Introduction

Many regulatory agencies can only inspect a small fraction of the entities they oversee, making the social value of inspections dependent on whether they affect outcomes at the many more entities that remain uninspected. Such spillovers may occur if inspections alter managers’ beliefs about inspection risk, disseminate knowledge of best practices, or prompt broader organisational responses. Identifying these spillovers is difficult because regulators target inspections where problems are expected, and nearby workplaces often experience common shocks. This research studies spillovers from inspections conducted by the US Occupational Safety and Health Administration (OSHA), exploiting quasi-random variation generated by the Site-Specific Targeting (SST) programme.

Institutional setting, background, and data

The SST programme operated from 1999 to 2014 in 29 states and targeted non-construction establishments in hazardous industries with high injury rates. It was OSHA’s largest inspection programme, accounting for more than 40 per cent of programmed inspections outside construction. OSHA used the OSHA Data Initiative (ODI) to collect annual injury data. Each year, roughly 3,500 establishments with the highest injury rates were placed on a primary inspection list. In 2004, for example, establishments with a DART rate of at least 15 or a DAFW rate above 10 were listed, well above the private sector averages of 2.8 and 1.6 respectively. Across 2001–2010, 32 per cent of establishments on the primary list were inspected before the following year’s list was created.

Workplace safety is measured by the annual number of days-away-from-work (DAFW) injuries. In 2005, 1.2 million such injuries occurred in the US private sector, causing an economic loss of over $80 billion (in 2026 dollars). DAFW injuries are more likely to be reported accurately than less serious injuries.

Estimating spillover effects of inspections

The researchers model spillover effects as a function of the share of an establishment’s peer group that received an SST inspection during the prior four years. To overcome endogeneity, they use the discontinuous increase in inspection probability for establishments just above the primary list cutoff. Having an injury rate just above the cutoff increased the probability of being listed by 97 percentage points and increased the probability of inspection by roughly 22 percentage points. The study uses a control-function approach with Poisson quasi-maximum likelihood to account for the count nature of injury data and high-dimensional fixed effects for establishment, region-year, and industry-year.

Spillover effects of OSHA’s SST inspections

The study finds large spillover effects on injuries among nearby establishments. One additional SST inspection averts an estimated 3.0 annual DAFW injuries across other establishments in the same ZIP code over the subsequent four years, representing an 11 per cent reduction. Because an inspection affects multiple nearby establishments, the aggregate spillover effect is several times larger than the direct effect of 0.48 injuries at the inspected establishment itself. Specifically, injuries averted due to spillovers are roughly six times larger than the direct effects.

Broadening the geographic scope to the county level, an additional SST inspection led to 4.3 fewer annual DAFW injuries across other workplaces. However, the results suggest that spillovers decay with geographic distance and are largely contained within ZIP codes. A naïve regression yields a coefficient of essentially zero, suggesting that traditional evaluations that do not account for endogeneity dramatically understate the social returns to enforcement.

Placebo, falsification, exclusion restriction, and robustness checks

Several tests support a causal interpretation. A falsification test shows that future peer inspections have no relationship with current injuries. Placebo cutoffs yield statistically insignificant effects. Tests of the exclusion restriction show that the instrument does not predict inspections in non-SST programmes or inspections of the focal establishment itself. The results are robust to alternative bandwidths, sample restrictions, and the inclusion of highly populous counties. Additionally, tests weigh against the concern that establishments merely manipulated reporting; total days away fell more than DAFW cases, which is consistent with real safety improvements rather than reporting changes.

Conclusion

Using quasi-random variation, the research demonstrates that regulatory inspections generate substantial spillover effects at nearby establishments. Inspections reduced injuries by roughly 11 per cent among establishments in the same ZIP code. Conventional evaluations focusing solely on directly inspected entities may substantially understate the social returns to enforcement. These findings highlight the importance of considering general deterrence when allocating scarce regulatory resources.