Federal Prison Residential Drug Treatment Reduces Substance Use And Arrests After Release - .Statistical Data Included - )

American Journal of Drug and Alcohol Abuse,  May, 2001  by Bernadette Pelissier,  Susan Wallace,  Joyce Ann O'Neil,  Gerald G. Gaes,  Scott Camp,  William Rhodes,  William Saylor

• E-mail
• Print
• Link

INTRODUCTION

In recent years, the number of persons incarcerated in the United States has increased dramatically. Between 1985 and 1996, the incarcerated population increased by an average of 7.8% per year (1). In addition, 62.2% of state prison inmates and 42.1% of federal inmates reported being regular drug users (2). The combination of a rapidly increasing prison population and the high percentage of prisoners with a history of drug use have led policymakers and correctional practitioners to call for increased funding of prison- and community-based drug treatment programs. As a result, prison-based treatment programs for substance-abusing offenders expanded greatly in the 1990s.

Evaluation of the effectiveness of treatment programs among criminal justice populations has largely focused on nonincarcerated populations and has examined the effectiveness of drug treatment as an alternative to incarceration or prosecution or as a condition of probation or parole (3-7). Even though in-prison residential programs have been designed and implemented throughout the United States, there are few evaluations of these programs. Outcome evaluations have been limited to programs in six state prison systems. Those evaluations typically reported that treatment lowered recidivism (as measured by arrests, reconvictions, and returns to prison), decreased postrelease drug use, and curtailed self-reported illegal activities (8-18; H. K. Wexler, G. DeLeon, G. Thomas, D. Kressel, and J. Peters, "The Amity Prison TC Evaluation: Reincarceration Outcomes," unpublished manuscript, 1997). There are, however, significant methodological weaknesses in the majority of these studies, the most common and most important being the lack of attention to the problem of selection bias.

Under ideal circumstances, causal inferences can be imputed from research designs that use random assignment of subjects to different intervention protocols. There are three related problems in implementing experimental designs in applied social settings. First, there are obvious administrative and ethical reasons that make it difficult to implement random assignment studies in applied settings. Second, even when random assignment is possible, a problem arises in that subjects inevitably select themselves, or are selected by others, out of treatment (and control) groups by dropping out of treatment or refusing follow-up interviews. This is often called the problem of noncompliance (19). Third, when random assignment is not possible, the problem of noncompliance is compounded by the inevitability of subjects selecting themselves into treatment. In our study, there were four prominent selection processes that filtered respondents into and out of the study: self-selection, administrative (or clinical) selection, treatment selection, and posttreatment selection.

The first process, self-selection, was created by internal motivational states and/or external incentives that disposed some people to volunteer for treatment. The second process, administrative selection, reflected the clinical judgment exercised by treatment providers and other administrators who determined whether someone was chosen for a program. The third process, treatment selection, weeded out clients who could not or refused to meet the program demands after entering treatment. Although clients weeded themselves out by dropping out of treatment, treatment providers in different programs exercised clinical judgment in expelling participants from treatment. The last selection pressure occurred when clients were lost to follow-up. The result of such selection influences was that the treatment and control groups differed for unintended and uncontrolled reasons, and these differences potentially introduced bias into the outcomes of the study.

Collectively, this set of selection processes is known as selection bias. When selection bias has changed the composition of the control group, the treatment group, or both groups, observed differences between the control group and the treatment group may be due in part, or entirely, to differences in the composition of the respective groups. In short, the effect of treatment is confounded by the effect of selection processes, and there is no simple means of unraveling the two (20, 21). One of the major challenges inherent in evaluation research is to find design, measurement, and analysis methods to minimize the effects of selection bias. Such methods must be able to differentiate the effects attributable to the selection processes from those attributable to treatment.

Many researchers assume that selection bias usually results in a treatment group composed of subjects who are more motivated to change and who have an inherently lower risk of postintervention failure. However, it is also possible that selection processes operate in an opposite manner. For example, there might be an incentive structure that encourages higher risk subjects, rather than lower risk subjects, to enter treatment. Another possibility is that treatment selection is tightly controlled by providers, who reserve treatment beds for the most difficult cases. Given the pervasive and difficult problems of selection bias, most evaluations have not employed methods that allow the evaluators to move beyond speculation about whether selection bias exists and, if it does, about the direction of the selection bias. We attempted to rectify these shortcomings in our study.