Weighting for non-monotonic response pattern in longitudinal surveys

In longitudinal studies, analysis can be based on any one of a large number of wavecombinations. However, only one set of non-response weights (often based on respondents from all waves up to the latest) is typically offered on public use data files. We refer to this as a single weighting strategy (SWS). This paper uses data from the British Household Panel Survey to illustrate the limitations of the SWS. We evaluate the effect of designing weights based on response to wave-combinations concerned with the same module of questions. The analysis shows that the use of SWS may lead to an unnecessary loss of respondents if used with a different combination of waves. This leads to less precision on some, but not all, of the survey estimates.