Distribution function estimation with calibration on principal components

Abstract

The calibration method is a convenient means of incorporating auxiliary information when several parameters must be estimated. This approach has recently been used to develop new estimators for the distribution function. However, the auxiliary information available may generate a large dataset, provoking a loss of e ciency in the estimators obtained, due to over-calibration. We propose adapting the calibration using principal components, in order to avoid the negative consequences of over-calibration when estimating the distribution function.