Title: Some advances on modal regression for circular data
Authors: Rosa Crujeiras - University of Santiago de Compostela (Spain)
Maria Alonso-Pena - Universidade de Santiago de Compostela (Spain) [presenting]
Abstract: Modal regression is a quite effective alternative to classical regression methods when the conditional mean or median (or any other quantile) is not an adequate summary of the behavior of a response variable with respect a certain covariate. This usually happens when the conditional distribution shows asymmetry and/or when more than a single (conditional) mode is present in the data, leading in some scenarios to a more complex estimator (actually, a multifunction) than in mean or quantile regression. We will show how multimodal regression estimation can be accomplished for regression models involving circular variables (response and/or covariate), from a nonparametric perspective. The algorithms will be described in detailed and some examples on the cylinder and the torus will be shown to illustrate the methods.