The Rate of Return to the HighScope Perry Preschool Program
Heckman, J. J., Moon, S. H., Pinto, R., Savelyev, P. A., & Yavitz, A. Q. · 2010
grade Dquasi-experimentdeveloper-involvedunreplicatednumbers spot-checked
Sample
123 (58 T / 65 C)
Population
Perry participants, with earnings extrapolated from age 40 to 65.
Design
A MODEL OUTPUT, not a measured effect — graded D on that basis. Requires extrapolating earnings beyond the last observed age and pricing crime (including murder). Corrects standard errors for the compromised randomization.
Key findings
The source of the famous 'preschool returns 7-10% a year' claim — and it is a model extrapolation, not a measured effect. Its own authors revised earlier 16-17% claims DOWN. The benefit-cost ratio is load-bearing on crime savings, which rest on the weakest part of the evidence (no adult Perry crime effect survives multiplicity correction).
Counterfactual
No preschool, no home visits.
Genetic confound
N/A (economic model).
Replication notes
Audit note: the per-gender x deadweight-loss IRR cells circulated in secondary sources do not reconcile with the paper (standard errors mis-paired across columns) and a widely quoted '7 to 12 dollars per person' benefit-cost phrase returns no hits in the full text — treat both as UNVERIFIED and cite only the 7-10% headline and the murder-pricing sensitivity.
Cost
~88% of the modeled public return is crime savings; ~93% attributable to males. Abecedarian's crime effect is null, so this channel does not replicate.
DOI / URL
10.1016/j.jpubeco.2009.11.001
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Headline social internal rate of return | % per annum | 'generally fall between 7 and 10%' — REVISED DOWN from prior claims of 16% (Rolnick-Grunewald) and 17% (Belfield) | unknown | adult | business-as-usual | adulthood | attainment |
| Benefit-cost ratio, sensitivity to how murder is priced | $ per $ at 3% discount | pooled 11.6 (high murder cost) vs 4.5 (low) — the ratio moves by 2.5x purely on this assumption, and there are FOUR murders in the entire record | unknown | adult | business-as-usual | adulthood | attainment |
| Cell IRR range across specifications | % per annum | 5.3-17.1%, standard errors 1.0-5.4pp | unknown | adult | business-as-usual | adulthood | attainment |
Cited by
- Perry Preschool and Abecedarian — what the famous studies actually establishmixedconf: mediumgc: low