Depth functions were developed to extend the univariate notions of median, quantiles, ranks, signs, and order statistics to the setting of multivariate data. Whereas a probability density function measures local probability weight, a depth function measures centrality. The contours of a multivariate depth function induce closely associated multivariate outlyingness, quantile, sign, and rank functions. Together, these functions comprise a powerful methodology for nonparametric multivariate data description, outlier detection, data analysis, and inference, including for example location and scatter estimation, tests of symmetry, and multivariate boxplots. Due to the lack of a natural order in dimension higher than 1, notions such as median and quantile are not uniquely defined, however, posing a challenging conceptual arena. How to define the middle? The middle half? Interesting competing formulations of depth functions in the multivariate setting have evolved, and extensions to functional data in Hilbert space have been developed and more recently, to multivariate functional data. A key question is how generally a notion of depth function can be productively defined. This talk provides a perspective on depth, outlyingness, quantile, and rank functions, through an overview coherently treating concepts, roles, key properties, interrelations, data settings, applications, open issues, and new potentials.
