Missing Data in Interactive High-Dimensional Data Visualization

We describe techniques for the interactive exploratory analysis of multi-variate data with missing values. The approach is to 1) provide trivial imputations such as fixed values, 2) accept multiple imputations computed elsewhere, and 3) provide a means for keeping track of the location of missing values in the data. The techniques have two major uses: First, they support the exploration of missing values, their correlations across variables and their associations with the variables of interest. Second, the techniques support the investigation and comparison of precomputed imputation schemes; in particular, they can be used to informally diagnose the adequacy of imputations. The techniques are illustrated with an implementation in the Xgobi software.