Targeted Refreshment Sampling
Longitudinal data, necessary for examining change, are subject to attrition. Depending on the process determining non-response, such attrition can be innocuous but, more likely, will result in biased inference. A simple way of addressing this is to use attrition weights. However, such weights are only appropriate when they control for all the variables causing non-response. Often, it is not possible to know whether all important influences have been captured. Refreshment samples are an alternative means of restoring the representativeness of a sample affected by non-response. They can also provide a means of testing whether attrition can be satisfactorily dealt with by re-weighting. This project artificially constructs refreshment samples from administrative outcomes linked to survey data. In so doing it, it identifies how linked administrative data can be used to test the type of non-response and therefore whether conventional weights are sufficient. It also explores the potential of refreshment samples and variants of re-weighting to correct for attrition bias.
Dorsett, R. (2010) Adjusting for nonignorable sample attrition using survey substitutes identified by propensity score matching: an empirical investigation using labour market data. Journal of Official Statistics 26(1): 105-125. PDF
Economic and Social Research Council