Missing data in interactive high-dimensional data visualization (Q1297827)

The authors discuss the problem of missing values visualization for multidimensional data scatterplots. They propose to use sample means instead of unknown values in 2-D projections of the data. An additional dataset of the same dimension as the main dataset is used to indicate the missing values. It consists of binary indicators of ``missingness'' of the data for each used variable. The indicators are perturbed by small random errors to spread the points as to avoid multiple overplotting. This dataset is called ``missing value shadow'' and can be plotted at projection scatterplots and gistograms as well as the main dataset. The authors describe how the interactive technique of linked views with brushing of subsets in the data allows the statistician to investigate the effects of missingness. Implementations are considered in XGobi and MANET software systems.