A Jack of All Trades But a Master of None? Evaluating the Performance of the Level of Service Inventory–Revised (LSI-R) in the Assessment of Risk and Need

ABSTRACT

We evaluate the performance of the LSI-R in assessing risk and need among 26,001 offenders released from Minnesota prisons between 2003 and 2011. Using multiple performance metrics, the results reveal the LSI-R incorrectly classified 41 percent of the offenders, and its area under the curve (AUC) was 0.628. Drawing on the extensive program evaluation research conducted on Minnesota prisoners over the last decade, we evaluated the LSI-R’s assessment of need by examining the relationship between LSI-R domain scores, participation in effective interventions, and recidivism outcomes. The findings provide, at best, moderate support for the LSI-R’s performance in assessing need. Interventions targeting the education/employment, alcohol/drug, and emotional/personal domains generally yielded better recidivism outcomes among offenders with higher scores in these domains.

KEYWORDS:

• Risk assessment
• needs assessment
• reliability
• predictive validity
• prison

Notes

1. The direction of the odds ratios for the 16 interventions remained the same across the nine domains for the highest need and lowest need groups, though there were a few differences as to whether the z score was statistically significant. For CD treatment, the z score was no longer significant for the alcohol/drug domain, but it reached significance for the financial domain. CIP was no longer significant for the attitude/orientation domain, although it achieved significance for the alcohol/drug domain. Visiting and PoP reached significance for the attitude/orientation domain, whereas secondary degree was not significant for the family domain.