Determinants of safety outcomes in organizations: Exploring ONET data to predict occupational accident rates

Although workplace safety research is common given the frequent occurrence of fatal and nonfatal occupational accidents, it has focused mainly on safety climate and lacks a unified approach when examining predictors of safety outcomes. We argue that adopting an integrated approach with job analysis data and using newer machine learning methods can support and extend findings from cross-sectional research studies using traditional statistical methods. The suggested approach is demonstrated by using three machine learning methods (elastic net, random forest, and gradient boosting) along with publicly available O*NET data to predict annual nonfatal occupational incident rates published by the US Bureau of Labor Statistics. Findings indicate that O*NET descriptors from several subdomains including abilities, work context, and work activities were significant in predicting occupational injury rates. The amount of variance explained by the predictors varied from 54.2% (gradient boosting) to 58.8% (elastic net) with 12 common predictors across the three methods. The exploratory approach with machine learning techniques supports past findings and helps uncover understudied predictors of safety outcomes. This study adds to the literature surrounding person- and situation-based antecedents to workplace safety and has several other implications for research and practice.

Automated summary

This study demonstrates the use of machine learning methods and job analysis data to predict occupational incident rates. The findings highlight the significance of various predictors, supporting past research and expanding understanding of workplace safety outcomes.