Ken Robertson, Larry Huff, and Albert Tou (1995) "A Study Of Donor Pools and Imputation Methods For Missing Employment Data," Proceedings of the Section on Survey Research Methods, American Statistical Association, forthcoming.
The Bureau of Labor Statistics Occupational Employment Statistics (OES) survey is a periodic survey of nonfarm establishments that collects occupational employment data on workers by industry. The survey uses a weighting cell adjustment procedure to adjust for unit nonresponse. Previous research has shown that this procedure works well with these data. A weakness of this procedure, as employed in this survey, is that it does not adjust for unit nonresponse in three-digit industries that have no responding units. The cell for the current procedure is, at most, the entire three-digit SIC within the sampled area.
First we will explore multiple donor pools, by allowing the donor pool to expand in increments to all areas and multiple SICs. Next we will apply alternative procedures such as mean imputation, hot deck (nearest neighbor), and hot deck (random selection in a cell) to adjust for nonresponse utilizing the expanded donor pools in the event that the original pool is empty. The first non-empty pool encountered will be used. The accuracy of each of these procedures in producing aggregate level estimates is studied. Of particular interest is the accuracy of aggregate level estimates that include three-digit industries having no responding units.
Last Modified Date: July 19, 2008