The Evidence on Teaching

The Perry Preschoolers at Late Midlife: A Study in Design-Specific Inference

Heckman, J. J., & Karapakula, G. · 2019

grade Breplicationdeveloper-involvedmixed
Sample
123 (58 T / 65 C); ~102 surveyed at late midlife (83% of sample; 12% deceased)
Population
Perry participants at age ~55.
Design
Age-55 follow-up using worst-case randomization inference. Monte Carlo shows standard asymptotic/bootstrap/permutation tests OVER-REJECT, which the authors say justifies 'some of the doubts of the critics'.
Key findings
Perry's crime, employment and health effects survive conservative design-specific inference at age 55 — an important defence of the study. But note the direct conflict with Anderson (2008): Heckman-Karapakula find effects strongest for MALES, Anderson finds the durable effects are FEMALE. Both cannot be the headline.
Counterfactual
No preschool, no home visits.
Genetic confound
RCT-based; the male-strongest pattern here CONTRADICTS Anderson's female-strongest multiplicity result — an unresolved tension worth recording.
Replication notes
Companion paper reports intergenerational gains in participants' children (observational within an experimental frame, grade C).

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Cumulative violent misdemeanour convictionsconvictions/person (AIPW)-0.53 at 30; -0.65 at 40; -0.69 at 50mixedadultbusiness-as-usualadulthoodbehaviour
Employment, health, cognition, personality at late midlifesurvival of worst-case testseffects survive; strongest for MALESmixedadultbusiness-as-usualadulthoodattainment

Cited by

The Perry Preschoolers at Late Midlife: A Study in Design-Specific Inference · The Evidence on Teaching