Nonparametric circular regression estimation with continuous and categorical predictors

A. Meilán Vila, M. Francisco Fernández

In this work, we propose a nonparametric circular regression framework that accommodates both categorical and continuous predictors through a product-kernel estimator specifically adapted to the circular setting. We study the theoretical properties of the proposed estimator, deriving expressions for its asymptotic bias and variance. Although these results provide insight into the estimator’s behavior, they do not directly yield feasible data-driven smoothing parameter selection rules. To address this, we develop a bootstrap-based bandwidth selection criterion tailored to circular loss functions and compare its performance with cross-validation and rule-of-thumb approaches through simulation experiments. Finally, we illustrate the practical utility of the proposed methodology by analyzing directional error data arising from a spatial orientation experiment under varying sensory conditions.

Keywords: kernel smoothing, bootstrap methods, bandwidth selection, circular data analysis, human orientation

GT Estadística no Paramétrica III: Inferencia no paramétrica para datos circulares
September 5, 2026  10:00 AM
Aula 29

A. Fernández de Marcos, E. García Portugués

J. S. Abston, Y. Larriba, I. Fernández

D. Bolón Rodríguez, D. Paindaveine