The analysis produces treatment effects that approximately double the effects of the arrest intervention on subsequent recidivism of abusers reported in earlier publications (1984, 1987). It is suggested that researchers undertaking randomized field studies should routinely anticipate implementation problems. A wide range of potential obstacles to implementation should be identified and discussed, and appropriate responses should be articulated and practiced. During the experiment, the assignment of cases to treatments should be carefully monitored, and documented problems should be addressed immediately. Data also should be collected on the implementation of random assignment in a form that can be used in later data analyses. Finally, statistical procedures should be used that give the impact of any faulty random assignment. Likelihood function equations are appended. 2 tables and 19 references. (Author abstract modified)