Empirical Evaluation of Imputation Methods on Quarterly Census of Employment and Wages (QCEW) Data

Marek W. Kaminski and Vinod Kapani


The U.S. Bureau of Labor Statistics’ Quarterly Census of Employment and Wages (QCEW) program currently uses each establishment’s year-ago trend in imputing missing employment and wages data. Ratio method is introduced which is using current trend of employment and wages. An empirical evaluation of well established methods, namely ratio and nearest–neighbor, is undertaken. This paper presents the analytical evaluation of these methods using current trends in the data. The reported data is simulated for imputing employment and wages on a random sample from QCEW Longitudinal Database (LDB). Both methods utilize exclusion criterion for removing influential observations from the computations. Finally, we offer comparisons of both methods, at stratum and aggregate levels, percentage relative errors.