The Effect of Grade Retention on Adult Crime: Evidence from a Test-Based Promotion Policy

14,728 students in the 30-point bandwidth; eighth graders 1998-99 to 2000-01, birth cohorts 1984-87

Louisiana eighth graders who failed a March LEAP exam, linked to the universe of Louisiana adult criminal convictions to age 25

Journal of Labor Economics 40(2), 361-395; full text read from NBER working paper 25384 (December 2018). Fuzzy RD using the July LEAP cutoffs as an instrument for eighth-grade retention, with outcomes drawn from state adult conviction records rather than self-report. The paper's most useful methodological contribution for this archive is that it does NOT stop at the cutoff: using Angrist-Rokkanen extrapolation it estimates effects away from the threshold, which is the standard objection to every RD in this literature.

Eighth-grade retention raises adult VIOLENT crime specifically. Conviction for a violent crime by age 25 rises 1.05 percentage points (SE 0.516), a 58.4% increase on the base rate; the number of convictions rises 0.025 (18.4%), 'primarily of a 0.017 (85.00%) violent crime increase'. Any crime rises 1.25 points (11%) but is not statistically significant, and property and drug crime are flat. The channel is educational, not directly criminal: dropout rises 7.17 percentage points (SE 2.33), absences three years later rise 11%, and disciplinary incidents rise at least 1.1 points, while JUVENILE conviction and juvenile felony are both null - the crime shows up in adulthood, after the dropout. The extrapolation result is the one policymakers should read: 'we find positive and statistically significant effects away from the cutoff for violent, property and drug-related crime... our regression discontinuity estimates UNDERSTATE the effects... one cannot simply move the passing threshold to eliminate the crime effect.'

Low at the cutoff and explicitly probed beyond it. Assignment is a statewide test threshold applied to students who have already failed once, so compliers on either side share the same measured failure. Criminal conviction is heritable-adjacent through impulsivity and conduct problems, which is exactly why the design matters: the comparison holds those traits fixed in expectation. The extrapolation to non-compliers relies on a conditional-independence assumption rather than the discontinuity, and is correspondingly weaker.

No independent replication of the adult-crime result exists; it is the only study in the literature linking retention to adult convictions. It is consistent in direction with the same team's dropout findings and with New York City's grade-7-and-8 attainment harms. It runs directly against Zhang and Huang's Chinese evidence that ABOLISHING retention raised adult property crime and rape - but that reform hit children in grades 1-3, so the two results together say the same thing the rest of this topic says: the grade at which it happens determines the sign.