Advances and Applications in Statistics
Volume 51, Issue 4, Pages 261 - 276
(October 2017) http://dx.doi.org/10.17654/AS051040261 |
|
SOME NEW PROPERTIES OF HELLINGER DISTANCE FOR VALIDATING APPROXIMATIONS IN BAYESIAN ANALYSIS
Ali. S. Gargoum
|
Abstract: In several applications in statistics and graphical models, we need to measure the information contained in one random variable about the value of another. Also of fundamental importance is to examine the proximity of one density function to its approximation. In graphical modelling contexts, there are different choices to calculate the ‘distance’ between two densities, and for measuring the strength of an edge connecting two variables in a conditional independence graph. Two measures are particularly important in parametric set up, because they can be written in an algebraic form for most common families of distributions. These are the Kullback-Leibler (K-L) measure of separation and the Hellinger distance. In this article, some new properties and results associated with these measures are discussed in a framework of graphical modelling. |
Keywords and phrases: Bayesian networks, graphical modelling, Hellinger distance, information distance. |
|
Number of Downloads: 394 | Number of Views: 1311 |
|