Reducing Racial Disparities in Bail Decisions - Pilot

Last registered on January 22, 2020


Trial Information

General Information

Reducing Racial Disparities in Bail Decisions - Pilot
Initial registration date
January 21, 2020

Initial registration date is when the trial was registered.

It corresponds to when the registration was submitted to the Registry to be reviewed for publication.

First published
January 22, 2020, 11:13 AM EST

First published corresponds to when the trial was first made public on the Registry after being reviewed.


There is information in this trial unavailable to the public. Use the button below to request access.

Request Information

Primary Investigator


Other Primary Investigator(s)

PI Affiliation
Harvard Law School
PI Affiliation
Harvard Kennedy School

Additional Trial Information

In development
Start date
End date
Secondary IDs
Racial disparities exist at every stage of the U.S. criminal justice system and are particularly prominent in the setting of bail. In this project, we are collaborating with courts from around the country to test the effectiveness of an intervention that can reduce racial disparities in bail decisions. We hypothesize that these disparities exist because of rushed judicial decision-making and a reliance on heuristics as well as biased beliefs about the relative risk of defendants based on race. Our intervention consists of three components. The first component provides judges with objective information on pretrial risk of white and nonwhite defendants. The second slows down and systematizes judicial decision-making to reduce reliance on inaccurate stereotypes that exaggerate the relative danger of black defendants. The third component provides detailed feedback to judges on their own outcomes over time, giving them the motivation, information, and tools necessary to reduce racial disparities in their pretrial decisions. We will estimate the causal effect of our interventions on pretrial release and misconduct rates (in the aggregate and racial gaps).

The pilot phase of this project includes seeking feedback about the clarity of our materials, ensuring feasibility and smooth roll-out in the court setting, and administering a survey to assess changes in pre- and post-intervention perceived risk of white and non-white defendants, as well as trade-offs between release and pretrial misconduct.
External Link(s)

Registration Citation

Dobbie, Will, Will Dobbie and Crystal Yang. 2020. "Reducing Racial Disparities in Bail Decisions - Pilot." AEA RCT Registry. January 22.
Experimental Details


Our intervention has three components intended to improve decision-making and reduce racial disparities at the pretrial justice stage.

(1) Objective Information on Misconduct Rates by Race. The first part of our intervention will provide information to judges on the actual risk of pretrial misconduct based on data by race. We aim to correct mistaken beliefs which can arise if judges erroneously assume that black defendants are riskier relative to white defendants than supported by data. We will create an informational handout that provides descriptive statistics of actual risk vs. "perceived" risk by race using historical data.

(2) "Benchard" checklist: The second component of our intervention will involve providing judges with a “benchcard” checklist that will prompt them to consider different areas important to the bail decision-making process.

(3) Individualized Feedback on Release and Misconduct Rates by Race: The third component of our intervention will provide judges with individualized feedback on their past performance, including their own release and pretrial misconduct rates. We will start by providing historical information but will also provide updated information on a regular basis. This therefore gives judges a chance to see if their changed behavior maps to improved outcomes.
Intervention Start Date
Intervention End Date

Primary Outcomes

Primary Outcomes (end points)
Average pre-trial release and misconduct rates, racial gaps in pre-trial release and misconduct rates
Primary Outcomes (explanation)

Secondary Outcomes

Secondary Outcomes (end points)
Secondary Outcomes (explanation)

Experimental Design

Experimental Design
We will conduct a randomized controlled trial (RCT) to identify the causal impact of our intervention on pretrial release and misconduct rates (on average and on racial disparities). We will randomize exactly when each judge receives the interventions in order to identify the causal impact of each treatment. After the conclusion of our planning period, our sample will be split into a treatment and control group.

Data Collection: First, we will collect survey data both during the planning and intervention stages. The survey will ask judges for baseline demographic information, questions about which factors they deem important to pretrial decision-making, questions assessing their current bail practices and perceived risk of defendants and will ask judges to make decisions for prototypical defendants in sample vignettes. We will also administer a baseline Implicit Association Test (IAT) to all judges, as the IAT has shown to be highly correlated with measures of bias and stereotyping behavior in the real world (Glover, Pallais, Pariente 2017, Carlana 2018). This survey will be repeated at the end of the study, where we will also ask additional questions in order to better understand how judges interpreted and utilized the intervention they received.

Second, both during and after the intervention, we will work with courts to obtain data on pretrial release decisions and misconduct outcomes at the case level. These data will also include judge and defendant identifiers, defendant demographics, and additional case and defendant details such as the crime type and prior criminal history.

Third, to assess whether the benchcard intervention slows down decision-making, we will send research assistants to each court to observe a random sample of bail hearings and record the number of minutes each hearing takes prior to a determination of bail conditions. We will also record the number of questions the bail judge asks at each hearing/arraignment.
Experimental Design Details
Not available
Randomization Method
Randomization done in office by a computer
Randomization Unit
Individual judge
Was the treatment clustered?

Experiment Characteristics

Sample size: planned number of clusters
Preliminary pilot will include 8-12 judges
Sample size: planned number of observations
Preliminary pilot will include 8-12 judges
Sample size (or number of clusters) by treatment arms
Preliminary pilot will include 4-6 judges in the treatment group with 4-6 in the control group
Minimum detectable effect size for main outcomes (accounting for sample design and clustering)

Institutional Review Boards (IRBs)

IRB Name
Harvard University Area Committee on the Use of Human Subjects
IRB Approval Date
IRB Approval Number