Dealing with Non-response Error in Longitudinal Sample Surveys: A comparison Between the Standard Weighting Approach and Sub-group Tailored Weighting Approach

Abstract

The standard weighting approach (SWA) for the treatment of unit non-response, being based on general variables for the whole sample, dose not take into account possible differences in response propensity in different sub-groups in the sample a thing that may lead to bias in estimates. In this research, the possibility of reducing bias in survey-based estimates has been investigated by using variables correlated with the non-response propensity in the subgroup on which analysis is to be conducted. This approach called “Subgroup-Tailored Weighting Approach (S-TWA) " led, in general, to estimates of propensity in sub-groups similar to those obtained for the whole sample by the SWA. However some of the estimates in sub-groups obtained by the S-TWA approach turned out to be superior to those obtained by the SWA approach as reflected in smaller standard errors. The S-TWA seems to perform best when the sample consist of sub-groups that vary drastically with respect to non-response propensity.