The Impact Of Missing Data In Sample Surveys And How To Deal With Missing Data

Abstract

Using anthropometric data from the Kenya Demographic and Health Survey, the research illustrates a model based approach to deal with missing numerical data. The procedure uses maximum likelihood estimates calculated using the Expectation maximization algorithm to generate multiple imputations under the Gaussian model. The variables in the data sets with and without the imputed missing data are then regressed on one another and the results compared. Finally the indicators for wasting, stunting and under weight are produced for data with and without missing data and compared