Does Preschool Pay Off Later? What the Long-Run Evidence Actually Shows

The question

Preschool is one of the most confidently sold policies in education. The pitch is simple: enroll disadvantaged children early, narrow the school-readiness gap, and watch the benefits compound into higher graduation rates, better jobs, less crime, and better health. The claim draws on famous studies — Perry Preschool, Abecedarian — that followed children for decades and reported returns of 6 to 10 percent per year.

But the same literature contains an awkward fact. The test-score advantages from preschool usually shrink or vanish within a few years. And the programs that produced the spectacular long-run results were small, intensive, expensive, and run decades ago for a narrowly selected group of children. Modern public programs, evaluated with modern methods, often produce much smaller effects — and occasionally negative ones.

This brief asks a bounded question: do preschool programs produce lasting gains, and what separates the programs that persist from those that fade or backfire? The short answer is that preschool reliably improves school readiness, that test-score fade-out is normal and not by itself evidence of failure, and that long-run effects are real but highly conditional. The average effect across programs is a poor guide to any particular program.

The short-run case is strong

The least controversial finding in the field is that preschool raises early academic skills. Randomized evaluations of Head Start, state pre-K programs, and model programs typically find gains of roughly 0.2 to 0.3 standard deviations in vocabulary, literacy, and early math at school entry, with the largest gains for children from low-income families. A meta-analysis of universal early childhood programs using natural experiments — 250 estimates from 30 studies published between 2005 and 2017 — found that effects were mixed overall but consistently more favorable for cognitive and academic outcomes, for publicly provided programs, and for children from lower socioeconomic backgrounds (van Huizen & Plantenga, 2018).

That short-run effect is real, and it is not trivial. A 0.2 to 0.3 standard deviation gain at age five is large enough to matter for school entry. The question is what happens next.

Fade-out is the norm, not the exception

Almost every preschool evaluation shows the same trajectory: effects peak at school entry, decline through the early grades, and are often statistically indistinguishable from zero by third grade. The Head Start Impact Study, the program's first large randomized evaluation, found modest gains at the end of the Head Start year; by first and third grade, the impacts were mixed or mostly gone (Puma et al., 2012). Tennessee's voluntary pre-K program, evaluated with an oversubscribed-site lottery, also found initial gains that did not survive the early grades.

Fade-out is sometimes treated as proof that preschool does not work. That inference is too quick, for two reasons. First, fade-out can reflect catch-up by the control group rather than decay of the treatment. If preschool gives a child a head start and elementary school then teaches everyone the same material, the gap narrows because the control group accelerates, not because the treated group lost anything. Second, fade-out on tests can coexist with persistent effects on outcomes that tests do not capture — attainment, behavior, health, and employment.

Still, fade-out is a real phenomenon and it is the default. Any claim that a program produced lasting academic gains has to explain why this program escaped a pattern that holds across decades of evaluations.

Long-run effects are real but narrower than the slogan

The strongest long-run evidence comes from two small model programs.

Perry Preschool randomized 123 African American children from low-income families in Ypsilanti, Michigan, in the 1960s; 58 were assigned to a high-quality, two-year preschool program with home visits. Follow-ups through age 40 found higher high school graduation, higher earnings, less crime, and a rate of return estimated at 6 to 10 percent per year, statistically significant for both males and females (Heckman et al., 2010). The sample is tiny, the randomization was compromised in ways that require careful adjustment, and the program cost far more per child than a typical public pre-K slot.

Abecedarian randomized 111 infants from low-income families in North Carolina to full-day, year-round care from infancy to age five, with a structured language-rich curriculum. At age 30, treated participants had completed about 1.15 more years of education (13.46 versus 12.31 years); income-to-needs ratios and criminal involvement did not differ significantly (Campbell et al., 2012). A biomedical follow-up in the mid-30s found markedly better cardiovascular and metabolic profiles, especially among males: mean systolic blood pressure was 126 mm Hg in the treated group versus 143 mm Hg in controls, and one in four control males met criteria for metabolic syndrome versus none of the treated males (Campbell et al., 2014). Those health results are striking, but they come from a small sample and required adjustment for multiple hypotheses and nonrandom attrition.

Both programs served children with few alternatives, at intensities that modern public programs do not match. Perry was two years of part-day preschool with well-trained teachers and home visits. Abecedarian was five years of full-day care starting in infancy. Neither is a template for a one-year universal pre-K program.

The long-run evidence at scale is more equivocal. Quasi-experimental studies of Head Start's earlier cohorts find higher schooling, higher earnings, and less crime (Deming, 2009; Carneiro & Ginja, 2014), even though the randomized Head Start Impact Study found test-score effects that faded. Reanalyses of that experiment find considerable variation by center and more persistent benefits for some subgroups, including Hispanic children and children with lower skills at entry. Boston's universal preschool program, evaluated with admissions lotteries, raised on-time college enrollment by 8.3 percentage points against a 46 percent baseline, increased ever-college enrollment by 5.4 points, raised high school graduation by about 6 points, and reduced juvenile incarceration — while producing no detectable effect on state achievement tests (Gray-Lobe et al., 2023). Effects were generally larger for boys.

Tennessee points the other way. The randomized evaluation of its voluntary pre-K program found negative effects on academic measures through sixth grade, with weighted intent-to-treat effect sizes of about −0.15 in English language arts, −0.21 in math, and −0.16 in science; treatment-on-the-treated math estimates reached −0.40. Children assigned to pre-K were also more likely to be placed in special education and to be disciplined for school rule violations (Durkin et al., 2022). Those are not small or ambiguous estimates. They are evidence that a scaled public program, delivered as implemented, can leave children worse off on the outcomes the program was designed to improve.

Quebec's universal child care expansion is the clearest case of a large program doing harm. The province introduced highly subsidized care in the late 1990s and expanded supply rapidly. Baker, Gruber, and Milligan (2019) found that the expansion produced lasting negative effects on children's non-cognitive outcomes, worse self-reported health and life satisfaction, and higher rates of criminal behavior, with the damage concentrated among boys. The program increased maternal employment, which was one of its goals, but the child-development results were bad.

Put together, the long-run record is not "preschool works" or "preschool fails." It is a distribution: large positive effects in some programs, null effects in many, and negative effects in a few. The distribution is not random, and the next sections describe what moves a program from one part of it to another.

The programs that worked were not typical

The Perry and Abecedarian results are often cited as if they establish the returns to preschool in general. They do not. Three differences matter.

The first is intensity and duration. Perry provided two years of a structured program; Abecedarian provided five years of full-day care from infancy. Most public pre-K programs provide one year of part-day or school-day instruction at age four. Dosage is not a minor implementation detail; it changes the size of the treatment.

The second is quality. The model programs had well-trained teachers, low child-to-adult ratios, structured curricula focused on language and reasoning, and in Perry's case home visits. Quality is hard to measure and easy to assert, but the evidence consistently finds that process quality — the quality of interactions, not just the presence of a certified teacher — predicts effects. The van Huizen and Plantenga meta-analysis found that quality mattered more than enrollment age or program intensity in explaining whether universal programs helped or hurt.

The third is the counterfactual. Perry and Abecedarian served children who, in the absence of the program, would have had little or no early education. A modern public pre-K program in a city where many children already attend private or Head Start programs is treating a different margin. If the control group is already receiving decent early care, the incremental effect of a public slot is smaller by construction.

The counterfactual explains much of the decline

The most persuasive recent account of why modern programs look weaker than the famous ones is not that preschool got worse. It is that the comparison group got better. Whitaker et al. (2026) review the evidence and argue that the most compelling explanation for weaker effects is the improvement in the early childhood conditions of children who do not enroll in public programs: more center-based care, more preschool, more early literacy in the home. As the counterfactual improves, the measured effect of a public slot shrinks even if the program itself is unchanged.

This has an uncomfortable implication for evaluation. A null result from a modern pre-K study may mean the program adds little for the children it serves, or it may mean the program is being compared with a control group that is already doing relatively well. The policy question is not "does preschool work in general" but "does this program add value relative to what these children would otherwise get."

The counterfactual also explains why the same program can look effective for one subgroup and useless for another. Children with the weakest home and neighborhood environments have the most to gain from a high-quality alternative; children from more advantaged families may gain little or may be displaced from care they already had. Boston's larger effects for boys and Tennessee's negative effects for a low-income applicant sample are consistent with the idea that effects depend on who is treated and what they would otherwise receive.

What separates persistence from fade

Several mechanisms recur in the literature, and they are not mutually exclusive.

Instructional alignment. If kindergarten and first grade repeat what children learned in preschool, the treated group has little new to learn and the control group catches up. Curricular redundancy is a plausible contributor to fade-out, and programs that build on preschool skills — rather than restarting — show better persistence. This is a school-side explanation, not a preschool-side one.

Peer composition. List and Uchida (2024) implemented a field experiment with two stages of randomization: first assigning preschool access, then assigning children to elementary school classrooms. They found that fade-out depended critically on the share of classmates who had also attended preschool. With enough treated peers, the classic fade-out pattern was muted. The result suggests that human capital accumulation is social: a treated child in a classroom of untreated peers may lose ground, while a cohort of treated children can sustain one another. It also implies that scaling a program can increase its effectiveness rather than dilute it, because scale changes the peer environment.

Outcome type. Test scores are narrow and proximal. Attainment, behavior, health, and employment are broader and more distal, and they can be affected through channels that tests do not capture: better non-cognitive skills, stronger attachment to school, lower incarceration risk, and improved health behaviors. Boston is the cleanest illustration: no state-test effect, sizable college and behavioral effects. Tennessee is the mirror image: negative test and behavior effects. The choice of outcome is not neutral; it determines which program looks successful.

Program quality and implementation. The same policy can be delivered well or badly. Tennessee's program had structural features that looked adequate on paper but produced negative effects in practice; the evaluators and later reviewers emphasized that process quality and instructional content were not well measured. Quebec's expansion was rapid, and the pattern is consistent with quality falling as supply grew; when a program scales faster than its quality infrastructure, the average treatment effect can fall below zero.

Statistical fragility. The model-program literature is small. Perry's sample of 123 children, with compromised randomization and many outcomes, is exactly the setting where multiple-hypothesis corrections matter. Anderson (2008) reanalyzed Perry, Abecedarian, and the Early Training Project with familywise error-rate and false-discovery-rate corrections and found that long-term benefits were concentrated among girls, with no significant long-term benefits for boys. Heckman and coauthors disputed the gender conclusion, and the disagreement has not been fully resolved. The episode is a reminder that "the returns to preschool" is an estimate with a standard error, not a constant.

What the evidence supports

The defensible synthesis is narrower than either side's talking points.

• Short-run academic gains: high confidence. Preschool raises school readiness for disadvantaged children by roughly 0.2 to 0.3 standard deviations.
• Test-score fade-out: high confidence. Gains typically shrink and often disappear on standardized tests within a few years.
• Long-run average effects of modern public pre-K: low-to-moderate confidence, probably small and positive on attainment and behavior, near zero on achievement tests, with substantial heterogeneity. Boston's college results are real but may not generalize to programs with different quality and different counterfactuals.
• Large long-run returns from intensive model programs: high confidence for the specific populations and programs studied; low confidence that the same returns would follow from scaling a one-year program.
• Negative effects: possible and documented. Tennessee and Quebec are not anomalies to be explained away; they are warnings that program design and implementation matter enough to reverse the sign.

A reasonable calibration: a typical one-year public pre-K program serving low-income children probably raises high school graduation by somewhere between 0 and 5 percentage points, probably has a small positive effect on adult earnings, and probably has little or no average effect on standardized achievement by adolescence. The confidence interval around those statements is wide, and the program-to-program variation is larger than the average.

What would change the answer

Three kinds of evidence would move this assessment.

First, more randomized or lottery-based evaluations of modern public programs with long-run follow-up. Boston is one; there need to be others in different cities, with different quality levels and different counterfactuals. The current long-run evidence is dominated by programs from the 1960s and 1970s and by quasi-experimental designs.

Second, better measurement of program quality and implementation. The field's inability to say why Tennessee's program produced negative effects — whether it was curriculum, teacher quality, peer composition, or something else — is a major limitation. Without process measures, "preschool" remains a black box whose average effect is an average over very different treatments.

Third, tests of the peer-composition and alignment mechanisms at scale. If List and Uchida's result holds, the optimal policy is not simply to expand access but to expand it in a way that creates sustained treated cohorts and aligned elementary instruction. That is a different policy from "fund more slots," and it is testable.

Until then, the honest answer to "does preschool pay off later" is: it depends, and the dependence is systematic enough to be useful. Preschool is a reliable short-run investment, a plausible but modest long-run one, and a policy whose sign can flip when quality is low, the counterfactual is decent, or the program is scaled faster than it can be delivered well. The strongest version of the pro-preschool case is not that every program produces Perry-sized returns. It is that well-designed programs for disadvantaged children have produced some of the best-documented long-run effects in social policy, and that the failures are informative about how to build the next one.

References

1. Durkin K, Lipsey MW, Farran DC, Wiesen SE. Effects of a state prekindergarten program on children's achievement and behavior through sixth grade. Developmental Psychology. 2022;58(3):470–484. doi:10.1037/dev0001301. https://doi.org/10.1037/dev0001301
2. Gray-Lobe G, Pathak PA, Walters CR. The long-term effects of universal preschool in Boston. Quarterly Journal of Economics. 2023;138(1):363–411. doi:10.1093/qje/qjac036. https://doi.org/10.1093/qje/qjac036
3. Baker M, Gruber J, Milligan K. The long-run impacts of a universal child care program. American Economic Journal: Economic Policy. 2019;11(3):1–26. doi:10.1257/pol.20170603. https://doi.org/10.1257/pol.20170603
4. Campbell FA, Pungello EP, Burchinal M, et al. Adult outcomes as a function of an early childhood educational program: an Abecedarian Project follow-up. Developmental Psychology. 2012;48(4):1033–1043. doi:10.1037/a0026644. https://doi.org/10.1037/a0026644
5. Campbell FA, Conti G, Heckman JJ, et al. Early childhood investments substantially boost adult health. Science. 2014;343(6178):1478–1485. doi:10.1126/science.1248429. https://doi.org/10.1126/science.1248429
6. Heckman JJ, Moon SH, Pinto R, Savelyev PA, Yavitz A. The rate of return to the HighScope Perry Preschool Program. Journal of Public Economics. 2010;94(1–2):114–128. doi:10.1016/j.jpubeco.2009.11.001. https://doi.org/10.1016/j.jpubeco.2009.11.001
7. Anderson ML. Multiple inference and gender differences in the effects of early intervention: a reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects. Journal of the American Statistical Association. 2008;103(484):1481–1495. doi:10.1198/016214508000000841. https://doi.org/10.1198/016214508000000841
8. Puma M, Bell S, Cook R, Heid C, Broene P, Jenkins F, Mashburn A, Downer J. Third Grade Follow-Up to the Head Start Impact Study: Final Report. OPRE Report 2012-45b. Washington, DC: Office of Planning, Research and Evaluation, Administration for Children and Families, U.S. Department of Health and Human Services; 2012. https://acf.gov/sites/default/files/documents/opre/head_start_executive_summary.pdf
9. Deming D. Early childhood intervention and life-cycle skill development: evidence from Head Start. American Economic Journal: Applied Economics. 2009;1(3):111–134. doi:10.1257/app.1.3.111. https://doi.org/10.1257/app.1.3.111
10. Carneiro P, Ginja R. Long-term impacts of compensatory preschool on health and behavior: evidence from Head Start. American Economic Journal: Economic Policy. 2014;6(4):135–173. doi:10.1257/pol.6.4.135. https://doi.org/10.1257/pol.6.4.135
11. van Huizen T, Plantenga J. Do children benefit from universal early childhood education and care? A meta-analysis of evidence from natural experiments. Economics of Education Review. 2018;66:206–222. doi:10.1016/j.econedurev.2018.08.001. https://doi.org/10.1016/j.econedurev.2018.08.001
12. Whitaker AA, Burchinal M, Jenkins JM, Bailey DH, Watts TW, Duncan GJ, Hart ER, Peisner-Feinberg ES. Why are preschool programs becoming less effective? Journal of Policy Analysis and Management. 2026;45(1):e70031. doi:10.1002/pam.70031. https://doi.org/10.1002/pam.70031
13. List JA, Uchida H. Here today, gone tomorrow? Toward an understanding of fade-out in early childhood education programs. NBER Working Paper 33027. October 2024, revised July 2026. doi:10.3386/w33027. https://doi.org/10.3386/w33027
14. Watts TW, Jenkins JM, Dodge KA, et al. Understanding heterogeneity in the impact of public preschool programs. Monographs of the Society for Research in Child Development. 2023;88(1):7–182. doi:10.1111/mono.12463. https://doi.org/10.1111/mono.12463