Online Program Home
  My Program

All Times EDT

Abstract Details

Activity Number: 55 - Advances in Bayesian Sparse Regression
Type: Contributed
Date/Time: Monday, August 3, 2020 : 10:00 AM to 2:00 PM
Sponsor: Section on Bayesian Statistical Science
Abstract #313660
Title: Global-Local Shrinkage Priors for Non-Gaussian Continuous-Valued Data
Author(s): Hongyu Wu* and Jonathan Bradley
Companies: Florida State University and Florida State University
Keywords: Bayesian; Spike and slab; Markov Chain Monte Carlo; Non-Gaussian
Abstract:

Sparsity inducing priors have been widely used for Bayesian modeling of Gaussian data. However, the use of these priors in skewed and heavy-tailed data settings is less developed. In this paper, motivated by the horseshoe prior, we propose a new continuous global-local shrinkage prior suitable for the case that the data model is non-symmetric. Specifically, we consider modeling the data according to the multivariate logit-beta distribution, which is a special case of the conjugate multivariate distribution. We show that this new specification can be considered as an extension of the horseshoe model. Furthermore, our approach involves an easy-to-implement Gibbs sampler. The simulation study suggests a superior performance of our proposed prior in a standard design setting against the horseshoe prior. Additionally, we conduct a statistical analysis of cloud fraction data from NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) instrument to demonstrate the effectiveness of the proposed prior.


Authors who are presenting talks have a * after their name.

Back to the full JSM 2020 program