In 2011 the state of California was ordered to dramatically reduce its prison population by the Supreme Court. The state responded to this court order with Public Safety Realignment, a transfer of authority and responsibility for non-violent, non-serious, and non-sexual offenders from the state to the counties. This rapid and universal policy change has created a compelling opportunity to utilize realignment-induced variation in county policies to identify the causal effects of these policies on recidivism outcomes.

The California Department of Corrections and Rehabilitation (CDCR) provided the study team with individual level prison release data from the pre- and post-realignment periods. The CDCR data includes county identifiers, demographic characteristics, criminal histories, current offense characteristics, risks and needs assessments, and recidivism outcomes through one year post release. We have utilized county realignment plans and budget allocations to characterize county policy environments based on the degree to which they emphasize and support the following practices: use of rehabilitative services to address criminogenic needs, intensive supervision for high risk populations, targeted alternative sanctions and traditional incarceration. We have merged the collected county data with CDCR individual level data to identify individuals exposed to different policy treatments.

We then address concerns that differences in the population compositions of treatment and control counties could produce differences in recidivism outcomes that are independent of county policy choices.  We use a non-parametric genetic matching strategy that relies on fewer assumptions than GLM approaches and highlights any failures of common support across treatment and control populations. We then run regressions for each identified policy treatment on our matched sample of individuals released from state prison.  We estimate the differential effect of each policy treatment on recidivism outcomes using a differences-in-differences estimator that controls for time-invariant county characteristics.

Our results will produce estimates of what is formally referred to as the “average treatment effect on the treated.” We will then take the analysis one step further to improve the generalizability of our findings to the control group. From a scientific perspective, there is no need to take this step. The analysis described above will reveal to us the extent to which there is an evidence-basis for the policy treatments pursued. However, from a policy perspective, we would like to be able to extend these findings to inform the policy choices of control counties. Therefore, the last stage of our analysis will be to estimate the “average treatment effect on the controls.” Under this framework, if our differences-in-differences estimator remains negative and significant for a policy treatment, then we can conclude the adoption of that policy by control counties is likely to lead to reductions in recidivism. This analysis can be conducted for the control counties as a group, as well as separately for individual control counties of interest.