The number of justice-involved women has risen sharply over the last 40 years, energizing efforts to identify the factors that are most relevant to their future reoffending. These efforts have shown that women share many gender-neutral risk factors with men, and that they also experience gender-responsive risk factors that uniquely impact their likelihood of recidivism. However, there is considerable variation in the impact of these risk factors on recidivism across studies. Research suggests that this variability may stem from methodological characteristics of the studies, differences between study sample populations, and how risk factors are conceptualized and measured. This paper engages a moderation analysis to examine the extent to which the effects of gender-responsive and gender-neutral risk factors on women’s recidivism vary by study, sample and risk level characteristics.
Data from a prior meta-analysis examining gender-neutral and gender-responsive risk factors for recidivism among justice-involved women were used for the present study. Fifty-seven manuscripts describing 64 unique studies provided 782 effect sizes, which were coded into 19 separate “risk domains”. Given the nesting of data, we used a three-level meta-regression framework to examine the moderating effects of 20 variables—grouped into study, sample, and risk factor characteristics—on the relationship between these risk domains and women’s recidivism. Meta-regressions were performed for the 14 risk domains that had a sufficient number of studies for analysis and were found to have significant heterogeneity, which was assessed using log-likelihood ratio tests .
Nine of the 20 moderators examined demonstrated a significant moderating effect on at least one risk domain. Race was found to moderate the effects of substance misuse on recidivism, such that the strength of this domain increased in proportion to the percentage of White women in study samples. Additionally, belonging to a racial or ethnic minority was associated with lower levels of recidivism only when studies used multidimensional recidivism measures. The inclusion of statistical controls for other risk factors was found to moderate risk domains of antisocial associates, family/marital problems, education/employment, and mental health, suggesting that their impact on recidivism was attenuated when models from primary studies included this type of statistical adjustment. The type of mental health diagnosis or victimization reported by participants in primary studies did not significantly moderate any risk domain.
Conclusions and Implications:
This study is the first moderation analysis of meta-analytic findings of gender-neutral and gender-responsive risk factors on women’s recidivism. Our findings have several important implications for social work practice, research, and policy. The majority of the variation in effect sizes among the risk factors examined in this analysis was related to methodological characteristics of primary studies rather than characteristics of study samples, suggesting that the risk factors examined in this study are applicable to diverse groups of women. Our findings also support the importance of using multi-dimensional indicators of recidivism rather than relying on only one indicator for analysis. Finally, victimization and mental health were shown to impact recidivism irrespective of the type of mental illness or victimization involved.