On the impact of contaminations in graphical Gaussian models

This work analyzes the impact of some kind of contaminants and wrong model assumptions on concentration graph models. The impact is measured in terms of model selection as the correct identification of the conditional independence structure of a vector of gaussian variables. Four different kinds of source of contamination were investigated, in order to consider both the case of occurrence of gross errors and model deviation. It is of interest to assess against which kind of contaminants graphical models have a more robust behavior. The analysis is based on simulated data. The simulation study shows that relatively few contaminated observations in even just one of the variables can have a significant impact on correct model selection, especially when the contaminated variable is a node in a separating set of the graph.