Association for Public Policy Analysis & Management

This paper builds a risk assessment model for the New York State Department of Health (DOH) as the agency seeks to determine who is acceptable to work, and who must be denied employment based on their criminal histories. The topic is of great importance as many employers currently attempt to conduct such assessments with little to no information and in many contexts, such as the one studied here, improved assessment can have large benefits for both the employer and potential employee. In this paper, we examine the three year arrest outcomes of individuals who were provisionally hired to work in the nursing home/home health care industry and were required to “pass” a state mandated criminal background check before beginning employment. We address the asymmetric costs of false negatives (individuals hired who pose real risks) and false positives (individuals not hired, but who pose no risk) and the inherent suppression effect of actually receiving clearance to work. We find that DOH is doing a good job of identifying individuals suitable for hire. We also find that our models can modestly improve predictive accuracy and allow for the clearance of an additional 10% of individuals for work without increasing the costs of making false negative predictions. We frame our findings in a more general policy discussion including which factors of the criminal record, such as time since last conviction and age at first conviction, are most relevant to risk assessment.