Why Does Education Increase Voting? Evidence from Boston's Charter Schools

9,560 charter-lottery applicants (main voting sample); subsamples 6,354-8,258 for test-score outcomes

Boston charter-school lottery applicants, projected high-school classes of 2006-2017, old enough to vote by 2016. Roughly 74% free/reduced-price lunch, overwhelmingly Black and Latino, baseline test scores well below the Massachusetts mean. Applied in grades 5-9; outcomes measured at ages 18-26.

Lottery-based 2SLS: randomly assigned initial and waitlist charter offers instrument for charter attendance, with lottery risk-set fixed effects. This is the strongest identification available in the whole civic-education literature - genes, family motivation and school-choice taste are balanced by randomization (covariate balance and a PREDICTED-voting placebo both pass; predicted-voting 2SLS is -0.005, SE 0.004). Outcome is VALIDATED turnout from Massachusetts and neighbouring-state voter files matched by name and birth date; ~90% of applicants matched, with no differential match by offer status. NOTE ON WHAT WAS READ: figures here are taken from NBER Working Paper 29308 (September 2021, revised July 2025), read in full; the version of record is Review of Economics and Statistics (2025), which was not re-read, so numbers may have moved slightly. Crucially, at the first presidential election there is NO difference in years of education between lottery winners and losers (<0.1 years, not significant), so the voting effect cannot be the education-quantity channel that Dee 2004 and Milligan et al. 2004 identify.

Charter attendance raised voting in the first presidential election after turning 18 by about 6 percentage points against a counterfactual of 35% - a 17% relative increase - on validated voter-file turnout. Share of presidential elections voted rose from 40% to 45%. The effect operates ENTIRELY on the turnout margin, not registration: 88% of control compliers were already registered and charters moved registration by nothing. Effects appear only in presidential general elections; nonpresidential generals and primaries (base rates 14% and lower) show nothing. The effect persists into the second presidential election and CONVERGES by the third. Sharp gender split: young women +11 points, young men ~0 (difference p=0.038). Mechanism tests across five candidate channels (cognitive skills, civic skills, noncognitive skills, social networks, policy feedback) point to NONCOGNITIVE skills: subgroups with the biggest voting gains are the subgroups with the biggest noncognitive-index gains, while cognitive gains do not track voting. Directly relevant to this cluster: charters whose MISSION STATEMENTS emphasise civics did not produce significantly bigger voting effects than charters that do not (first-presidential 2SLS 0.075, SE 0.039 vs 0.034, SE 0.036; difference p=0.391) - although the civics-mission schools did move AP US History/Government taking sharply (+0.159, SE 0.025). Academic gains were large and confirmed (test scores, +8.5pp four-year college enrolment, +0.26 final years of education), but arrived AFTER the voting effect.

Low. Charter offers were randomly assigned within lottery risk sets, so allele distributions and family motivation are balanced in expectation. The authors additionally show that predicted voting (built from demographics and baseline test scores) is unaffected by the offer, which is a direct test that selection on voting propensity is not driving results.

Independently replicates the direction of Gill et al. 2020 (Democracy Prep, NYC) using a different city, a much larger lottery sample and a different voter-file vendor. The two lottery studies are the only randomized evidence that any schooling treatment moves VALIDATED turnout, and they agree. They disagree with the compulsory-schooling null literature (Pelkonen 2012, Lindgren et al. 2019, Berinsky and Lenz 2011), but not necessarily inconsistently: those test education QUANTITY and this tests school QUALITY at fixed quantity.