Duwe (2010) used a retrospective quasi-experimental design to examine the effectiveness of prison-based chemical dependency (CD) treatment provided by the Minnesota Department of Corrections (MNDOC) to reduce recidivism among chemically abusive or dependent offenders. The study compared recidivism outcomes between treated offenders and a matched comparison group of untreated offenders.
The main outcome of interest was recidivism, which was defined as rearrest, felony reconviction, or reincarceration for a new sentence. Recidivism data was collected on offenders through Dec. 31, 2008. The follow-up periods ranged from 36–48 months. Data on arrests and convictions was collected from the Minnesota Bureau of Criminal Apprehension. Data on reincarceration was collected from the Correctional Operations Management System maintained by the MNDOC.
Study participants included 3,499 offenders who were directed to enter CD treatment when they were admitted to prison after 2001. All of the offenders were released during 2005. Of these 3,499 offenders, 1,164 participated in CD treatment, and the remaining 2,335 were not offered treatment, primarily due to a lack of available space in treatment (35 were removed from this sample because they refused treatment). The study used the propensity score matching (PSM) method to control for selection bias and match treated offenders to nontreated offenders. PSM involves finding a predicted probability of selection (or a propensity score) that is estimated by a logistic regression model in which selection into the CD treatment is the dependent variable and the predictor variables included 17 control variables that could have had an impact on the selection process (such as sex, race, and prior felony conviction).
The PSM procedure resulted in matching 926 offenders in the treatment group with 926 offenders in the non-treatment group. The matched treatment group was 90 percent male and 43.5 percent minority, with an average age of 33.44 years at release from prison. The matched untreated group was 88.5 percent male and 45 percent minority, with an average age of 33.3 years at release. Due to the PSM process, there were no significant differences between the two groups.
The study also looked at the effects of treatment by comparing completers and dropouts in the treated group, and the effects of program duration by separating out treated offenders who participated in short-, medium-, and long-term programming. Separate PSM were estimated for each of the five measures of treatment:
· Treatment completers (n= 843)
· Treatment dropouts (n= 321)
· Short-term program participants (n= 671)
· Medium-term program participants (n= 393)
· Long-term program participants (n= 100)
Untreated offenders were matched with treated offenders for each of the five treatment measures.
Survival analysis models were used to analyze the outcome results because they use time-dependent data, which was important to help determine not only whether offenders recidivated but also when they recidivated. The study also used the Cox regression model, which included both “time” and “status” variables in estimating the impact of treatment on recidivism. The “time” variable measures the amount of time from the date of release until the date of the first rearrest, reconviction, or reincarceration. For those who did not recidivate, the date of Dec. 31, 2008, was used. The “status” variable measured whether offenders reoffended during the period they were at risk to recidivate. The time offenders spent in prison as a result of violating supervised release was subtracted from their at-risk period, but only if it preceded a rearrest, reconviction, and reincarceration for a new offense or if the offender did not recidivate prior to Jan. 1, 2009.