Mixed graphical models with missing data and the partial imputation EM algorithm (Q2742758)

The following model is considered. Let \(G =(V, E)\) denote a graph, where \(E\) is the set of edges, \(V\) the set of vertices, and \(V\) is partitioned as \(V = \Delta \cup \Gamma\) into a dot set \(\Delta\) and a circle set \(F.\) A dot denotes a discrete variable and a circle denotes a continuous variable. Thus the random variables are \(X_V = (X_v)_{v\in V}.\) The absence of an edge between a pair of vertices means that the corresponding variable pair is independent conditionally on the other variables which is the pairwise Markov property with respect to \(G.\) The authors use a set of hyperedges to represent an observed data pattern. A normal graph represents a graphical model and a hypergraph represents an observed data pattern.NEWLINENEWLINENEWLINEIn terms of mixed graphs the decomposition of mixed graphical models with incomplete date is discussed. The authors present a partial imputation method which can be used in the EM algorithm and the Gibbs sampler to speed up their convergence. For a given mixed graphical model and an observed data pattern a large graph decomposes into several small ones so that the original likelihood can be factorized into a product of likelihoods with distinct parameters for small graphs. For the case where a graph cannot be decomposed due to its observed data pattern the authors impute missing data partially such that the graph can be decomposed.