Evaluating an alcohol and drug treatment program for the homeless: An econometric approach

The New Orleans Homeless Substance Abusers Project (NOHSAP) was designed as a randomized field experiment to test the effectiveness of a residential alcohol and drug treatment program on the sobriety, employment, housing, and social integration of homeless substance abusers. However, program staff sabotaged randomization into treatment and control groups, and research attrition was also non-random. Non-random assignment to treatment and non-random research attrition threaten internal and external validity by biasing OLS estimates of the effects of treatment and necessitate use of econometric selection bias correction modeling techniques. Results of these corrected models are then used in subsequent estimates of treatment effects on a variety of outcome measures, After correction, positive treatment effects prove relatively modest. However, subsequent analysis suggests that NOHSAP exerted a critical indirect effect on outcomes by facilitating subject's participation in outside substance abuse groups. We conclude with some observations on the policy implications of the substantive results. (C) 1997 Elsevier Science Ltd.