Spatial Difference-in-Differences with Bayesian Disease Mapping Models.

Area of Science:

• Epidemiology
• Biostatistics
• Spatial Analysis

Background:

• Bayesian disease-mapping models account for spatial correlation in small-area epidemiology.
• Difference-in-differences (DID) methods are common for estimating treatment effects but often ignore spatial dependence.
• Small-area evaluations require methods that address both spatial structure and causal inference.

Purpose of the Study:

• To integrate Bayesian disease-mapping models into a difference-in-differences (DID) framework for small-area evaluations.
• To address spatially structured residual variation and enhance precision in causal effect estimation.
• To develop a method enabling causal identification equivalent to fixed effects DID while incorporating spatiotemporal random effects.

Main Methods:

• An imputation-based DID framework incorporating Bayesian disease-mapping models.
• Utilized two-way Mundlak estimation for causal identification.
• Employed Integrated Nested Laplace Approximation (INLA) for efficient Bayesian computation with flexible spatiotemporal structures.

Main Results:

• The integrated approach improves precision and interval coverage compared to standard DID methods when spatiotemporal structure is correctly specified.
• Simulations demonstrate the enhanced performance of the proposed Bayesian disease-mapping DID framework.
• The method was successfully illustrated evaluating local ice cleat distribution programs.

Conclusions:

• Integrating Bayesian disease mapping with DID methods offers a powerful tool for small-area causal inference.
• The developed framework enhances the precision and reliability of treatment effect estimates in spatially correlated data.
• This approach advances causal panel data methods for epidemiological and public health research.