Association for Public Policy Analysis & Management

The use of criminal background checks in employment decisions has skyrocketed in recent decades.  While employers are concerned about potential harm to their production and operations, advocates of ex-offenders claim that such background checks actually hinder the process of desistance by discouraging employment.  However, much of the discussion in this area is conjecture, since researchers have not been able to study the direct impact of a criminal background check on either employment or subsequent crime (EEOC 2012).  Recently, the NY State Department of Health has cooperated with an evaluation of their background screening process for the approximately 80,000 people a year who have, for the first time, been offered provisional employment to work as non-licensed direct care providers in health care facilities or for home health care firms.  In the first stage of this project, research using a fixed effect design has shown that the decision to clear an individual with a record leads causally to sizeable increases in employment and earnings over the following three years (Bushway et al. 2015).

This paper takes the next step by performing a recidivism analysis on the same sample of individuals with a conviction record who face the DOH background check.  There appears to be no plausible exogenous variation in who obtains the clearance decision, but using detailed institutional knowledge about the DOH’s background screening process, we build structural assumptions on the potential outcomes for different sub-samples in our data.  Based on these assumptions, we are able to partially identify the average treatment effects for our sample, controlling for all observable information. Using a denial in the background check as the treatment, our estimates indicate that on an average, being denied to work based on a criminal background increases the likelihood of a re-arrest by 1.3 to 2.2 percentage points in the following three years.  We also perform an analysis on short term outcomes, estimating the impact on arrests one year post DOH's decision.  Although the upper bound on the average treatment effects is small in the short term, we do find heterogeneous upper bounds, suggesting that much of the average effect is coming from one sub-sample of the population.  Policy implications of these results are discussed.