Survey undercoverage is often thought to be an intractable problem. The surveys conducted by the Bureau of the Census are widely believed to provide as good or better population coverage than those carried out by other survey organizations, but its surveys miss high percentages of people in certain groups. As a result, the Bureau is investigating various methods to improve the coverage of its household surveys.

Undercoverage occurs in two main forms -- 1) entire households are missed and 2) individuals are missed in households that are partially covered. Whole-household undercoverage occurs when a household eligible for interview is excluded from the survey, typically because the dwelling where the household resides has been omitted from the listing of addresses for the area. Within-household undercoverage occurs when one or more members of a sample household are omitted from the household roster.

This paper focuses on within-household undercoverage, which was the major topic of a study conducted on behalf of the Bureau of the Census by the National Opinion Research Center (NORC). Funding was provided primarily through the Survey of Income and Program Participation (SIPP) with additional support by the National Center for Health Statistics. In brief, the study used 3 questionnaire versions. One was a control version and the other two used an expanded set of experimental roster questions. One of these experimental versions 3 required full names and the other allowed "anonymous" responses. The most important findings are that the anonymous questionnaire resulted in a much higher average number of Black males listed as usual household residents (on the order of 33% higher). There was no evidence, however, that the expanded roster questions by themselves had any effect.