← Studies Suggest ⚖️ Policy

Banning the Criminal-Record Box Was Supposed to Fight Racism. A 15,000-Application Experiment Found It Made the Racial Gap Six Times Worse

Economists Amanda Agan and Sonja Starr sent fictitious job applications to employers in New Jersey and New York City before and after ban-the-box laws took effect. At companies that removed the criminal-history question, the Black-white callback gap grew from 7% to 43%, as hiring managers substituted race-based guesswork for the missing information.

By Marcus Reeves, Policy & Economics · September 25, 2026

For decades, the American job application opened with one brutal question: have you ever been convicted of a crime? Check yes, and your application died before a human read the rest of it. The box fell hardest on Black men, incarcerated at nearly six times the rate of white men, so reformers settled on an elegant fix: ban the box, delay the question until later in hiring, and give every applicant a foot in the door. The logic felt airtight: employers cannot discriminate on information they do not have.

Amanda Agan and Sonja Starr, then at Rutgers and the University of Michigan, tested that logic with an experiment: roughly 15,000 fictitious online job applications went to employers in New Jersey and New York City, timed to ban-the-box adoption in both places, each describing a young man seeking a low-skill job; two details were randomized, a distinctively Black or distinctively white name and whether the résumé disclosed a felony conviction.

The design is what makes the result hard to dismiss: a randomized audit study layered over a natural policy experiment. Employers that asked about criminal history and were forced by law to stop formed the treatment group, while employers that never asked formed the control group, and the control group's hiring patterns barely moved, so whatever changed at the treated firms, the law caused it.

The racial gap in callbacks changed violently. Before ban-the-box, white applicants at box-using employers received about 7% more callbacks than equally qualified Black applicants; after the law forced those employers to drop the question, the gap ballooned to 43%, six times wider than before, while the control group showed no such jump.

The mechanism has a name in economics: statistical discrimination, judging individuals by group averages when individual data is missing. That is what some hiring managers did once denied criminal-history data, substituting assumptions for evidence. The authors' interpretation is blunt: employers relied on exaggerated impressions of real racial differences in conviction rates, with Black applicants without records paying the price by losing the one cheap way to signal a clean history. The paper's own data show white ex-offenders captured the largest callback gains.

For scale, compare the most famous résumé audit ever run. In 2004, Marianne Bertrand and Sendhil Mullainathan sent fictitious résumés to employers in Boston and Chicago and found white-sounding names received 50% more callbacks than Black-sounding names. Agan and Starr's experiment suggests banning the box recreated 43 of those 50 points through information removal alone, meaning a policy built to shrink racial disparity rebuilt, nearly in full, the gap the landmark study measured as pure bias, though in different cities and a different decade, so treat this as a benchmark rather than an equation.

The stakes are not abstract: thirty-seven states, the District of Columbia, and more than 150 cities and counties have adopted some version, though only 15 states extend the rule to private employers. The follow-up literature keeps the debate honest, with Doleac and Hansen's 2020 employment analysis finding the policy cut employment 3.4 percentage points for young, low-skilled Black men, while Seattle's administrative data, public-sector hiring analyses, and later-year estimates find smaller, mixed, or null effects on actual employment. The callback gap is solid, but whether it becomes fewer jobs is still contested.

The Strongest Counterargument

The fiercest critics do not dispute the arithmetic; they dispute what it means. The National Employment Law Project's Beth Avery and Maurice Emsellem argue these studies reveal entrenched racism in hiring, not a failed policy: scrapping ban-the-box restores a world where every applicant with a record is screened out at step one. The box was a proven barrier, and Agan and Starr's own data confirm it: employers that asked about records were 63% more likely to call back applicants with clean records, a penalty so severe that checking "yes" ended a candidacy before qualifications were weighed. Banning the box demonstrably helped people with records get callbacks, and even the authors refused to call for repeal: Starr said the results do not definitively argue against the policy, since the benefits for people with records are real and may justify the cost.

Then there is the measurement objection, and it has teeth: Agan and Starr measured callbacks, not hires. Plenty happens between application and offer, not least because employers learn an applicant's race eventually whatever the form asks; some researchers argue statistical discrimination simply migrates to later hiring stages rather than being created by the policy, and the employment studies refuse to speak with one voice, showing negative effects for young Black men in some analyses, no detectable effect in Seattle, positive effects in public-sector hiring, and shrinking effects over time. If the callback gap were destiny, the employment data would agree, and it does not.

The deepest objection is moral, not empirical: even if every number holds, the policy question is distributional, trading opportunities for Black applicants without records against opportunities for applicants with records, a disproportionately Black group previously shut out almost entirely. Choosing between them is a values judgment no experiment can make for you.

What We Didn't Prove

This was a correspondence study: it measured callbacks, not hires, and cannot say how many lost callbacks became lost jobs. It covered two labor markets and only young men seeking low-skill work, and the 43% is a relative gap among affected employers rather than a national average. It cannot weigh the policy's benefit to ex-offenders against its cost to Black non-offenders, because that comparison needs a value judgment no dataset contains; the follow-up literature is genuinely mixed, so the mechanism is better established than its consequences.

The Bottom Line

Two things are true at once, though the debate keeps demanding you pick one. The criminal-history box was a real barrier that crushed job prospects for people with records, and removing it helped them get callbacks, but information has a price: when the law took criminal-history data away, hiring managers replaced it with race-based guesswork, and the racial callback gap grew sixfold. The policy did not fail because its goal was wrong; it failed, in part, because it assumed prejudice needs information to operate. Sometimes prejudice only needs the absence of it.

What You Can Do

If you hire, audit your own funnel: compare callback rates by name-inferred race before and after any application change, and you will know quickly whether your process carries the same disease. Consider pairing a delayed record check with name-blind initial screening, the authors' own suggested fix, since removing the proxy when you remove the information keeps one reform from recreating the problem the other was meant to solve. Check whether your state offers certificates of rehabilitation or employability: an audit study found applicants holding certificates were called back at nearly the same rate as applicants with clean records.

If you make policy, do not assume your city's results match New Jersey's. The mixed follow-up literature is an instruction to measure, not a permission to ignore: pair fair-chance rules with enforcement and outcome tracking, which is the National Employment Law Project's position, because the answer to statistical discrimination is stronger anti-discrimination enforcement alongside the box ban, not the ban alone.

If you carry a record and are job hunting, know your local fair-chance law. Thirty-seven states plus the District of Columbia restrict when employers can ask, and in many places an expungement or certificate changes the legal calculus entirely.