Model-assisted Weighting for Survey Data with Multiple Response Modes

• Activity Number: 109
• Type: Contributed
• Date/Time: Monday, August 12, 2002 : 10:30 AM to 12:20 PM
• Sponsor: Section on Survey Research Methods*
• Author(s): Eric Slud*+
• Affiliation(s): U.S. Census Bureau/University of Maryland
• Keywords: Nonresponse adjustment ; logistic models ; survey weights ; misspecified models ; mean-squared error ; American Community Survey

Abstract

Several large surveys administered by the Census allow responses in several different modes (e.g., mail, telephone, personal interview), in some cases--as in the American Community Survey (ACS)--after sub-sampling at some stages from that part of the sample that has not responded at any earlier stage. Analysis of the survey resullts usually proceeds by weighting responses roughly by inverse unconditional probability of response to the survey. Often, nonresponse in the different modes can be modelled meaningfully in terms of demographic and geographic variables, such as dwelling type and aggregated characteristics related to ethnicity and socieoeconomic status, and the results of such models derived from previous recent surveys could be used to improve the estimation of population totals and domain subtotals, which is ordinarily the goal of large surveys. This paper applies the concepts of sample-survey theory to describe the magnitudes of improvements to nonresponse adjustments, which would be feasible, assisted by correctly and incorrectly specified models, and to illustrate the results numerically using data and models relating to the 1990 decennial census.