Project 5: Statistical Adjustments for Nonresponse

• Project Directors: Rod Little, Sonya Vartivarian

A basic survey estimation strategy is to adjust survey weights to compensate for nonresponse.  The most common nonresponse adjustment procedure is to multiply the probability-based weight assigned to the case by the inverse of the weighted response rates in a set of adjustment cells.  That is, within each cell, the ratio of the sum of the sampling weights of the respondents and nonrespondents to the sampling weights of the respondents is found and this adjustment factor is used to expand the sampling weights for the respondents.

Nonresponse and statistical adjustments that attempt to reduce its impact are extremely common features of surveys on health and other topics.  The traditional procedures used to reduce nonresponse and to mitigate its impact are the subject of increasingly widespread doubts.  At the same time, response rates are continuing to decline, heightening concerns about the impact of nonresponse bias.  This project takes a fresh look at the problem of statistical procedures to reduce the impact of nonresponse; it promises to yield results that are wide interest to health researchers and survey statisticians.

This project examines this common procedure and compares it with several alternatives.  Both theory and simulation results indicate that the use of sample-weighted response rates is either unnecessary or incorrect.  A better approach is to model nonresponse as a function of the design variables (e.g., the strata) that determine the sampling weights and the auxiliary variables used to define adjustment cells and to estimate the nonresponse adjustment as the inverse of the fitted response probability from this model.  This approach can be implemented by creating adjustment cells that include the design variables in the cross clarification; if this approach would produce too many cells, response propensity weighting can be used instead.

Publications:

On Weighting the Rates in Nonresponsive Weights

On Weighting the Rates in Nonresponsive Weights (revised)