An Investigation of the Extent and Consequences of Measurement Error in Labor-Economic Survey Data
This paper summarizes results from a validation study in which administrative records from a large manufacturing company are used to validate survey responses of a sample of workers from that company. Relatively little evidence of bias is found in reports of prior-year earnings and unemployment, the more salient fringe benefits, and union status. Quite diverse amounts of error variance are found in the survey reports, ranging from very small for reports of job tenure, annual earnings, and annual unemployment to very large for work hours and especially for a measure of hourly earnings constructed by dividing annual earnings by annual work hours. Furthermore, it is found that the conventional assumptions regarding measurement error in earnings models are not justified. In particular, there appears to be an important correlation between the measurement error in the reports of earnings and the level of job tenure, producing a bias in the estimated payoff to job tenure of roughly 30%. This result suggests a need for a greater understanding of the correlates of survey response errors.