A0373
Title: Flexible Bayesian quantile regression for semiparametric geoadditive mixed models
Authors: Jiang Xuejun - Souther University of Science and Technology (China) [presenting]
Abstract: A geoadditive mixed regression model is built under the flexible Bayesian quantile regression framework. Instead of considering the asymmetric Laplace distribution as residual distribution, the error term is assumed to be distributed as an infinite mixture of Gaussian densities. A stochastic constraint is imposed to ensure inference on the quantile of interest. Based on that, a Bayesian version of semiparametric geoadditive mixed regression model is proposed, where both one-dimensional curve and two-dimensional surface fitting for modeling interactions are developed by using P-splines. In addition, unobserved heterogeneity is incorporated as spatial random effects to account for variation among different regions. The main advantage of our method is that it not only allows for the full span of quantile-restricted error distribution for Bayesian quantile regression with more flexibility and feasibility, but also can describe the large-scale geographic trend along with the local spatial correlation. Finally, a data set of earthquake in Mainland China is used to illustrate the usefulness and effectiveness of the proposed methodology.
