U.S. flag

An official website of the United States government, Department of Justice.

Police Officer Attitudes toward Pre-arrest Behavioral Health Diversion Programs: Identifying Determinants of Support for Deflection Using a Machine Learning Method

NCJ Number
Policing: An International Journal of Police Strategies & Management Volume:   Online Dated: APR 2024
Date Published
April 2024

In this study, researchers use the kernel-based regularized least squares (KRLS) method to explore police officer attitudes toward pre-arrest behavioral health diversion programs.


This study demonstrates the usefulness of the kernel-based regularized least squares (KRLS) method for practitioners and scholars seeking to illuminate patterns in police attitudes. It further underscores the importance of agency leadership in legitimizing deflection as a pathway to addressing behavioral health challenges in communities. Researchers explore the determinants of police officer support for pre-arrest/booking deflection programs that divert people presenting with substance use and/or mental health disorder symptoms out of the criminal justice system and connect them to supportive services. The study analyzes responses from 254 surveys fielded to police officers in Delaware. Questionnaires asked about views on leadership, approaches toward crime, training, occupational experience and officer’s personal characteristics. The study applies a new machine learning method called KRLS for non-linearities and interactions among independent variables. Estimates from a KRLS model are compared with those from an ordinary least square regression (OLS) model. Support for diversion is positively associated with leadership endorsing diversion and thinking of new ways to solve problems. Tough-on-crime attitudes diminish programmatic support. Tenure becomes less predictive of police attitudes in the KRLS model, suggesting interactions with other factors. The KRLS model explains a larger proportion of the variance in officer attitudes than the traditional OLS model. (Published Abstract Provided)

Date Published: April 1, 2024