Behavioral-Based Predictors of Workplace Violence in the Army STARRS.

摘要

Over the past year we used data from the Army STARRS Historical Administrative Data System (HADS), New Soldiers Study (NSS), All-Army Study (AAS), and Pre-Post Deployment Study (PPDS) to (i) estimate standardized rates of several types of workplace violence perpetration and victimization, (ii) examine bivariate associations of administrative and survey predictors with workplace violence outcomes, and (iii) develop risk prediction equations using data mining. Standardized rates of perpetration and victimization across the HADS, NSS, AAS, and PPDS suggested that male soldiers have higher rates perpetrating workplace violence, while female soldiers have higher rates of victimization. Most bivariate associations with our outcomes were consistent with expectations. The most noteworthy findings from the past year pertain to our development of risk prediction equations. We used data mining methods to maximize the prediction of 15 perpetration and victimization outcomes in the HADS and NSS. Overall, these models performed quite well. Prediction accuracy was much better than would be expected by chance (AUCs = 0.7-0.9), and soldiers who were predicted to be high risk accounted for a large proportion of all workplace violence perpetrations and victimizations (18.4-72.9% of all first occurrences of perpetration/victimization occurred among soldiers in the top 5% of predicted risk).