Prediction regions for the visualization of incomplete data sets (Q1861586)

A multivariate data set with missing values is considered. The missing data are replaced by randomly sampled values with conditional distribution (given the observed data) based on some probabilistic model of the sample. The procedure of replacement is repeated many times and so a collection of datasets with different imputations of missing data is produced (this is called a multiple imputation technique). For the visualization of such data, the authors propose to use principal component plots at which the missing data are replaced by prediction regions. In the considered examples, these regions are simply ellipses or intervals (lines). A NORM package was used within the S-PLUS statistical environment for the imputations and graphical presentations.