program. This would include the routine use of outside researchers to both assess the evaluation methodologies and to conduct independent validation studies.
Bloom, Michalopolos and Hill (2005) compare experimental estimates of program impacts with estimates from non‐experimental studies and conclude that successful non‐ experimental studies have several key components. First, comparison and treatment groups need to be drawn from the same economic or institutional settings. Second, outcome measures for both groups need to be comparable and come from a common source. Third, successful analyses require longitudinal data on outcome measures. Fourth, nonparametric methods should be used to choose comparison groups that are observationally similar to treatment group members.
To ensure that the appropriate data are available to researchers evaluating job training programs, we would impose several requirements on the entities administering the programs as well as on state officials. First, administrators of training programs would be required to maintain accurate data on participants including information on sex, race, educational attainment, marital status, household composition, industry of current/previous job, occupation of current/previous job, detailed geographic information, the types of training services received, and a unique identifier that can be matched to other administrative sources. With appropriately robust provisions to ensure privacy, we would provide these individual‐level records to researchers.
Second, and again with robust privacy provisions, we would require state officials participating in federal training programs to provide researchers with administrative data on individuals participating in other government programs such as unemployment insurance claimants, ES participants, individuals participating in the Temporary Assistance for Needy Families program, and food stamp recipients. These data are important for selecting a comparison sample of individuals that are observationally equivalent to individuals receiving training. These data would need to contain similar information as the data on participants, including a unique identifier that allows for a link to other administrative data.