B0956
Title: Strictly positive empirical expectile depth
Authors: Ignacio Cascos - Universidad Carlos III de Madrid (Spain) [presenting]
Abstract: Data depths have become popular tools in multivariate statistics. They associate each point in the multivariate space with its degree of centrality with respect to a multivariate sample allowing to compare any two points in terms of such centrality. The greater the depth of a point is, the better it fits the data cloud, while outliers attain low-depth values. The empirical counterpart of several popular depth notions (halfspace, simplicial, zonoid, expectile, etc.) assumes zero value for any point out of the convex hull of the dataset, which becomes a problem when comparing points in that region. Some alternative constructions to the halfspace and zonoid depths have been proposed in the literature to circumvent the problem, while empirical expectile depths are presented which are always strictly positive.
