Using Experiments to Estimate the Effects of Education on Voter Turnout

1,636 subjects pooled: Perry Preschool 123 (60 T / 63 C), Project STAR 1,455 (429 T / 1,026 C), I Have a Dream 58 (19 T / 39 C)

Three cohorts of disadvantaged American children: Perry Preschool (low-SES Ypsilanti MI preschoolers, 1960s), Project STAR (Tennessee kindergarteners, 1985 on), I Have a Dream (Boulder CO fifth-graders, 1992 on). Turnout measured in adulthood, 2000-2004.

Re-analysis of two genuine RCTs (Perry Preschool; Project STAR class-size reduction) plus one quasi-experiment (I Have a Dream scholarship/mentoring, compared against neighbouring cohorts). None was designed to study voting. Subjects were matched to public voting records, so the outcome is VALIDATED turnout - a real strength. The fatal weakness is power: 1,636 subjects total, only 58 in the IHAD arm and 123 in Perry, and the causal chain runs treatment to high-school graduation to voting, which loses precision at every step. Cohodes and Feigenbaum, citing this paper as their closest predecessor, note the total sample is under 20% of theirs. Secondary summaries of this paper disagree about whether the pooled estimate is statistically distinguishable from zero, which is itself a signal about the precision. READ DEPTH CAVEAT: the AJPS article is paywalled; details here come from the AEA registry entry, J-PAL's listing, Campbell's 2019 review and Cohodes and Feigenbaum's description.

Across three interventions that raised high-school graduation, the treated groups voted at higher rates as adults on validated records. Reported per-site differences: Perry Preschool +46% graduation and +44% turnout in 2000/2002; I Have a Dream +28% graduation and +26% turnout through 2004; Project STAR +6% graduation and 10-11% higher voting. The authors concluded that 'exogenously induced changes in high school graduation rates have powerful effects on voter turnout rates' and that education is causal rather than proxy. The estimates are large but rest on tiny samples, and secondary sources differ on their statistical significance - which is the honest reading: the sign is consistent, the precision is poor, and the magnitudes are implausibly large next to every later well-powered study.

Low to medium. Two of the three arms are randomized (Perry, STAR), so genes are balanced. The IHAD arm is a cohort comparison and carries selection risk. The bigger threat is not confounding but sampling error at n=58 and n=123.

The headline claim - that exogenously raising high-school graduation raises turnout - is contradicted by every subsequent well-powered non-US natural experiment (Pelkonen 2012 Norway null; Lindgren et al. 2019 Sweden average null on 417,000 register-validated voters) and by Berinsky and Lenz 2011 in the US. It is supported by the two charter lottery studies, but those move QUALITY not quantity. Treat as mixed.