Missing data are a ubiquitous problem for everyone working with data. To evaluate the performance of methods dealing with missing data, researchers perform simulation studies. An important aspect of these studies is the generation of missing values in complete data (i.e. the amputation procedure).

Missingness occurs with all sorts of underlying mechanisms. In this talk I will introduce the three types of missingness: MCAR, MAR and MNAR mechanisms. And I will explain how these missingness mechanisms create difficulties for statistical analyses.

Then, I will show how to generate legitimate missingness to evaluate the performance of missing data methods. Until recently, missing values were generated one variable at a time. Especially when it is the intention to make multiple variables incomplete, this stepwise univariate amputation approach is not always sufficient in creating reliable missing data problems.

We implemented a multivariate amputation procedure into R-function ampute (available in R-package mice). ampute enables the generation of missing values in multiple variables, based on multiple variables, with any desired missingness percentage and much more. With ampute, we have an efficient amputation method to accurately evaluate missing data methodology.