for drug use (based on use of major drugs), which we also expect to test in our models.
When one considers the almost innumerable possibilities for model specification and analysis represented in the diagram in Table 8, one can more fully appreciate the importance of the objective of this paper, that is, to use an organizing framework based in public management theory to identify important structural, management, and primary work factors in substance abuse treatment programs that might be related to post-treatment outcomes. The analyses described in this paper showed statistically significant and substantively interesting relationships among measures of organizational structure and mission, financial management variables (e.g., revenues per client and revenue sources), human resources management (e.g., staffing levels, the use of case managers, etc.), and measures of service technology (e.g., the provision of supportive services, counseling intensity, etc.). In substance abuse treatment programs and in the model shown in Table 8, the impact of these variables on patient outcomes is mediated through the characteristics and pre-treatment histories of the patients. Researchers using the ODATS data were unable to investigate these types of relationships in the absence of patient-level data. Other researchers using the DATOS and SROS data, with rich client-level data, found their program-level measures to be lacking.
We are fully aware that there are limitations to the NTIES data, too. Only a little more than one-tenth of the full set of service delivery units can be linked to the patient-level data, although this subset still represents more than 4,000 patients. There will be some missing information for some of the program- and patient-level variables, and we will be limited by degrees of freedom in our empirical efforts to fully explore the large number of potential