Bayesian Networks Model Averaging for Bes Indicators

Abstract

The measure of the equitable and sustainable well-being (Bes) is of growing interest in the last years. The National Institute of Statistics (Istat) provides, for Italy, a wide set of indicators describing each domain of well-being that is, by definition, a multidimensional concept. In this study, we propose the use of Bayesian networks to deal with basic and composite Bes indicators. Its capability to model very complex multivariate dependence structures is useful to describe the relationships between indicators belonging to different domains and, being a probabilistic expert system, the estimated network could be also useful for probabilistic inference and what-if analysis. In this study, all the Bayesian networks structures have been estimated by means of the hill climbing algorithm based on bootstrap resampling and model averaging in order to prevent bias due to deviations from the normality assumption.