Pointless spatial modeling.

Area of Science:

• Spatial statistics
• Epidemiology
• Social sciences

Background:

• Area-level aggregated data analysis is prevalent in epidemiology and social sciences.
• Markov random field models are typically used for spatial dependence but have ad hoc neighborhood definitions for irregular areas.

Purpose of the Study:

• To develop a continuous spatial modeling approach for aggregated data that overcomes limitations of discrete area-based models.
• To enable reconstruction of continuous underlying surfaces and handle supplementary point data.

Main Methods:

• Utilizing stochastic partial differential equations for continuous spatial modeling.
• Implementing Bayesian inference with integrated nested Laplace approximations for linear links.
• Employing Hamiltonian Monte Carlo and empirical Bayes (Laplace approximations) for nonlinear links.

Main Results:

• The continuous spatial modeling approach provides a data-driven smoothing mechanism for irregular areas.
• The method allows for the reconstruction of continuous surfaces, though interpretation requires careful consideration of data quality and configuration.
• The model effectively handles both aggregated and supplementary point data.

Conclusions:

• Continuous spatial modeling offers a robust alternative to traditional discrete area models, particularly for irregular spatial units.
• The proposed Bayesian inference methods provide efficient implementations for complex spatial models.
• The approach is validated through simulation and applied successfully to real-world epidemiological data (Scottish lip cancer).