The research literature suggests that adolescents placed in residential programs due to their delinquent behavior are at high risk for drug and alcohol use and abuse. Research is rare, however, on the effects of residential delinquency-treatment programs on drug- and alcohol-related behaviors. This study examined the comparative effects of community-based, group-home, delinquency-treatment programs in Kansas on participants' self-report measures of drug and alcohol use and abuse, and of some prosocial behaviors. The results indicated that youths (n = 82) participating in group homes using the broadly disseminated Teaching-Family approach had better during-treatment outcomes than youths (n = 103) participating in a set of comparison group homes. Corroboratively, a sample of Teaching-Family youths (n = 28) also had better during-treatment outcomes than a matched no-treatment comparison group of their friends (n = 28), while, in contrast, a sample of youths (n = 33) in the comparison group homes did not differ on during-treatment measures from a matched no-treatment comparison group of their friends (n = 33). In the post-treatment year, however, no outcome difference was evident for any of the comparisons. Regression analyses were conducted to attempt to account for variation in the during treatment drug use and prosocial behavior measures. Significant regression coefficients were found for some measures of treatment process and for measures of youths' pretreatment behavior. In a second set of regression equations, the process and pretreatment measures were less effective in accounting for variance in posttreatment outcomes. The limitations of the research and its implications for the treatment and prevention of drug and alcohol use and abuse in group-home delinquency-treatment programs are discussed.

The American Psychological Association's “Psychology as a Health Care Profession” (1979) contains a review of several studies on the cost-effectiveness of psychotherapy. As we document, that review does not consider viable alternative interpretations or qualify findings in line with the studies' substantial limitations, e.g., their lack of controls for regression or selection biases. Following our critique, we suggest that in light of the financial contingencies surrounding conclusions about the effectiveness of psychotherapy, special care must be taken to ensure that psychology's public stance on the value of psychotherapy contains the qualifications suggested by the scientific evidence. Public statements by the American Psychological Association on the issue of the effectiveness of psychotherapy might be based on a review of the scientific merits of pertinent studies by an independent body of research methodologists.

Cummings's 1985 response to our prior article suggests that the American Psychological Association was justified in not adequately describing or qualifying medical-cost-offset research findings, because the “preponderance of evidence” in the over-all literature is “quite persuasive.” We argue in return that one's willingness to accept a body of evidence as persuasive does not excuse one from providing important qualifications for findings, especially when significant threats to validity exist. We question whether the evidence is indeed persuasive. A review of the evidence suggests that the potential for serious regression and selection biases is not limited to the studies described in the APA document. Cummings acknowledges that studies of offset have weaknesses but argues that collectively they provide evidence analogous to that establishing a causal link between smoking and disease. On the contrary, as we describe, the evidence linking smoking and disease is vastly stronger. We do not find compelling the analogy pairing the Surgeon General's conclusion about smoking with a government agency's report concluding that psychological care reduces medical costs. We appreciate that one would not want to use a no-treatment control group in the kind of research Cummings is currently conducting, but there may be alternative, quasi-experimental designs which allow testing for absolute effects while controlling for possible regression.