom participated in the ChildParent Center (CPC) Education Program. CPC is the second
oldest (after Head Start) federally funded preschool, and has
been implemented in the Chicago Public Schools since 1967
(20). In addition to providing comprehensive services to
economically disadvantaged families, the program has
preschool and school-age components that enable
assessments of the timing and length of participation.
In this study, we investigate links between CPC
participation and well-being by age 28. While previous
studies of publicly-funded programs (21) including CPC (22)
have showed positive evidence, due to the age of assessment
in early adulthood, a full range of economic, health, and
family outcomes has not been assessed. Moreover, unlike
previously, we examine differential effects by timing and
length of intervention as well as child and family attributes.
We also take into account through propensity score analysis
the potential biasing effects of attrition and selection bias.
Our major questions are: (1) Is CPC participation beginning
in preschool and continuing into school-age associated with
multiple domains of well-being? (2) Do estimated effects
vary by child and family characteristics as well as dosage
levels? (3) Are effects consistent across models for reducing
bias in estimates?
Born in 1979–1980, the CLS sample is the entire cohort of
989 children who completed preschool and kindergarten
(half- or full-day) in all 20 CPCs and 550 low-income
children who did not attend the program in preschool but
participated in a full-day kindergarten intervention in five
randomly selected schools. 15% attended Head Start with
most others in home care. That the comparison group
participated in an enrichment program minimizes bias in
group selection. First- to third-grade program services are
offered to all students. Table 1 shows the patterns of
participation and for inclusion in the adult follow-up study
(13).
In this alternative-intervention, quasi-experimental design,
groups matched on age, eligibility for intervention, and
family poverty. In support of the interpretability of estimates,
group comparisons at the beginning of the study and at
follow-up show similarity on preprogram characteristics
(Table 1 and table S2). Sample characteristics have been
consistent over time.
Located in or close to elementary schools, the CPC
program provides educational and family-support services
between the ages of 3 and 9. The key goal stated by founder
Lorraine Sullivan is that the centers “are designed to reach the
child and parent early, develop language skills and self-

confidence, and to demonstrate that these children, if given a
chance, can meet successfully all the demands of today’s
technological, urban society” (23). The program emphasizes
basic skills in language arts and math through relatively
structured but diverse learning experiences that include
whole-class instruction, small-group and individualized
activities, and frequent field trips. All teachers have
bachelor’s degrees and are certified in early childhood
education. Classes are small (17 in preschool; 25 in
kindergarten to third grade) and are staffed by teacher aides.
In addition to the head teacher in each site, the parent
resource teacher and outreach representative direct multifaceted and intensive services in the parent resource room.
The scope of services helped ensure high participation. Heavy
outreach by staff also led to participation by families most in
need (13). Preschool and kindergarten were funded by Title I
of the Elementary and Secondary Education Act of 1965
(P.L. 89-10; 79 Stat. 27) while school-age services were
funded by the State of Illinois and Chicago Boards of
Education. In 2011 dollars, the average costs per child were
as follows: preschool ($9,233), school-age over and above
regular instruction ($4,113), and preschool plus school-age
relative to lesser program services ($5,600).
As shown in Table 1, 90.1% of the original sample had
follow-up data on educational attainment or socioeconomic
status (mean age 28.3 years). Recovery rates for the groups
were nearly identical. They ranged, in the overall sample,
from 80% to 96% for other outcomes (table S1). Well-being
was assessed in five domains: educational attainment,
socioeconomic status (SES), health status and behavior, crime
and justice system involvement, and family outcomes (tables
S3–S5). High school completion, for example, was a high
school diploma or equivalent. One indicator of SES was a
composite index of education and income. Measures were a
combination of administrative and survey data from many
sources (e.g., education, crime, and income records) and are
theoretically related to the ultimate goal of economic
independence.
We estimated effects using probit, linear, and negative
binomial regression analysis adjusted for 15 preprogram
attributes and weighted by attrition propensities through
Inverse Probability Weighting (IPW) (24). IPW has been
shown to yield the most efficient estimates (25). The weight
was 1/p1, where p1 is the predicted probability of being in the
recovery sample (Ri = 1; otherwise 0) for each outcome (i) as
a function of 26 predictors known to influence attrition (table
S6 and fig. S6). Standard errors were corrected for site
clustering. Robustness of estimates was fully assessed (table
S7).
A summary of findings for select outcomes in four
domains is shown in Table 2, including preschool, schoolage, and extended intervention. For brevity, our measure of

/ www.sciencexpress.org / 9 June 2011 / Page 2 / 10.1126/science.1203618

extended intervention here is for 4 to 6 years versus fewer.
Findings for an alternative measure of extended intervention
(extended-2) and for other outcomes including family status
are shown in table S9. Unadjusted group differences are also
reported (table S8). We emphasize domains in which two or
more indicators shows significance at the 0.05 level.
Relative to the comparison group, the preschool group had
significantly higher levels of educational attainment for 3 of 4
select outcomes in Table 2 and 4 of 6 attainment outcomes
overall (table S9). This included highest grade completed
(12.15 vs. 11.88; p = 0.03), attendance in a 4-year college
(14.7% vs. 11.2%; p = 0.04), high school completion (81.5%
vs. 75.1%; p = 0.007), and on-time high school graduation
(44.3% vs. 36.6%; p = 0.018). These educational advantages
translated to higher economic status, including occupational
prestige (2.8 vs. 2.5; p = 0.03), SES composite score
(education and income) of 4 or higher (34.4% vs. 28.6%; p =
0.03; scale of 0–8), and average annual income in 2007
dollars ($11,582 vs. $10,796; p = 0.001). Moreover, a higher
percentage had an occupational prestige level of 4 or higher
(28.2% vs. 21.4%; p = 0.01), synonymous with postsecondary
training. No differences were detected for degree completion,
employment, or a combined measure (table S9).
School-age participation was associated with a higher rate
of on-time high school graduation (44.4% vs. 35.3%; p =
0.011) while extended intervention was linked to highest
grade completed (12.21 vs. 11.95, p = 0.02), high school
completion (82.7% vs. 77.2%; p = 0.01), on-time graduation
(48.6% vs. 31.3%; p = 0.001) as well as the SES composite
score of 4 or higher (35.9% vs. 30.3%; p = 0.036), and the
occupational prestige index (3.1 vs. 2.7, p = 0.017; table S9).
Based on the alternative extended-intervention contrast, only
on-time high school graduation differed between groups
(table S9). This conservative test minimizes any possible
synergistic effect of intervention, however, because
kindergarten achievement was included in the model.
The preschool group had a higher rate of health insurance
coverage (75.9% vs. 63.9%; p < 0.01), including private
insurance (49.1% vs. 39.5%; p = 0.01). They also had
significantly lower rates of substance abuse (13.7% vs.
18.9%; p = 0.01) and drug and alcohol abuse (16.5% vs.
23.0%; p = 0.004; table S9). The extended program group had
a higher rate of private health insurance coverage (51.8% vs.
42.2%; p = 0.001) but this difference was not found for the
alternative contrast.
The preschool group also had lower rates of crime and
justice system involvement for 2 of 3 select outcomes
including any arrest (47.9% vs. 54.3%; p = 0.03) and felony
arrest (19.3% vs. 24.6%; p = 0.02) as well as any
incarceration or jail history (15.2% vs. 21.1%; p = 0.04)
(table S9). Because the latter two outcomes were measured
from official records, they are more severe and have higher

costs. No differences were detected for the number of arrests,
arrests for violence, or convictions. School-age and extended
intervention were unrelated to justice involvement. For public
aid and family outcomes, no meaningful differences were
found (table S9).
Although subgroup differences were detected, they were
limited to specific outcomes and intervention components
(table S11). The most consistent evidence was for gender and
parent education. Figure 1 shows the primary findings. Male
preschool participants showed substantially greater wellbeing than the male comparison group for high school
completion (77.5% vs. 63.5%; p = 0.002) and substance
abuse (33.7% vs. 42.9%; p = 0.002) whereas female program
groups had similar rates. In contrast, females showed
comparatively greater effects of school-age intervention than
males. Because this latter finding was not found for another
outcome, cautious interpretation is warranted.
In addition, preschool participants whose parents were
high school dropouts showed significantly larger effects than
participants of graduates for high school completion, felony
arrest, and substance abuse. For example, preschool
participants of high school dropouts had a rate of felony arrest
(13.9%) that was nearly half the rate for the comparison
group of school dropouts (25.2%). A similar risk indicator—4
or more family risks—also moderated preschool impacts on
felony arrest and substance abuse. While these findings
support the compensatory value of intervention, we found no
differences by race/ethnicity, early home environment, and
other factors. A similar pattern was found for extended
intervention.
For program dosage within components, length of
preschool was unrelated to nearly all measures of well-being
(table S12). School-age participation for 2 or 3 years was
linked to higher rates of on-time high school graduation
(41.5% vs. 28.5%; p = 0.025). Relative to 4 years, extended
intervention for 5 or 6 years was linked to a lower rate of
arrest for violence (13.4% vs. 20.8%; p = 0.002), and this was
also found for the alternative contrast (14.1% vs. 19.3%; p =
0.019).
To assess the robustness of estimates, we tested five
additional model specifications for each intervention contrast,
ranging from no adjustment on preprogram attributes to
inclusion of covariates, and IPW-attrition and IPW-selection
adjusted models. For the latter, the inverse of the estimated
propensity score for program participation (17 predictors;
table S6) was multiplied by IPW-attrition and this product
(double correction) was the model weight. Other propensity
methods such as matching yielded similar findings (table S7,
fig. S4).
We found evidence of consistency across model
specifications. The predominant pattern is shown in Figure 2
for moderate or higher SES and felony arrest. This

/ www.sciencexpress.org / 9 June 2011 / Page 3 / 10.1126/science.1203618

generalized to subgroup estimates reported above. Among the
four specifications shown, the unadjusted group differences
for SES (7.1 points) and felony arrest (6.4 points) are slightly
higher than the adjusted rates but the type of adjustment,
including the double correction, did not affect estimates in
any meaningful way. The reduction over the comparison
group in felony arrest was 27% whereas for SES it was an
increase of 20%. These findings strengthen confidence in the
beneficial effects of intervention.
The interpretation of findings as the impact of intervention
is further supported by corroboration that five sets of
mediators can account for effects. In this model, participation
impacts well-being through the accumulation of cognitive
skills, social adjustment, motivation, and family and school
support behaviors from school entry up to early adulthood
(26). We found that these mediators explained 60% or more
of the observed effects of preschool and nearly 40% or more
for extended intervention (tables S13 and S14). The
mediators completely accounted for effects on SES,
education, and felony arrest. The process of influence is
initiated by the impact on cognitive skills at age 5 and parent
involvement and continues through socio-emotional
adjustment, school quality, and reductions in problem
behavior. These paths have been found for outcomes at
younger ages (20, 27).
Overall, we found that the most consistent and enduring
effects were for preschool participation, which started at ages
3 or 4. Its impact was broad, including education, SES, health
behavior, and crime outcomes. Since the program affected
multiple indicators within these domains, impacts are unlikely
to be artifacts of measurement. Findings for later intervention
were limited primarily to education while those for extended
intervention were exclusive to education and economic wellbeing. Because of the high avoidable costs of school dropout
and related problems (28, 29), our findings strengthen
evidence that sustained, publicly-funded early education can
be a cost-effective strategy for promoting well-being.
The enduring effects of the program were observed within
a social context characterized by high levels of risk that
substantially counteract the positive influences of early
experience (30, 31). In addition to residing in neighborhoods
of persistent poverty where the majority of students fail to
complete high school, over half of participants changed
schools frequently and only 25% of participants attended
schools of relatively high quality. That the program,
especially in preschool, showed such broad and practically
significant effects on well-being despite these environmental
challenges is encouraging for prevention programming.
That male participants and those from higher risk families
showed the largest preschool effects is consistent with prior
studies (3, 4, 7), and given our estimation, cannot be due to
differential attrition. The advantage for males was found even

with no initial group differences (table S2). These findings
suggest that early interventions can reduce health disparities,
especially if they impact educational attainment, a key path to
later health and SES (10, 32). One implication is that national
goals of increasing quality and years of healthy life can be
achieved in part through access to quality educational
programs.
The study also shows the potential limits of the long-term
effects of dosage within program components. Although
extended intervention linked to well-being, the number of
years of preschool and extended services was unrelated to
most outcomes. Consistent with other studies (2, 33), greater
dosage of school-age intervention was linked to high school
graduation. These results suggest that among high quality
programs there may be a threshold beyond which effects
diminish. In previous studies (34, 35), however, preschool
and extended-intervention dosage was associated with
improved child and adolescent well-being including school
readiness, remedial education, child maltreatment, and
delinquency.
In conclusion, early education programs can impact lifecourse outcomes necessary for economic success and good
health. The findings of this study indicate that while there are
limits to the effects of the CPC program for particular
outcomes and groups, impacts which endured provide a
strong foundation for the investment in and promotion of
early childhood learning.
References and Notes
1. G. Camilli, S. Vargas, S. Ryan, W. S. Barnett, Teach. Coll.
Rec. 112, 579 (2010).
2. J. A. Temple, A. J. Reynolds, Econ. Educ. Rev. 26, 126
(2007).
3. E. Zigler, W. S. Gilliam, S. M. Jones, A Vision for
Universal Preschool Education (Cambridge University
Press, New York, 2006).
4. L. A. Karoly, M. R. Kilburn, J. S. Cannon, Early
Childhood Intervention: Proven Results, Future Promise
(RAND, Santa Monica, CA, 2005).
5. E. C. Melhuish et al., Science 321, 1161 (2008).
6. W. T. Gormley, D. Phillips, T. Gayer, Science 320, 1723
(2008).
7. A. J. Reynolds, J. A. Temple, Annu. Rev. Clin. Psychol. 4,
109 (2008).
8. Consortium for Longitudinal Studies, As the Twig Is Bent
... Lasting Effects of Preschool Programs (Erlbaum,
Hillsdale, NJ, 1983).
9. L. J. Schweinhart et al., Lifetime Effects: The High/Scope
Perry Preschool Study Through Age 40 (High/Scope
Press, Ypsilanti, MI, 2005).
10. P. Braveman, C. Barclay, Pediatrics 124, S163 (2009).
11. U.S. DHHS, Head Start Impact Study: Final Report (U.S.
DHHS, Washington, DC, 2010).

/ www.sciencexpress.org / 9 June 2011 / Page 4 / 10.1126/science.1203618

12. R. C. Pianta, J. Belsky, R. Houts, F. Morrison, Science
315, 1795 (2007).
13. Materials and methods are available as supporting
material on Science Online.
14. M. E. Lachman, Annu. Rev. Psychol. 55, 305 (2004).
15. M. Nation et al., Am. Psychol. 58, 449 (2003).
16. M. C. McCormick et al., Pediatrics 117, 771 (2006).
17. D. L. Johnson, J. Blumenthal, J. Prim. Prev. 25, 195
(2004).
18. G. W. Imbens, J. M. Wooldridge, J. Econ. Lit. 47, 5
(2009).
19. Chicago Longitudinal Study, User’s Guide, Version 7
(University of Minnesota, Minneapolis, 2005).
20. A. J. Reynolds, Success in Early Intervention: The
Chicago Child-Parent Centers (University of Nebraska
Press, Lincoln, 2000).
21. E. Garces, D. Thomas, J. Currie, Am. Econ. Rev. 92, 999
(2002).
22. A. J. Reynolds et al., Arch. Pediatr. Adolesc. Med. 161,
730 (2007).
23. N. Naisbitt, Child-Parent Education Centers: ESEA Title
I, Activity I (unpublished report, Chicago, 1968).
24. J. M. Robins, M. A. Hernan, B. Brumback, Epidemiology
11, 550 (2000).
25. K. Hirano, G. W. Imbens, G. Ridder, Econometrica 71,
1161 (2003).
26. A. J. Reynolds, S. Ou, Child Dev. 82, 555 (2011).
27. A. J. Reynolds, S. Ou, J. W. Topitzes, Child Dev. 75,
1299 (2004).
28. M. A. Cohen, The Costs of Crime and Justice (Routledge,
New York, 2005).
29. M. E. O’Connell, T. Boat, K. E. Warner, Preventing
Mental, Emotional, and Behavioral Disorders Among
Young People (National Academy Press, Washington, DC,
2009).
30. R. J. Sampson, P. Sharkey, S. W. Raudenbush, PNAS 105,
845 (2008).
31. P. L. Chase-Lansdale et al., Science 299, 1548 (2003).
32. A. Singh-Manoux et al., Int. J. Epidemiol. 33, 1072
(2004).
33. J. D. Hawkins et al., Arch. Pediatr. Adolesc. Med. 162,
1133 (2008).
34. A. J. Reynolds, Early Child. Res. Q. 10, 1 (1995).
35. A. J. Reynolds, S. Ou, J. A. Temple, Child Welfare 82,
379 (2003).
Acknowledgments: This research was supported by grants
from the National Institute of Child Health and Human
Development (HD034294), McKnight Foundation, and the
Human Capital Research Collaborative (a partnership of
the University of Minnesota and Federal Reserve Bank of
Minneapolis). We thank the Chicago Public Schools for
invaluable collaboration on the study over the years. We

also are grateful to the Illinois Departments of
Employment Security, Human Services, Corrections,
Child and Family Services, and Public Health; Circuit and
Juvenile Courts of Cook County; City Colleges of
Chicago; Chapin Hall Center for Children at the
University of Chicago; and the Health Survey Center at the
University of Minnesota for assistance and cooperation in
data collection.
Supporting Online Material
www.sciencemag.org/cgi/content/full/science.1203618/DC1
Materials and Methods
SOM Text
Figs S1 to S6
Tables S1 to S14
References
31 January 2011; accepted 19May 2011Published online 9
June 2011; 10.1126/science.1203618
Fig. 1. Well-being for selected outcomes by maternal
education, gender, and program groups. Error bars represent
±1 SE. Means and rates on the outcome are adjusted for 15
preprogram characteristics (table S2) and attrition by IPW.
Extended intervention is 4 or more years of CPC participation
from preschool to third grade versus participation for 3 years
of less. Outcomes were measured by age 28 from multiple
source including administrative data and adult surveys.
Mothers education was measured by age 3 of the study
participant.
Fig. 2. Robustness estimates for SES and felony arrest by
model specification. Error bars represent ±1 SE. The y axis
represents marginal effects in percentage points. Models are
adjusted for 15 preprogram characteristics (table S2) except
Model 1. Models 1 is unadjusted. Model 2 is adjusted with
covariates. Model 3 is adjusted for attrition by IPW. Model 4
is adjusted for attrition and program selection by IPW. Base
rates of unadjusted comparison group are (A) 29.3%. (B)
32.4%, (C) 31.4%, (D) 25.6%, (E) 21.5%, and (F) 22.7%.

/ www.sciencexpress.org / 9 June 2011 / Page 5 / 10.1126/science.1203618

Table 1. Patterns of participation and sample recovery of Child-Parent Center (CPC) Education Program and
comparison groups in the Chicago Longitudinal Study. Cases for program participation span the 6-year period
(1983–1989) that defines enrollment in the CPC intervention. Among the comparison group, 389 cases were from
randomly selected schools participating in an alternative kindergarten intervention. 176 cases in the comparison
group were eligible to receive limited services in CPC kindergarten but enrolled in different classrooms. Some cases
in the comparison group participated in the school-age program because it was open to any child enrolled in
elementary school from first to third grade. Cases were lost during post-program years primarily because they
moved from Chicago and could not be located, were deceased, or did not have sufficient identifying information to
track. The number of equivalent covariates shown is for the preschool group (table S2a). The respective numbers for
the school-age and the extended intervention groups were 18 and 15 (tables S2b and S2c). For the latter, one
covariate was equivalent (adult administrative records) after adjustments for child and family preprogram
characteristics. Only 15 of the 20 covariates were included in the analyses.
Study category

Total sample

CPC
intervention
group

Comparison
group

Original sample

1539

989

550

1073
74.2
850 (55)
553 (36)
171 (11)

989
59.9
684 (69)
553 (56)
104 (11)

(84)
100
166 (30)
0 (0)
67 (12)

1386 (90)
1233, 1473
80, 96
28.29

900 (91)
808, 950
82, 96
28.27
18

486 (88)
425, 523
77, 95
28.32

Program participation
No. in center-based preschool (Head Start)
Full-day kindergarten (%)
No. with CPC school-age participation (%)
No. with CPC extended intervention (4–6 years) (%)
No. lost due to mobility, mortality, or other (%)
Sample recovery and characteristics by age 28
No. with educational attainment/employment (%)
No. (min, max) for family, health, and justice outcomes
Percentage sample recovery for min, max (%)
Average age on 31 August 2008 (years)
No. of covariates equivalent with comparison group (of 20)

Table 2. Means and group differences for selected adult outcomes by age 28 adjusted for attrition by inverse probability weighting (IPW) and preprogram
characteristics. See table S9 for all outcomes and findings for the alternative extended intervention contrast. Extended intervention-1 = Child-Parent Center
(CPC) participation for 4 to 6 years (preschool to second or third grade) versus fewer years. Unadj and adj. diff. = unadjusted and adjusted group difference. All
adjusted models used robust standard errors, and attrition was taken into account through IPW as a sampling weight in the model for each study outcome. Sample
sizes vary by measure. Ages of assessment were are follows: educational attainment (28.3), socioeconomic status (SES; 27.6), health status and behavior (27.6),
and crime and justice system involvement (26.6). SES ≥ 4 = index of education and income from 0 to 8.
Adult outcome

Unadj.
diff.

CPC preschool1
Interv
Comp

Adj.
diff.

Unadj.
diff.

CPC school-age2
Interv
Comp

Adj.
diff.

CPC extended intervention-13
Unadj.
Interv
Comp
Adj.
diff.
diff.

Educational attainment
Highest grade completed in years

0.33**

12.15

11.88

0.27*

0.16t

12.07

12.03

0.04

0.33**

12.21

11.95

0.26*

High school completion

7.6**

81.5

75.1

6.4**

3.8t

80.0

78.5

1.5

6.6**

82.7

77.2

5.5**

On-time high school graduation, %

9.6**

44.3

36.6

7.7*

7.9**

44.4

35.3

9.1*

12.1**

48.6

31.3

17.3**

BA or AA degree, %

0.8

8.4

8.5

-0.1

1.4

8.8

7.4

1.4

2.1

9.5

8.3

1.2

7.1*

34.4

28.6

5.7*

2.8

32.8

31.6

1.2

6.7*

35.9

30.3

5.6*

932**

11,582

10,796

786*

-28

2.5

49.1

44.8

4.3

Any health insurance coverage, %

10.0*

75.9

63.9

Substance abuse (excluding alcohol), %

-6.5**

13.7

Any arrest (including self reports), %

-6.2*

Felony arrest, %
Any conviction, %

Socioeconomic status
SES ≥ 4, %
Average annual income (2007 dollars)
Food stamp participation, ages 24–27, %

11,250

11,278

1,102t

11,822

10,942

-2.9

54

43.9

52.0

-8.1*

-1.8

45.0

48.9

880
-3.9

12.0**

0.3

70.5

73.7

-3.1

6.7*

75.7

69.6

6.1**

18.9

-5.2*

-0.1

16.1

14.7

1.4

-3.6

14.3

16.2

-1.9

47.9

54.3

-6.4*

4.9t

52.4

47.5

4.9t

-1.5

51.1

49.7

1.4

-6.4**

19.3

24.6

-5.3*

1.2

21.6

20.4

1.2

-3.2

19.5

21.2

-1.7

-5.2*

25.1

28.8

-3.7

1.1

27.0

25.9

1.1

-4.4t

24.1

26.7

-2.6

Health status and behavior

Crime and justice system involvement

1

Adjusted for school-age participation, 8 indicators of pre-program risk status (table S2), sex of child, race/ethnicity, child welfare history by age 4, neighborhood
poverty at 1980, a dummy-coded variable for missing data on risk status, and home environment problems at ages 0–5. 2Adjusted for preschool participation, 8
indicators of pre-program risk status (table S2), sex of child, race/ethnicity, child welfare history by age 4, neighborhood poverty at 1980, a dummy-coded
variable for missing data on risk status, and home environment problems at ages 0–5. 3Adjusted for 8 indicators of pre-program risk status (table S2), sex of
child, race/ethnicity, child welfare history by age 4, neighborhood poverty at 1980, a dummy-coded variable for missing data on risk status, and home
environment problems at ages 0–5. ** p < 0.01, * p < 0.05, and t < 0.10.

Rates by maternal education and CPC preschool program

100 A. High school completion 30
25
90
~ 80
20
ro

e. Substance abuse

30

25
20

~ 70

15

15

rf. 60

10

10

50

5

5

o

o

e

40
Mother
completed
HS

Mother not
completed
HS

Mother
Mother not
completed completed
HS
HS
• Preschool 0 No preschool

Mother
completed
HS

Mother not
completed
HS

High school completion rate by gender and CPC program

100

D.

epe preschool

100

E.

epe school-age

100

90

90

~ 80

80

80

70

70

60

60

ro

~u 70
Q; 60

a..

50
Female

Male

• Preschool 0 No Preschool

epe extended

90

50

50

40

F.

40

40
Female

Male

• School-age 0 No school-age

Female

Male

• Extended 0 Less extended

A.

epe preschool program

SES
B.

~4

epe School-age program

e. epe Extended participation

9

4.5

9

8

4

8

;i7

3.5

7

en 6

3

6

&5
<1>

2.5

5

(ii4

2

4

'~3

1.5

3

:2:2

1

2

1

0.5

1

o

o

o

~

t5
c

ctl

1

2
3
Model

-0.5

4

1

2
3
Model

4

2
3
Model

1

4

Felony Arrest
D.

epe preschool program

E.

epe school-age program

F.

epe extended participation

o

3

o

...-.--1

2.5
2

-0.5
-1

1.5

-1.5

~

--2
en

~-3

:1=

<1>-4
(ii

·~-5

1

-2

0.5

-2.5

o +-JI1lfllL-r-

2'

-3

~-6

-0.5

-7

-1

-3.5
-4

-8

-1.5

-4.5

234
Model

234
Model

1

2

3

Model

4