‘Dead man working’: A Place-based approach to occupational safety and health
利用机器学习与意大利数据构建工作场所死亡风险地图，发现当前检查与补贴政策与高风险区域匹配不足，基于机器学习的精准干预每年可预防约70例死亡（占年死亡6%），产生显著社会效益。
Despite increasingly stringent regulations, there has been a concerning stagnation in reducing workplace fatalities. Can place-based policy targeting help? By coupling machine learning techniques with comprehensive data from Italy, we develop a place-based approach to workplace fatalities. Harnessing accurate machine forecasts, we construct a granular risk map and compare it to the allocation of on-site work inspections and public subsidies for occupational safety, uncovering limited overlap. Counterfactual estimates reveal that current public policies are effective only in areas flagged as high-risk by ex-ante machine predictions. A back-of-the-envelope exercise based on these estimates suggests that machine-learning-based targeting could prevent approximately 70 workplace fatalities per year, equivalent to around 6% of annual deaths, generating substantial social benefits. AI-powered territorial targeting can reduce the incidence and economic burden of this chronic issue while lowering the costs of policy implementation.