A flexible Bayesian nonconfounding spatial model for analysis of dispersed count data.

Area of Science:

• Biostatistics
• Spatial Statistics
• Epidemiology

Background:

• Spatial regression models for count data face challenges with covariate collinearity and data dispersion (over/under-dispersion).
• Classical Poisson models are often inappropriate for dispersed count data, leading to potential inference errors.
• Existing methods for spatial confounding may not adequately address dispersion issues.

Purpose of the Study:

• To propose a flexible Bayesian hierarchical modeling approach for spatial count data.
• To integrate nonconfounding spatial methodology with dispersed count modeling derived from renewal theory.
• To provide a robust framework for analyzing count data with both spatial effects and dispersion.

Main Methods:

• Developed a Bayesian hierarchical model incorporating a gamma distribution for waiting times.
• Formulated the model as a latent Gaussian model for computational efficiency.
• Utilized the integrated nested Laplace approximation (INLA) for fast computation.
• Extended existing spatial count data analysis methodologies.

Main Results:

• The proposed model effectively controls for spatial confounding and handles data dispersion.
• Comparison of different spatial confounding approaches in the presence of dispersion was performed.
• Simulation studies demonstrated the merits of the proposed Bayesian approach.

Conclusions:

• The flexible Bayesian hierarchical model offers a robust solution for analyzing spatial count data with dispersion.
• The method provides more reliable inferences compared to classical approaches when dealing with complex count data.
• The approach is validated through real-world applications in crime and cancer incidence studies.