A comparison of some weighting adjustment methods for panel nonresponse

In some surveys, many auxiliary variables are available for respondents and nonrespondents for use in nonresponse adjustment. One decision that arises is how to select which of the auxiliary variables should be used for this purpose and another decision involves how the selected variables should be used. Several approaches to forming weighting adjustments for nonresponse are considered in this research. The methods include those based on logistic regression models, categorical search algorithms, and generalized raking. These methods are applied to adjust for panel nonresponse in the Survey of Income and Program Participation (SIPP). The estimates from the alternative adjustments are assessed by comparing them to one another and to benchmark estimates from other sources.