Abstract

This publication represents a technical summary report of the Urban Institute’s evaluation of efforts with the Minnesota Department of Corrections (MnDOC) to improve the surveillance system in two state prisons: Stillwater Correctional Facility (STW) and Moose Lake Correctional Facility (ML). The goal of this study was to conduct a rigorous process and impact evaluation of the steps STW and ML took to optimize their surveillance systems, which included repositioning existing cameras, installing new cameras, and making other infrastructural upgrades. In addition, ML integrated an audio analytic technology in their system that would alert on-unit security staff through a visual and audio alert when it detected sounds associated with anger, fear, or verbal aggression.

The evaluation used a mixed-methods research design. Qualitative collection included stakeholder interviews and in-depth observations of the camera operations at ML and STW before, during, and after the upgrades. We interviewed wardens, supervisors and officers working in the intervention units, and numerous other individuals who oversaw operations, investigations, information technology, and camera installation and configuration in ML and STW. We also collected quantitative administrative data from ML and comparison facilities and employed comparative interrupted time-series (CITS) analyses to examine changes in two outcomes following the intervention: (1) total misconduct incidents and (2) guilty dispositions. To support the CITS, we identified another unit in ML that did not upgrade its surveillance system but was similar to the intervention housing unit in terms of population, architecture, and misconduct levels (internal comparison unit), and used the synthetic control method to create another comparison unit derived from the three other medium-security prisons operated by MnDOC (external comparison unit).

Findings from the CITS yielded limited evidence that the intervention reduced misconducts. However, we did find that the installation of cameras in new locations to reduce blind spots increased the number of guilty dispositions. These upgrades seem to support misconduct investigations by helping staff identify people who committed or witnessed these incidents. We also found challenges with the integration and use of the audio analytic technology in ML; all the alerts during the study period were either nuisance alerts (those triggered by people talking loudly to each other or other non-aggressive sounds) or false alerts (those without an obvious or perceptible audio trigger), with none of the alerts being triggered by fights or aggressive behavior as intended.