A Bayesian study of safety conditions at United States navy shipyards: exploration of statistical models for the severity of injuries and safety deficiencies

This study quantitatively examines how personal and environmental factors affect workplace safety, through the Markov chain Monte Carlo simulation of two Bayesian statistical models, one for United States Navy shipyard injuries, and one for safety deficiencies. Each shipyard used two separate databases to track these incidents. The shipyards each maintained their own independent databases while applying a similar records structure. The study models injury severity using a bivariate numeric response level where the responses were days of labor lost by the workplace, and days spent away from the workplace in recovery by the injured worker. The response for the safety deficiency model is the severity rating of one, two, or three, where one means the deficiency poses a serious safety risk, and three means the deficiency poses only a minor risk. Both models operated successfully during simulations, making them candidates for future improvements to refine them further. Both models’ simulation outputs contain unexpected parameter estimates that undermine confidence in the other results. The provided data lacks internal consistency since the variables recorded are not entirely consistent between shipyards, and the shipyards do not use standardized data naming conventions. The data scarcity in the records used also suggests that using data from a wider period of time will also improve the models’ performances. The models provide strong results given the limited information imposed by privacy, security, and contracting concerns associated with military data. Despite these restrictions, the models provide important insights into the underlying system behavior. The results strongly suggest that these models have the potential to accurately represent workplace safety conditions, even for an environment as complex as that of an industrial Navy shipyard.