The 1990 Post-Enumeration Survey: Operations and Results

Abstract The Census Bureau has struggled for decades with the problem of undercount in the population census. Although the net national undercount has been greatly reduced in recent censuses, it still tends to display important differences by race, ethnic origin, and geographic location. The 1990 Post-Enumeration Survey (PES) was designed to produce Census tabulation of states and local areas corrected for the undercount or overcount of population. The PES was the subject of litigation between the federal government and a coalition of states and local governments. Because of the litigation, the PES was conducted under specific guidelines concerning timing, prespecification, and quality. The PES measured Census omissions by independently interviewing a stratified sample of the population. It measured Census erroneous enumerations by a dependent reinterview of a sample of Census records and by searching the records for duplicates. A dual-system estimator (DSE) was used to prepare estimates of the population by post-strata. Adjustment factors were computed as the ratio of these estimates to the census count. These factors were smoothed using a generalized linear model and then applied to the census counts by block and post-strata to produce adjusted census estimates. Although the government decided not to release these numbers as the official census results, the Census Bureau has conducted further research to improve these estimates to incorporate them into the postcensal estimates program. The revisions have included new post-strata and corrections of errors found in the original estimates. The results of the PES show a differential undercount by race and ethnic group and by owner/nonowner status. They also demonstrate differences in undercount by geography.