Racial Bias in Criminal Records

Abstract

Objectives: Government officials use criminal records as proxies for past conduct to decide who and how to investigate, arrest, charge, and punish. But those records may be racially biased measures of individual behavior. This paper develops a theoretical definition of bias in criminal records in terms of measurement error. It then seeks to provide empirical estimates of racial bias in official arrest records for a broad swath of offenses.

Method: I use official arrest and self-reported crime data from the Pathways to Desistance study to estimate Black-to-white and Hispanic-to-white crime ratios conditional on arrest. I also develop a novel, theory-based empirical test of differential reporting across racial and ethnic groups.

Results: Compared to white subjects with the same number of arrests, Black subjects committed 53, 30, 23, and 56% fewer property, violent, drug, and DUI offenses, respectively, and Hispanic subjects committed 19 and 46% fewer drug and DUI offenses. The analysis finds relatively little evidence of differential reporting that would bias my estimates upwards, with the possible exception of drug trafficking offenses.

Conclusion: The results provide evidence that Pathways subjects’ arrest records are racially biased measures of their past criminal behavior, which could bias decisions of criminal justice officials and risk assessment algorithms that are based on arrest records.

Keywords: criminal records, racial bias, algorithms, risk assessment

JEL Classification: K14