A multivariate long-term model of drug-related behaviors and social interventions was employed, and the model's forecasting performance was examined. Data used for model development were based on retrospective self-report information for various drug-related behaviors. This information was provided by methadone maintenance patients in southern California. For time-series analysis, each subject's longitudinal addiction history was aggregated to provide group-level data which consisted of 99 bimonthly period for five variables: abstinence from drugs, daily drug use, property crime, legal supervision, and methadone maintenance. Postsample forecasting performance was compared between the long-term time-series model and a more common time-series model which captures only short-term relationships. Overall, results demonstrated superior performance by the long-term model, indicating its adequacy to explain system dynamics among the five variables. Simulated responses of variables to hypothetical policy changes in methadone maintenance revealed an asymmetric nature between methadone maintenance increase and decrease. Policy implications of the model for understanding drug use and drug treatment are discussed. Additional data on the model are appended. 15 references, 8 footnotes, 5 tables, and 4 figures