Predicting Recidivism: Continuing To Improve the Bureau of Prisons’ Risk Assessment Tool, PATTERN

With bipartisan support, the First Step Act (FSA) was signed into law in December 2018.[1] The principal goal of the FSA’s prison reforms was to reduce recidivism by directing programs and other services to people in the custody of the Federal Bureau of Prisons (BOP). Title I of the FSA mandated the development and implementation of a risk and needs assessment system for people in federal custody. In response, development consultants for the National Institute of Justice (NIJ) created a risk assessment tool, the Prisoner Assessment Tool Targeting Estimated Risk and Needs (PATTERN), to assess individuals’ risk of engaging in crime once they are released from federal custody.

After its initial development in July 2019, PATTERN was subjected to public review and comment. Based on participant input and suggestions, several revisions were introduced, including the removal of two items believed to exacerbate racial disparities. The revised version of the tool, referred to as PATTERN 1.2, was implemented in January 2020.[2]

Title I of the FSA also requires the Department of Justice (DOJ) to review, validate, and release publicly its risk and needs assessment system on an annual basis. In January 2021, NIJ’s review and revalidation expert consultants reported that they had discovered discrepancies with some of the measures used to create PATTERN 1.2. NIJ’s expert consultants collaborated with staff from BOP’s Office of Research and Evaluation to correct these issues.

In December 2021, NIJ released the annual review and revalidation report which documented the coding and specification errors detected in version 1.2 and described the development of a revised version 1.3.[3] PATTERN 1.3 comprises 15 variables[4] spread across four tools: (1) general recidivism for males, (2) violent recidivism for males, (3) general recidivism for females, and (4) violent recidivism for females.[5]

Findings from the 2021 review and revalidation of PATTERN 1.3 are promising:

• PATTERN 1.3 risk scores display a high level of accuracy in their ability to predict recidivism. Findings also indicate that the risk level categories — the probability that an individual has a minimum, low, medium, or high likelihood of recidivating — provide meaningful distinctions of risk. In other words, individuals in higher risk level categories have higher average recidivism rates.
• Individuals are capable of changing risk scores and levels during confinement. And importantly, these changes relate to recidivism outcomes (i.e., individuals who reduced their risk scores and levels from first to last assessment were generally less likely to recidivate).
• PATTERN 1.3 shows relatively high predictive accuracy across racial and ethnic groups. That is, the risk scores predict recidivism well for white, Black, Hispanic, Native American, and Asian individuals.

Although these findings are encouraging, there is work to be done. Results also demonstrate evidence of differential prediction across racial/ethnic groups. This includes the overprediction of Black, Hispanic, and Asian males and females on some of the general recidivism tools, and the underprediction of Black males and females and Native American males, relative to whites, on some of the violent recidivism tools.[6]

To this end, NIJ’s expert consultants are actively pursuing state-of-the-art approaches to address issues of bias with the goal of reducing or eliminating differential predictions with PATTERN. NIJ’s expert consultants will conduct the annual review and revalidation of PATTERN and findings will be released annually. This report will once again assess and summarize the predictive validity, dynamic validity, and racial and ethnic neutrality of PATTERN. The FSA states that revisions or updates can be made to PATTERN to ensure that disparities “are reduced to the greatest extent possible.” As we move forward, DOJ is committed to addressing this important issue and increasing the racial and ethnic neutrality of the tool.

Notes

note 1] [The First Step Act of 2018, U.S.C. § 115-391 (2018).

note 2] Office of the Attorney General, “[The First Step Act of 2018: Risk and Needs Assessment System – UPDATE,” Washington, DC: U.S. Department of Justice, Office of the Attorney General, January 2020.

note 3] National Institute of Justice, “[2021 Review and Revalidation of the First Step Act Risk Assessment Tool,” Washington, DC: U.S. Department of Justice, National Institute of Justice, December 2021. The errors in PATTERN 1.2 were reported in the 2020 review and revalidation report released in January 2021 on the NIJ website.

note 4] For the list of variables included in PATTERN, see National Institute of Justice, “[2021 Review and Revalidation of the First Step Act Risk Assessment Tool.”

note 5] General recidivism is defined as a rearrest or a return to BOP custody within three years of release from BOP custody, and violent recidivism is defined as a rearrest for a suspected act of violence within three years of release from BOP custody. See National Institute of Justice, “[2021 Review and Revalidation of the First Step Act Risk Assessment Tool,” 10.

[note 6] Overprediction means that the instrument assigns a higher recidivism risk score for the race group relative to white group and underprediction means that the instrument assigns a lower recidivism risk score for the race group relative to white group.

Where this page came from

This page was imported from National Institute of Justice. Published by the National Institute of Justice, U.S. Department of Justice, whose content is not copyrighted (its reuse policy); its pictures, stock photography, are not copied.

Nobody has written it yet — it is the source material at a new address, which is why search engines are asked to skip it and why no one earns from it. It is up for grabs: take it on, and it is yours to rewrite and to earn from.