ABSTRACT
Survey data can be imperfect in various ways. Sampling, noncoverage, interviewer error, and features of the survey design and administration can affect data quality. In particular, surveys typically have missing-data problems due to nonresponse. This can be in the form of unit nonresponse, which occurs when a selected unit (person, establishment, etc.) does not respond to any of the survey items, because of noncontact, refusal, or some other reason; or it can be in the form of item nonresponse, which occurs when a unit responds to some items but not to others. Discussions of several issues related to nonresponse in surveys are contained in the three volumes produced by the Panel on Incomplete Data of the Committee on National Statistics in 1983 (Madow, Nisselson, & Olkin, 1983; Madow, Olkin, & Rubin, 1983; and Madow & Olkin, 1983) as well as in the volume stimulated by the 1999 International Conference on Survey Nonresponse (Groves, Dillman, Eltinge & Little, 2002).
