Mathematikgebäude, Hörsaal E28

Abstract: Standard dependence measures considered in the (mostly non-mathematical) literature like Pearson correlation, Spearman correlation, the Maximal information coefficient (MIC), and Schweitzer and Wolff’s famous sigma are symmetric, i.e. they assign each pair (X, Y) of random variables the same dependence as they assign the pair (Y, X). Independence of two random variables is a symmetric concept modelling the situation that knowing X does not yield any information gain about Y and vice versa - dependence, however, is not. Thinking, for instance, of a sample (x1, y1), ..., (xn, yn) roughly in the shape of a noisy letter V, it is without doubt (on average) easier to predict the y-value given the x-value than vice versa.
In 2010 a Markov kernel based, scale-free dependence measure quantifying directed dependence of pairs of random variables was introduced in [1]. In 2018 a checkerboard-based estimator for this dependence measure was derived and implemented in the R-package qad (short for quantification of asymmetry in dependence). Given a sample sample (x1, y1), ..., (xn, yn) the R-package estimates the dependence of the second variable on the first one and vice versa, and, as a byproduct, calculates the asymmetry in dependence of the underlying dependence structure.
After recalling some background on copulas the talk will sketch the basic ideas behind the qad-estimator, show that the estimator is strongly consistent (without any regularity assumptions), and illustrate its asymptotic as well as its small sample properties by some examples.

[1] W. Trutschnig: On a strong metric on the space of copulas and its induced dependence measure, Journal of Mathematical Analysis and Applications 384, 690-705 (2011), doi:10.1016/j.jmaa.2011.06.013

Vortrag im Rahmen des 50. Geburtstags der Fakultät für Mathematik

Hinweis: Der Vortrag fällt leider aus. Ein neuer Termin wird rechtzeitig bekanntgegeben.

Prof. Dr. Wolfgang Trutschnig

Paris-Lodron-Universität Salzburg