Abstract
In this talk the interest is in estimation in regression models with circular covariates (and continuous or discrete real-valued response), where the conditional distribution of the response given the covariate is specified through a parametric model. The discussed approach extends the idea of (extended) generalized linear models to the case of circular covariates. The estimation of the conditional mean, or a transformation of it, is carried out nonparametrically, by maximizing the circular local likelihood. The methodology can be extended to joint estimation of conditional mean and dispersion. Among the important issues are to study bias and variance of the local estimators, and a data-driven selection of the concentration parameter(s). The generality and practical use of the approach is illustrated with several real-data examples.
Collegamento con Microsoft Teams
Organizzatore
Sabrina Mulinacci