What works! A meta-analysis of general risk prediction instruments for adult criminal offenders

Thumbnail Image



Journal Title

Journal ISSN

Volume Title


University of New Brunswick


The current dissertation used meta-analytic techniques to summarize the predictive validities for general recidivism using risk assessment studies generated between 2000 and 2017. In total, 263 studies contributed 746 effect sizes involving over one million justice-involved individuals. Given this scale, the results provide comprehensive and relevant information related for the choice and implementation of risk assessment tools/practices to appraise general recidivism across criminal justice contexts. Findings supported the use of theoretically-based, second, third, and fourth generation risk tools as these had predictive validity for appraisal of general recidivism. Tools that were entirely static or that combined static with dynamic factors had larger predictive validities than those containing only dynamic items. For assessment administration method, appraisals conducted using structured professional judgment processes provided comparable predictive validity to actuarial tools. Results also provided continued support for multi-sourced methods of gathering appraisal information, though these predictive validities were at par with single-sourced methods (i.e., file review and client self-report questionnaires). For sample diversity, predictive validities were comparable for mixed ethnicity, Indigenous and Caucasian samples, whereas validities were not as accurate for Black and Hispanic samples. There were no differences in predictive validity categories for client age, type of correctional client (general, forensic-psychiatric, sexual, and violent), or for male and female samples. For outcomes, there was no difference in predictive validity by type of officially-recorded general recidivism and very minimal support for self-supported general recidivism. Stronger predictive validities resulted from follow-up periods exceeding two years duration. The results contained herein will provide policy decision makers and assessors with information to guide choices related to risk assessment tools relevant to their individual target populations and contexts.