64th ISI World Statistics Congress - Ottawa, Canada

64th ISI World Statistics Congress - Ottawa, Canada

Bias corrected imputation method for missing not at random response mechanism using local polynomial regression

Conference

64th ISI World Statistics Congress - Ottawa, Canada

Format: CPS Poster

Keywords: caliberrfimpute, non-response, nonparametric, population, predictive-mean-matching-imputation, response-probability, sampling, super-population

Session: CPS Posters-02

Monday 17 July 4 p.m. - 5:20 p.m. (Canada/Eastern)

Abstract

Many studies have been conducted to properly handle nonresponse. One method of handling the nonresponse is the nonresponse imputation. A lot of nonresponse imputation methods have been developed and practically used. Most imputation methods assume MCAR(missing completely at random) or MAR(missing at random). Recently, MNAR(missing not at random) nonresponse that are affected by the study variable has occurred frequently, but there are relatively few studies on imputation method for MNAR. The MNAR nonresponse causes bias and reduces the accuracy of imputation whenever response probability is not properly estimated. Practically we do not know both the response probability and the response probability model, and so the best way to impute the missing value is using all information of the available auxiliary variables. In this study we propose a bias corrected imputation method for MNAR nonresponse under non-informative sampling. To estimate the bias, we assume a linear response probability model. The advantage of using a linear response model is that we can obtain the theoretical bias under any super-population model. The obtained bias is applied to some existing imputation methods and we get the bias corrected imputed value. Through simulation studies, we confirm that the suggested bias corrected method gives better results than the existing imputation methods.