Journal of Official Statistics, Vol.7, No.4, 1991. pp. 475–498
A Comparison of the Missing-Data Treatments in the Post-Enumeration Program
Joseph L. Schafer
Abstract:Missing data were a major source of uncertainty in the U.S. Census Bureau's Post-Enumeration Program (PEP) to measure coverage error in the 1980 census. A variety of reweighting and imputation techniques led to widely different estimates of the undercount. This paper outlines the important features of missing data in the PEP undercount estimation, identifying why undercount estimates were sensitive to the various missing-data treatments. The twelve sets of PEP estimates should not be regarded as equally plausible alternatives, nor should their range be regarded as a measure of the uncertainty due to missing data. A distinct possibility exists that all of the missing-data treatments may have been inherently conservative, causing all of the estimated rates of gross undercount to be biased downward.
Keywords:Dual-system estimation; ignorable nonresponse; imputation; undercount.
Copyright © Statistics Sweden 1996-2018. Open AccessISSN 0282-423XCreated and Maintained by OKS Group