A test for selection in matched administrative earnings data

Abstract

The matching of administrative records to labor-market surveys is both commonly done and represents an important innovation in the measurement of earnings. However, a potential problem is that this process might introduce sample-selection bias. Individuals typically must give informed consent to have their earnings matched, and consenters may display systematically different labor-market behavior than non-consenters. In this paper, we use the differential timing of the consent process to test whether individuals in the Health and Retirement Study who consented represent a non-random, thus biased sample. In particular, we find that for both men and women there is a general pattern of negative selection across three measures of pre-entry labor-market behavior: labor-force participation, self-employment, and earnings.