Prep School for Poor Kids: The Long-Run Impacts of Head Start on Human Capital and Economic Self-Sufficiency
Bailey, M. J., Sun, S., & Timpe, B. · 2021
grade Bnatural-experimentindependentmixednumbers spot-checked
Sample
22.48 million children born 1950-1980 in the continental US, restricted to ages 25-54 at observation
Population
US cohorts exposed to Head Start's county-level rollout in the 1960s-70s.
Design
Event-study/rollout design on a very large administrative sample (restricted long-form 2000 census + 2001-2018 ACS linked to SSA Numident) — the strongest long-run Head Start design available. Identification is a three-part linear spline in age at programme launch with a tested trend break at age 6. Two caveats the headline numbers hide: (1) every quoted effect is an ATET obtained by dividing the ITT by an estimated take-up of 0.149 (SE 0.022), so the ITT on years of schooling is +0.096 years, not +0.65; (2) the preferred estimates are additionally pre-trend adjusted. The data contain no family background covariates at all.
Key findings
The best long-run Head Start evidence. Real attainment gains — ATET +0.65 years of schooling, +12 pp college completion (+39% off a 30% base), +2.4 pp high school completion, human capital index +0.18 SD — plus economic self-sufficiency +0.09 SD, adult poverty -23% and public assistance receipt -27%. But directly measured wage earnings show NO effect (log labor income +0.043, CI -0.022 to 0.12), and incarceration is a precise-ish null (-0.39 pp, CI -1.2 to 0.49) that excludes the large crime effects claimed for Perry. Every headline number is an ATET = ITT / 0.149 take-up; the ITT on years of schooling is +0.096. Note the era: this is a counterfactual of near-zero alternative preschool (public kindergarten was not yet universal), which no longer exists.
Counterfactual
Same-county children too old to be age-eligible when the program arrived, plus not-yet-treated counties.
Genetic confound
Rollout timing is plausibly exogenous to family genetics — a genuine strength versus the sibling designs.
Replication notes
Direction replicates across independent designs (Garces-Thomas-Currie sibling PSID, Ludwig-Miller RD, Thompson spending, Johnson-Jackson, De Haan-Leuven), but magnitude does not: the authors themselves report their high-school-completion estimate is one-quarter the size of Deming (2009), one-sixth Johnson-Jackson and one-seventh Ludwig-Miller, and their college-enrollment estimate one-quarter of Ludwig-Miller. Meanwhile Pages et al. (2020) is a failed direct replication of the Deming sibling result on later cohorts. Hence mixed rather than replicated: same sign, order-of-magnitude disagreement, and one failed replication of a neighbouring finding.
DOI / URL
10.1257/aer.20181801
Effects
| Outcome | Metric | Value | Measure | Timing | Vs | Horizon | Class |
|---|---|---|---|---|---|---|---|
| Adult human capital index (Table 1) | SD | +0.18 SD ATET [0.097, 0.28] (pre-trend adjusted); underlying ITT only +0.027 SD | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Years of completed schooling (Table 1) | years | +0.65 years ATET [0.40, 1.01]; ITT is +0.096 years, scaled by take-up 0.149 | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| High school completion or GED (Table 1) | percentage points | +2.4 pp ATET [0.40, 5.2] on a control mean of 92% (= +2.7%) | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| College enrollment / attended some college (Table 1) | percentage points | +5.4 pp ATET [0.092, 11] on a control mean of 64% (= +8.5%) | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| College completion (Table 1) | percentage points | +12 pp ATET [6.7, 19] on a control mean of 30% (= +39%); the base is 30%, not a very low base | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Economic self-sufficiency index (Table 4) | SD | +0.092 SD ATET [0.041, 0.15] (pre-trend adjusted) | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Log labor income (wage earnings, actually measured; Table 4) | log points | +0.043 ATET, 95% CI [-0.022, 0.12] — null; authors attribute this to negative selection into employment | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Adult poverty (Table 4) | percentage points | -2.7 pp ATET [-5.2, -0.71] on a control mean of 12% (= -23%) | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Public assistance receipt (Table 4) | percentage points | -3.2 pp ATET [-5.7, -1.2] on a control mean of 12% (= -27%); for men alone -4.8 pp (-42%) | administrative | ages 25-54 | business-as-usual | adulthood | attainment |
| Incarceration in adulthood (Table 6) | percentage points | -0.39 pp ATET, 95% CI [-1.2, 0.49] — NULL; CIs often exclude the large crime reductions reported for Perry and by Barr-Gibbs | administrative | at time of census/ACS observation | business-as-usual | adulthood | behaviour |
Cited by
- Preschool at scale — Head Start, state pre-K, and what universal provision deliversmixedconf: highgc: low