To impute or not impute: that's the question

Researchers often encounter missing values in their datasets, yet they frequently usesuboptimal methods to tackle this missing data problem. This might be explainedby either ignorance of more sophisticated missing data techniques, or anxiety aboutthe complexity of those methods. The main goal of this paper is to present a numberof more sophisticated missing data techniques in non-statistical terms, so that it canbe a guide for researchers or methodological consultant who want to tackle a missing data problem. We first give an overview of older missing data treatments andillustrate their limitations and possible usefulness. After that we wll focus on fourmodern missing data treatments: mean imputation, regression imputation, multipleimputation and a maximum likelihood technique. We illustrate the usefulness of eachtechnique with a practical example.