The effective evaluation of correctional programs is critically important. However, research in corrections rarely allows for the randomization of offenders to conditions of the study. This limitation compromises internal validity, and thus, causal conclusions can rarely be drawn. Increasingly, researchers are employing propensity score matching (PSM) to mitigate this problem. PSM involves matching offenders in an experimental condition with offenders in a control condition on several relevant covariates. Comparisons between these two groups are then more meaningful because we can assume a relative equivalency between groups on the matched characteristics. When utilized appropriately, this procedure can mimic randomization and allow for causal conclusions. Although it is encouraging that this technique is increasingly employed in corrections, it is often misunderstood or misreported. This article provides an overview of the basic methodology and considerations for PSM. In addition, it provides an example of a correctional education program evaluation using PSM strategies. The inability to maintain experimental control is a fundamental impediment to research in corrections (Sherman et al., 1997). A lack of experimental control is due, primarily, to the fact that random assignment of subjects to conditions of a study is generally not feasible. This limitation makes the determination of causality untenable. For instance, while many correlates to the risk of recidivism are known, without proper experimental control we cannot determine if these variables have a causal connection with recidivism. This methodological issue plagues program evaluations as well. If one wishes to compare the three-year recidivism rate of offenders who enroll in vocational training versus those who do not pursue training, preexisting variability between the groups, as opposed to the program's efficacy, could explain post-treatment differences. However, methodologies can be utilized to maximize the strength of empirical research, given the inherent limitations that render randomization unfeasible. This paper addresses the use of PSM, as it can mimic true randomization. This is a basic overview, as the nuances of PSM are expansive. The following is meant to motivate a more systematic reporting of such research and to provide a starting point for those interested in utilizing this procedure.
Available at: http://works.bepress.com/jason-piccone/4/