Checked 18 August 2026
We checked
Claim: The Donohue–Levitt hypothesis about abortion and crime stands up to current scrutiny
No clear yes: the hypothesis remains genuinely contested, not settled in its favor. Donohue and Levitt's own updated data still show the correlation, but methodological critiques (coding errors, arrest-rate corrections, international crime-decline patterns) have never been resolved, and a DOJ-funded review and economists as recently as 2024 say the most rigorous tests don't support it — this is an active dispute, not a case of individual claims failing while the core idea holds.
Claim: The Donohue–Levitt hypothesis about abortion and crime stands up to current scrutiny
Open questionDonohue and Levitt's 2019/2020 retrospective, using 17 additional years of data, reported that the abortion-crime coefficient was as strong or stronger than in their original 2001 paper, and some independent replications (e.g., a Bayesian tree-ensemble reanalysis) found support for the pattern. But credible economists have continued to contest the causal interpretation on methodological grounds since then: Foote and Goetz's peer-reviewed comment found the effect vanished once a coding error was fixed and arrest rates rather than raw totals were used, a U.S. Department of Justice-funded review concluded the most credible tests do not support the hypothesis, and criminologists were still publicly disputing the claim as recently as 2024, citing confounders like the international timing of crime declines that don't track U.S. abortion legalization.
The core problem isn't a lack of data, it's identification: abortion legalization happened at a specific historical moment alongside other things that plausibly cut crime in the 1990s (the end of the crack epidemic, more aggressive policing and incarceration, and the phase-out of leaded gasoline, which affected the same birth cohorts). Any statistical model has to somehow separate the abortion effect from all of that, and reasonable researchers using different specifications, controls, and time windows get different answers, including some that flip sign or vanish. That's the signature of a question where the natural experiment is too entangled with confounders to be cleanly settled by more regression work.
The strongest case for the hypothesis is that it isn't just an American 1990s story: similar cohort-timing patterns show up in some other countries after their own abortion legalizations, which is harder to explain away as coincidence with US-specific factors like lead or policing. That doesn't resolve the identification problem, but it's the real reason the hypothesis hasn't been abandoned outright.
The hypothesis originates in a May 2001 Quarterly Journal of Economics paper by John Donohue and Steven Levitt, "The Impact of Legalized Abortion on Crime," which found legalized abortion appeared to account for up to half of the drop in violent and property crime and predicted further declines. The claim spread rapidly beyond academia after being popularized in Levitt's 2005 book Freakonomics, and it resurfaced in wider public debate whenever abortion law was politically live, including after the 2022 leak of the draft opinion overturning Roe v. Wade. Its academic life since has been a back-and-forth of challenges, a real coding error, and a 2019/2020 update by the original authors reasserting the finding.
ReferencesReviewed 17 September 2026 — nothing had changed.
Claims on this page are re-checked periodically; any change of finding is reviewed by a person before it appears here.