Validating RefUSA micro-data with the Longitudinal Employer-Household Dynamics Data

This paper validates the reliability of employment data in a frequently used establishment panel database assembled by InfoGroup by comparing it with employment and establishment data from the publicly available Longitudinal Employer-Household Dynamics (LEHD) and County Business Patterns (CBP) at the three-digit industry-by-state-by-year and two-digit industry-by-county-by-year levels between 1997 and 2013. We document substantial differences in both their cross-sectional and time series properties. Through an application involving the evaluation of the employment effects of state corporate tax rates, we also illustrate that the inclusion of fixed effects does not eliminate the bias associated with the extrapolation and/or other measurement error. These results suggest that both descriptive evidence and causal inference from the RefUSA data are unreliable. (C) 2017 Elsevier B.V. All rights reserved.