Outlier Robust Imputation of Survey Data via Reverse Calibration

Download

PDF Download (185Kb)

Description/Abstract

Outlier robust methods of survey estimation, e.g. trimming, winsorization, are well known (Chambers and Kokic, 1993). However, such methods do not address the important practical problem of creating an “outlier free” data set for general and public use. In particular, what is required in this situation is a data set from which the outlier robust survey estimate can be recovered by the application of standard methods of survey estimation. In this paper we describe an imputation procedure for outlying survey values, called reverse calibration, that achieves this aim. This method can also be used to correct gross errors in survey data, as well as to impute missing values. The paper concludes with an evaluation of the method based on a realistic survey data set.