B0726
Title: Joint modeling of conditional mean and dispersion with a circular predictor
Authors: Maria Alonso-Pena - Universidade de Santiago de Compostela (Spain) [presenting]
Irene Gijbels - KU Leuven (Belgium)
Rosa Crujeiras - University of Santiago de Compostela (Spain)
Abstract: The simultaneous and flexible estimation of the mean regression function and the dispersion function is considered in situations where the response is a count variable, and the predictor variable is circular. The estimation approach is based on the maximization of the circular local likelihood function, without the assumption of any parametric forms for the regression functions. The conditional distribution is assumed to belong to the double exponential family, which allows us to model overdispersion, underdispersion or a combination of both, where the amount of overdispersion (or underdispersion) may change with the value of the circular predictor. We apply this methodology to study how the number of neuronal spikes in a macaque monkey changes with the direction of a visual stimulus. The new approach allows the flexible estimation of not only the expected number of spikes, but also of the variability in the number of spikes, both as a function of the direction of the stimulus.