Alternatives of the A.C.E. Missing Data Evaluation

• Activity Number: 348
• Type: Topic Contributed
• Date/Time: Wednesday, August 14, 2002 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Anne Kearney*+ and Thomas Belin and Rita Petroni and Donald Keathley and Bruce Spencer
• Affiliation(s): U.S. Census Bureau and University of California, Los Angeles and U.S. Census Bureau and U.S. Census Bureau and Northwestern University
• Keywords: noninterview adjustment ; late data ; logistic regression ; nonignorable missingness

Abstract

The Accuracy and Coverage Evaluation relies on dual system estimation (DSE) to estimate coverage in Census 2000. As in most surveys, missing data in the A.C.E. result from noninterviews and item nonresponse. The A.C.E. must account for missing data to calculate the DSEs. It does this by implementing a set of procedures to account for household noninterviews, missing demographic characteristics, and unresolved match, resident, or enumeration status. The A.C.E. spreads noninterviewed household weights over interviewed households. It uses distributions and hot decks to impute for the missing demographic characteristics. Finally, it uses an imputation cell procedure to impute missing status probabilities.

We developed alternative procedures to examine the spread of DSEs when these are used on the same A.C.E. data. We developed the following alternatives: alternative noninterview adjustment (NIA) cells; nearest neighbor NIA; late-arriving data to account for missing data; logistic regression for assigning match, residence, and enumeration status probabilities; and nonignorable missingness procedures. This paper discusses the motivation and development of these alternatives.