Research on the implementation of the Job Training Partnership Act (JTPA) included a sample of more than 20,000 people in 16 service delivery areas (SDAs). The random assignment model for the study produced estimates of the impact of the overall JTPA program in this sample of SDAs, plus separate impact estimates for individuals recommended for classroom occupational training and on-the-job training. Because some SDAs declined to participate due to fears that the study would disrupt or have a later negative impact on their operations, the U.S. Department of Labor and the research team altered the exclusion rules for the control group, provided technical assistance to help SDAs and service providers meet enrollment goals, simplified the research design, increased the compensation for sites' efforts, and urged states to administer the performance standards system flexibly. The following lessons for future research were developed: (1) when research goals conflict, priorities must be set quickly so that a realistic plan can be developed; (2) if generalizable findings are necessary, major adjustments may have to be made in the study design to recruit a representative sample; (3) if participation by sites is voluntary, sites must be offered incentives to participate; (4) implementing a random assignment research design changes the operation of an ongoing program; and (5) technical assistance and staff training are important. (Appendices provide an overview of JTPA, profiles of the sites, tables comparing participating SDAs with others not participating; and 45 references.) (KC)