ABSTRACT
Calibration of weights (Deville and Särndal, 1992; Deville, 1993) is a technique for using the whole cohort in analysis of a subsample. In common with many other techniques from survey sampling, it is simple and requires no assumptions for valid inference, but may be less efficient than approaches relying more heavily on models. In this chapter I will explain how calibration works; how it is related to other, possibly more familiar, techniques; and how to do it in R. The name ‘calibration’ is unfortunate, since it is used for so many unrelated techniques across statistics (and science). It is often better to use the less ambiguous, if more obscure, term generalised raking, a name that comes from one of the computational algorithms used for an early version of the method. Not many people will know what it means, but search engines will find the right results.
