688 – Novel Regression Methods
Wage Imputation in the OES Survey: A Model-Assisted Approach Incorporating Data from the Quarterly Census of Employment and Wages
Jane Osburn
Bureau of Labor Statistics
A proposed revision of wage imputation procedures in the OES Survey is based on relatively accurate estimates of area and establishment wage levels obtained by using the OES and QCEW datasets together in the multivariate EBLUP estimator of Lohr and Prasad (2003). Current OES procedures define donor cells by the same time period /MSA / Industry / Establishment-size /Occupation as the non-respondent cell, and the mean wage distribution of this donor cell is imputed to the non-respondent. In the case of insufficient response, the cells are collapsed over industry and size groupings. Where necessary, the cells are reformed by State / Industry / Establishment Size, and the process repeats as for the MSA base-level strata. The base-level strata default to USA in a similar manner. The proposed method replaces the OES base-level strata with three successively more aggregated sets of establishment groupings based on multivariate EBLUP estimates of establishment wage levels. Simulation is used to compare the performance of the new approach against the OES method and a variety of alternatives. The results suggest that imputations based on the new approach outperform the current method .