The Evidence on Teaching

Early Childhood Intervention and Life-Cycle Skill Development: Evidence from Head Start

Deming, D. J. · 2009

grade Clongitudinalindependentfailednumbers spot-checked
Sample
1,251 (sibling fixed-effects identifying sample)
Population
NLSY children; siblings discordant for Head Start attendance.
Design
Sibling fixed effects. IMPORTANT LIMIT: this removes shared family environment and about half of additive genetic variance, but does NOT control child-specific endowment — parents who enrol one sibling and not another plausibly select on child traits.
Key findings
The most-cited evidence that Head Start has durable adult benefits despite test-score fadeout. It should now be read alongside its failed replication: the result does not hold in later cohorts, and the identifying variation is within-family selection that a sibling design cannot fully remove.
Counterfactual
Siblings who attended other preschool or no preschool; the headline 0.23 SD is versus NO preschool.
Genetic confound
HIGH for a within-family design: sibling FE does not remove child-specific genetic endowment or differential parental selection.
Replication notes
FAILED REPLICATION — see Pages et al. (2020): the effect is +0.17 in Deming's own cohorts, -0.15 in later cohorts, and ~zero combined.
DOI / URL
10.1257/app.1.3.111

Effects

OutcomeMetricValueMeasureTimingVsHorizonClass
Young-adult summary index (schooling, idleness, crime, teen parenthood)SD+0.23 vs siblings with no preschoolmixedadultbusiness-as-usualadulthoodattainment
High school graduationpp+8.5ppmixedadultbusiness-as-usualadulthoodattainment
Cognitive test index, ages 5-6 -> ages 11-14SD+0.15 -> +0.06 — Deming's own data show the test-score fadeoutstandardizedgrade-6+business-as-usualover-2yrdomain-skill

Cited by