Weighting of household survey data: A comparison of various calibration, integrated and cosmetic estimators
This paper compares the bias and efficiencies of different estimation techniques aimed at creating a unique set of case weights, which can be used to estimate characteristics for both person and household variables in a household sampling survey. The resulting estimates are compatible with external population information and thus attempt to correct for differential non-response (i.e. under-representation of certain parts of the population). The existing techniques of calibration and integrated weighting and cosmetic estimators are combined and modified to give improved weighted estimators. The theoretical properties of these estimators are outlined, and the performance of the estimators is investigated via simulation studies and by application to a data set. The effect of using auxiliary information on persons only, or on both persons and households, is evaluated. The performance of the estimators under condition of differential non-response has been simulated by using unequal probability sampling to mimic such conditions.