An Alternative Strategy for Non-response Weighting Adjustment in Longitudinal Surveys

Abstract

Longitudinal survey organisations can offer data for analysis from many subsets of data that are related to the longitudinal population at the latest wave, from a large number of wave-combinations. However, only one set of non-response weights (which is often based on respondents from all waves up to the latest) is often offered to be used with any of the wave-combinations. This use of non-response weights is a single weighting strategy. Weights are derived based on information from one subset of waves but used for analysis with other subsets of waves. In this paper, the limitations of the single weighting strategywas illustrate. Creating subsets of weights for all the possible combinations of waves is impractical. However, weights are more useful for some combinations than others. a criterion of designing subsets of weights based on considering wave-combinations that are concerned with the same module of questions was evaluated . Data from the British Household Panel Survey (wave 1 to 8) were used to conduct the investigation.I found that the use of a single weighting strategy may lead to an unnecessary loss of respondents and hence less precision and bias on some, but not all, of the survey estimates.