'Dead Man Working': A Place-based Approach to Workplace Fatalities

Abstract

This paper proposes a place-based approach for addressing workplace fatalities. We employ machine learning (ML) techniques and a comprehensive panel dataset from Italy to detect systematic patterns that pinpoint areas most at risk of onthe-job fatalities. The empirical analysis demonstrates that ML algorithms can accurately forecast, ex-ante, the number of workplace deaths out of sample. We then create a granular risk map based on the ML forecasts and compare it with the actual allocation of on-site work inspections and public subsidies to improve occupational safety, finding minimal overlap. This mismatch suggests that public effort is currently not more prevalent where it is most needed, at least with respect to onthe-job deaths. Lastly, we assess the impact of on-site inspections on the number of workplace deaths via double/debiased machine learning and show that inspections appear to be effective only in areas flagged as high risk by the ML forecasts. Overall, these findings suggest that by replacing current allocation rules with machine predictions, it would be possible to significantly improve the cost-effectiveness of public interventions and boost the deployment of deterrent and preventive measures aimed at enhancing occupational safety and health.

Keywords: machine learning, place-based policy targeting, workplace deaths

JEL Classification: I18, J28, R50