Bayesian semiparametric model with spatially-temporally varying coefficients selection

Area of Science:

• Statistics
• Spatiotemporal Analysis
• Bayesian Modeling

Background:

• Covariate effects in spatiotemporal analysis often vary across geographic areas and time.
• Traditional models may face estimation biases due to normality assumptions on spatially varying coefficients.
• Spatial configurations can be influenced by random intercepts and spatially varying coefficients.

Purpose of the Study:

• To propose a flexible Bayesian semiparametric space-time model for analyzing complex covariate effects.
• To address limitations of existing models, including normality assumptions and potential biases.
• To incorporate variable selection for spatially-temporally varying coefficients.

Main Methods:

• Developed a Bayesian semiparametric space-time model with coefficients decomposed into fixed, spatially varying, and temporally varying components.
• Utilized area-specific Dirichlet process priors for nonparametric modeling of spatially varying coefficients.
• Employed a dynamic model for temporally varying coefficients and a variable selection procedure for coefficient inclusion uncertainty.

Main Results:

• The proposed semiparametric model demonstrated improved performance compared to standard Bayesian spatial-temporal models with normality assumptions.
• Outperformed models relying solely on Dirichlet process priors for the random intercept.
• Simulation studies confirmed the effectiveness of the new approach.

Conclusions:

• The Bayesian semiparametric space-time model offers a more robust and accurate approach for spatiotemporal data analysis.
• The model effectively handles varying covariate effects and addresses limitations of conventional methods.
• The approach provides valuable insights, as illustrated by the low birth weight data application.