An Experiment in Background Check Adjudication

• In Event: Work, Crime, and the Consequences of Criminal Records: Causal Evidence on Labor Market Impacts

Abstract

Criminal background checks are a standard feature of hiring in the United States, yet little evidence exists on whether firms use background check data effectively to select candidates. We partner with a large online services marketplace that conducts criminal background checks on all providers seeking entry to the platform. In collaboration with the firm, we implement a randomized experiment that suppresses a set of non-violent criminal charges for randomly selected applicants, increasing platform access by 3.3 percentage points (5%) among applicants with criminal records. Cross-sectional comparisons show that workers with criminal records generate lower customer engagement and higher rates of adverse customer incidents. Yet, local average treatment effect estimates from the experiment show that randomly admitted marginal service providers perform no worse than average on either dimension, meaning that marginal applicants screened out by criminal records are not less productive or riskier than the average service provider admitted onto the platform. Because individual criminal charges are heterogeneous and offer limited statistical power for predicting adverse customer outcomes, fine-grained screening rules will have high rates of both false negatives and false positives at the margin. This pattern suggests that commonly used screening rules may over-exclude workers whose realized performance is similar to that of admitted workers. Our paper contributes to this literature in three ways. First, we provide experimental evidence on the consequences of relaxing criminal record screening thresholds, allowing us to isolate outcomes for marginally admitted workers rather than relying on comparisons among observationally selected workers. Second, we study a setting with rich measures of on-the-job performance, including engagement, productivity, and safety outcomes that are directly relevant to firm decision-making. Third, because the intervention operates through the firm’s existing adjudication rules, we are able to link specific features of criminal records, such as charge type and time since offense, to both hiring decisions and subsequent outcomes.

Author

• Amanda Yvonne Agan, Cornell University
  Presenting Author