Nonresponse Adjustment Using Logistic Regression: To Weight or Not To Weight?

• Activity Number: 464
• Type: Contributed
• Date/Time: Wednesday, August 9, 2006 : 2:00 PM to 3:50 PM
• Sponsor: Section on Survey Research Methods
• Author(s): Eric A. Grau*+ and Frank Potter and Stephen R. Williams and Nuria Diaz-Tena
• Companies: Mathematica Policy Research, Inc. and Mathematica Policy Research, Inc. and Mathematica Policy Research, Inc. and Mathematica Policy Research, Inc.
• Keywords: unit nonresponse ; weighting ; propensity modeling ; nonresponse adjustment ; survey bias and precision ; community tracking study

Abstract

Unit nonresponse in sample surveys is accommodated by reallocating the weights of unit nonrespondents to respondents. One way of doing this is to develop logistic regression models to predict the probability of response. The inverses of the predicted probabilities from these models are then used to adjust the sampling weights. In rounds two and three of the Community Tracking Study (CTS) Household and Physician Surveys, nonresponse adjustments to the weights were carried out using weighted logistic regression models. In the fourth round of the survey, unweighted logistic regression models were used to adjust for nonresponse, with design variables, basic sampling weights, and higher order interactions included in the models, following a methodology introduced in papers by Vartivarian and Little (2003). In this paper, we compare nonresponse adjustments using the two methods.