Data integration by graphical models

Statistical matching aims at combining information obtained from different non-overlapping sample surveys. The main target is in constructing a complete synthetic data set where all the variables of interest are jointly observed. In this paper we propose the use of Bayesian Networks to deal with the statistical matching problem. Bayesian networks admit a recursive factorization of a joint distribution useful both for data integration and for evaluating the statistical matching uncertainty in the multivariate context.