Probabilistic Models for Detecting Census Duplication at the Person and Household Levels

• Activity Number: 453
• Type: Contributed
• Date/Time: Thursday, August 7, 2003 : 8:30 AM to 10:20 AM
• Sponsor: Section on Government Statistics
• Author(s): Robert E. Fay*+
• Companies: U.S. Census Bureau
• Keywords: census duplication ; erroneous enumeration ; computer matching ; Census 2000

Abstract

Computer matching was used in the Accuracy and Coverage Evaluation (ACE) to detect duplicate enumerations of persons, in an attempt to estimate the census's net coverage error, and in Census 2000 itself. The computer matching relied on reported name and date of birth, but these characteristics are not unique identifiers in many cases. In other words, persons with the same name may coincidentally share the same date of birth. A previous paper developed probabilistic models at the person level to distinguish census duplications from matches of different persons coincidentally sharing the same date of birth. But the models did not consider household-level information--in fact, many duplicated persons are members of duplicated households. This paper extends the probabilistic models to reflect household-level information for (1) households linked by exact matches (name, date of birth) for one or more members, and (2) for households that can be identified as duplicated on the basis of close but not exact agreement for two or more household members. The methods are of potential importance for the design of the census in 2010.