Does OSHA Enforcement Actually Work? What the Randomized Evidence Shows

For fifty years, the argument over OSHA has run on ideology: one side calls inspections essential, the other calls them job-killing red tape. In 2012, three researchers did something the debate rarely allows — they ran the experiment. The result was decisive. Randomly inspected workplaces had 9.4% fewer injuries and 26% lower injury costs than otherwise-identical uninspected ones, with no detectable loss of jobs, sales, or business survival. That is not a correlation or an advocacy estimate; it is causal evidence from a randomized controlled trial, the same design used to test medicines. This analysis walks through what the evidence actually shows, what a follow-up study found about doing it better, and why — with enforcement capacity fixed at roughly one visit per workplace every two centuries — the entire game is now targeting.

At a glance: In a randomized trial (Levine, Toffel & Johnson, Science, 2012), OSHA-inspected workplaces had 9.4% fewer injuries and 26% lower injury costs — with no detectable loss of employment, sales, or survival. A 2023 follow-up: each inspection prevented about 2.4 serious injuries over five years, and better targeting could avert up to 2× as many — ~16,000 fewer injured workers and up to ~$1B in social value over a decade. A randomized experiment published in Science in 2012 found that OSHA inspections reduced workplace injuries by 9.4 percent and injury costs by 26 percent, with no detectable loss of jobs or sales.

The experiment that ended the argument

Almost every claim about OSHA's effect suffers from the same flaw: the workplaces OSHA inspects are not like the ones it doesn't. Inspected firms may be more dangerous to begin with, or more troubled, so any simple comparison confuses the effect of the inspection with the reason for it. The only clean way around this is randomization — and California's high-hazard inspection program provided it, selecting eligible workplaces for inspection essentially at random.

Levine, Toffel, and Johnson compared those randomly inspected establishments with the eligible-but-not-inspected ones over the following years. Because the two groups were statistically identical at the start, any later difference is caused by the inspection. The results follow.

The two numbers on the left are the benefit; the number on the right is the rebuttal. The persistent political claim is that safety enforcement destroys jobs. The researchers went looking for exactly that harm — in employment, sales, credit ratings, and firm survival — and could not find it. Inspections made workplaces safer and cheaper to run, and the businesses were no worse off. On the evidence, the trade-off that dominates the debate largely does not exist.

What one inspection buys

A 2023 follow-up in the American Economic Journal: Applied Economics extended the picture from "does it work" to "how much." Studying OSHA's own randomized Site-Specific Targeting inspections, the authors estimated the yield of a single visit.

The second and third lines are the important ones. They say the binding constraint on OSHA's impact is not whether inspections work — that is settled — but which workplaces get the limited supply of them. Using machine learning to predict where injuries are most preventable, the researchers showed OSHA could roughly double its injury-prevention yield with the resources it already has. Enforcement effectiveness, in other words, is now an information problem more than an authority problem.

The capacity bind: effective, but rare

To see why targeting matters so much, put the effectiveness next to the reach. Federal OSHA and the State Plans together field 1,651 compliance officers for about 12 million workplaces, and federal OSHA's own staffing works out to one inspection of each workplace in its jurisdiction about once every 191 years.

This reframes what "enforcement" even means. With visits this rare, OSHA cannot police its way to safety establishment by establishment; its reach comes from deterrence and from concentrating inspections where the expected injuries are highest. The research validates both halves of that strategy — the visits change behavior, and smarter targeting multiplies the change — and explains why OSHA has moved toward data-driven Site-Specific Targeting, built on the electronic injury data of 29 CFR 1904.41, rather than random coverage. (For the reach side of this story, see OSHA's inspection capacity: once every 191 years.)

The levers an inspection pulls

The deterrence the researchers measured runs through a handful of rules. An inspection checks the injury records an employer keeps (1904.7(a), 1904.29(b)(1)) and the hazards those records point to, and every step below is written down:

What the evidence means for an employer

It would be a mistake to read "inspections are rare" as "the odds are in my favor." The same research that proves inspections work also explains why the smart posture is to behave as if one is coming. In records terms that means four habits: log each recordable case within seven calendar days (1904.29(b)(3)), post the annual summary from February 1 to April 30 (1904.32(b)(6)), report severe injuries on time (1904.39(a)), and certify cited hazards as abated (1903.19(c)).

• The controls pay for themselves. The abatement an inspection forces — guarding, fall protection, training, documentation — is what produced the 9.4% injury drop and the 26% cost drop. Those savings accrue whether or not an inspector ever arrives, which is the real reason to invest in them. (See what a workplace injury really costs.)
• Targeting cuts both ways. OSHA increasingly selects by risk signals — injury history, industry, and prior citations. A workplace with a clean, documented program is both safer and less likely to surface in that selection.
• Be ready before the knock. The record is the defense. Keep the program scored, the audit current, and the hazard analysis done, so an inspection finds what it is supposed to find.

Inspections cut injuries; the open question is where OSHA sends them

The question "does OSHA enforcement work" has an answer, and it is yes — established not by advocacy but by a randomized experiment that found 9.4% fewer injuries, 26% lower costs, and no job loss, then confirmed and quantified by a follow-up showing each inspection prevents about 2.4 serious injuries and that better targeting could double the benefit. The open question is no longer effectiveness; it is allocation. OSHA reaches any given workplace so rarely that its impact rides almost entirely on choosing the right ones — and on employers deciding, correctly, that the controls an inspection would demand are worth putting in place before it ever arrives.

Related: Once Every 191 Years: OSHA's Inspection Capacity · Anatomy of an OSHA Construction Inspection · What a Workplace Injury Really Costs · How OSHA Calculates the Penalty · OSHA Data & Research hub

Sources & verification (verified 2026-07-13): Randomized-trial findings from David I. Levine, Michael W. Toffel, and Matthew S. Johnson, "Randomized Government Safety Inspections Reduce Worker Injuries with No Detectable Job Loss," Science 336, no. 6083 (2012): 907–911 — inspected California workplaces had 9.4% fewer injuries and 26% lower injury costs, with no statistically detectable difference in employment, sales, credit ratings, or firm survival. Per-inspection yield and targeting estimates from Matthew S. Johnson, David I. Levine, and Michael W. Toffel, "Improving Regulatory Effectiveness through Better Targeting: Evidence from OSHA," American Economic Journal: Applied Economics 15, no. 4 (2023): 30–67 — about 2.4 fewer serious injuries (~9%) per inspected establishment over five years, with better (machine-learning) targeting able to avert up to roughly twice as many injuries, on the order of 16,000 fewer injured workers and up to ~$1 billion in social value over the decade examined. Inspection-frequency framing ("once every 191 years", federal OSHA) and the 1,651 inspector count from the AFL-CIO Death on the Job report, 2026 edition. The studies concern the causal effect of inspections; individual results vary by industry and program. Not legal advice.