440 – Missing Data, Imputation, and Calibration
An Evaluation of Backwards Imputation for the Annual Survey of Public Employment & Payroll
Junilsa Toribio
U.S. Census Bureau
The Annual Survey of Public Employment & Payroll (ASPEP) provides state and local government data on full-time and part-time employment, part-time hours worked, full-time equivalent employment, and payroll statistics by governmental function. For nonresponding general purpose governments, dependent and independent school districts, and special district governments with no historical data available (i.e. births), missing data are imputed using the hot-deck imputation method. Since the hot-deck imputation method entails the use of a random donor for imputation, it is possible that the imputed data do not accurately reflect the data of the unit that was imputed. In an effort to increase data quality, research was conducted to determine if a more accurate imputed value could later be obtained through backwards imputation. This paper describes the results of this research. It discusses the backwards imputation methodology as implemented for the survey and compares the current hot-deck methodology with the backwards imputation methodology.