Abstract

Excerpted From: Jessica Saunders and Greg Midgette, A Test for Implicit Bias in Discretionary Criminal Justice Decisions, 47 Law and Human Behavior 217(February 2023)(2 Footnotes/References)(Full Document)

There is no denying the existence of stark racial and ethnic disparities in the criminal justice system, but the reasons driving those disparities are still heavily debated. One mechanism that has received a great deal of attention is the influence of implicit, or unconscious, biases on the hundreds of decisions that criminal justice system actors make across the course of any criminal case, from decisions to stop and search an individual because they look suspicious or match the description of a suspect to decisions to revoke community supervision and grant parole. Researchers have been called on to study this topic, particularly to help understand whether and how implicit bias is contributing to racial inequalityand what can be done to reverse it.

Understanding and quantifying the unique contribution that implicit biases make in the observed racial disparities in incarceration is fraught with methodological challenges, but this article describes a unique approach to disentangling the marginal impact that implicit bias has on discretionary decisions within community supervision. This procedure allows the identification of the additional bias associated with increased disambiguation and discretion, but it does not provide an answer about the absolute size of bias nor is it able to isolate the contribution of explicit bias to our observed outcomes. Importantly, the setting in which we developed the method to identify implicit bias offered a unique set of characteristics that enabled us to validate our findings using a strong quasi-experimental design that will likely not be available in all settings. However, given that the results in the main model and model validation were consistent, we believe that this testing strategy could be generalized to other activities that involve varying levels of discretion by criminal justice system actors, particularly in situations in which there is limited interaction between the community member and justice system actor, including traffic stops, prosecutorial charging, and sentencing.

In this study, we used two complementary quasi-experimental methods to look for evidence of implicit bias in community supervision discretionary decision-making using a novel testing strategy. This procedure uses exogenous variation on two dimensions to isolate the impact of reduced information on racial and gender disparities in the experience of technical violations--defined generally as violations for failure to appear for a scheduled appointment with a court or corrections officer, failure to report information deemed pertinent to supervision, and failure to pay fines or fees on time--relative to new criminal offenses, which are more likely to be recorded by officers as violations. First, we used plausibly random assignment of clients to supervision officers to identify the effect of supervision intensity on the relative rate of high discretionary technical violations compared with low discretionary violations associated with criminal charges. Second, we used a regression discontinuity(RD) design based on the supervision-level thresholds in clients' risk assessment scores.

We begin with a high-level overview of justice system trends and what drives disparities. Although there is a strong consensus that racial disparities are due to a variety of factors, this article will focus on the role of biased decision-making and how we can test for its unique contribution to the problem. Therefore, the next section discusses how decisions within the community supervision context are vulnerable to bias. Next, we describe our study, including how and why we believe the community supervision context enabled us to develop a framework to test for implicit bias that can be validated using quasi-experimental methods. We then describe the testing procedure and provide data to establish the validity of the method, discuss how systems can be put in place to identify and reduce bias, and finally describe how this approach could be generalized to other decision points.

[. . .]

This article offers a unique testing procedure that incorporates concepts from social psychology and applied work on identifying racial bias within the criminal justice system while controlling for common threats to validity such as differential base rates and other potentially influential variables. Taking advantage of the natural variation in both discretion and the amount of information a decision-maker has, we demonstrated that bias influences decisions when there is less interaction between the supervision officer and their client, which is consistent with decades of social science lab experiments mimicking similar information ambiguity. In our study, implicit bias influenced decisions when community supervision officers spent less time with their clients, and its impact was larger for women than men. This is particularly troubling because the conditions in which implicit biases are more influential characterize most criminal justice system interactions. Because we established the validity of this approach using RD, a strong quasi-experimental method, we believe that this testing framework can also be applied to other activities that involve discretion by criminal justice system actors, such as traffic stops, prosecution decisions, and sentencing. This angle reveals a different policy lever--alleviating disparities might be best achieved by introducing policies that reduce ambiguity, limit discretion, and promote oversight and accountability.

_________

Stephane Shepherd served as Action Editor.

Jessica Saunders(iD) https://orcid.org/0000-0001-7611-9559

Greg Midgette(iD) https://orcid.org/0000-0003-2038-2950