580 – Imputation Methods: Challenges with Complex Designs and Complex Data
Sequential Regression Multivariate Imputation in the Current Population Survey Annual Social and Economic Supplement
Charles Hokayem
Centre College
Trivellore Raghunathan
University of Michigan
Jonathan Rothbaum
U.S. Census Bureau
The Current Population Survey Annual Social and Economic Supplement (CPS ASEC) serves as the data source for official income, poverty, and inequality statistics in the United States. The Census Bureau has used a "hot deck" procedure to impute missing income values since 1962. This paper implements an alternative model-based methodology, sequential regression multiple imputation (SRMI), to impute missing income values in the CPS ASEC. SRMI offers several potential advantages over the current hot deck method, including 1) greater flexibility to add additional covariates and 2) using multiple imputation to account for uncertainty in the imputation process. We implement a baseline SRMI with data from the CPS ASEC and then augment this with tax records on earnings from the Social Security Detailed Earnings Records (DER) file. We compare imputed income values from SRMI to those from the hot deck procedure along several dimensions including the mean, median, variance, and common official statistics derived from income (poverty and inequality).