*Names in bold indicate Presenter
This paper addresses the following two questions:
- Do administrative and survey earnings data yield similar impact estimates on employment and earnings during the periods that both data sources cover?
- Are the administrative and survey impact estimates more similar for some key population subgroups (defined, for example, by gender, age, race/ethnicity, educational level, arrest history, and residential/nonresidential status) than for others?
The impact results using the survey and administrative data could differ because of survey nonresponse bias or reporting differences in the two types of data. For instance, administrative records do not cover all types of jobs, and during the survey, some sample members may not have accurately recalled their earnings over a long follow-up period, and some may have systematically over- or underreported them.
We find that the pattern of earnings impacts using the administrative and survey data are similar in periods covered by both data sources, although the survey-based impact estimates are larger and more often statistically significant. This occurs primarily because reported earnings levels are substantially higher according to the survey data.
Our results indicate that the survey and tax data provide complementary earnings information for low-income youths. It is difficult to assess which data source provides more accurate information. Reported earnings levels for the Job Corps sample are nearly double in the survey data, suggesting that considerable amounts of earnings are not captured in the tax data. This pattern emerges across broad groups of youths defined by their demographic and job characteristics, and the undercount appears to be especially large for those in short-term casual jobs that offer low wages and few fringe benefits. On the other hand, survey-based earnings measures appear to be biased upward, because of overreporting of hours worked and survey nonresponse bias (which was more pronounced for treatments than controls).
Differences between the survey and tax data are of policy concern, because these data sources are typically used to measure the performance of programs that serve disadvantaged youths and to measure youth poverty rates. Additional research in this area is needed to provide more complete explanations for the survey-tax earnings differences for this population.