Dealing with missing data

Abstract

What is done with missing data? Does the missingness mechanism matter? Is it a good idea to just use the default options in the major statistical packages? Even some highly trained statisticians do this, so can the non-statistician analysing their own data cope with some of the better techniques for handling missing data? This paper shows how the mean and standard deviation are affected by different methods of imputation, given different missingness mechanisms. Better options than the standard default options are available in the major statistical software, offering the chance to 'do the right thing' to the statistical and non-statistical community alike.