Association for Public Policy Analysis & Management

Inspections are the primary tool government agencies use to assess how well companies are complying with regulations. Budget constraints typically require agencies to choose a small subset of the companies under their jurisdiction to actually inspect. But how can we know whether agencies’ prevailing targeting decisions are maximizing the effectiveness of inspections? We investigate this question in the context of a large-scale targeted inspection program of the U.S. Occupational Safety and Health Administration (OSHA), which from 2001 to 2010 prioritized for inspection establishments that had recently experienced high injury rates. Leveraging a portion of inspections that were randomly assigned, we find inspections, on average, led to 9 percent fewer serious occupational injuries and illnesses over the following five years. To examine the potential for improved targeting, we employ a machine learning algorithm called Targeted Maximum Likelihood Estimation to estimate heterogeneous treatment effects of inspections. We use these estimates to conduct policy counterfactuals that prioritize inspections to facilities with the largest predicted benefit. We find OSHA could avert 66 percent more injuries at inspected establishments, without increasing costs or sacrificing general deterrence, which would have created a social value of roughly $1.2 billion. We outline how our approach can generalize to assess the extent to which other regulatory agencies are maximizing the benefits of their limited inspection resources.