A stochastic neighborhood conditional autoregressive model for spatial data.

Area of Science:

• Spatial statistics
• Geostatistics
• Statistical modeling

Background:

• Conditionally autoregressive (CAR) models are standard for spatial processes.
• CAR model neighborhoods are typically deterministic, based on distances or boundaries.
• Limitations exist in deterministic neighborhood selection for complex spatial relationships.

Purpose of the Study:

• To propose an extension of the CAR model called the Stochastic Neighborhood CAR (SNCAR) model.
• To introduce parameter-dependent neighborhood selection for greater flexibility.
• To improve the estimation of covariance structures in spatial data.

Main Methods:

• Developed the Stochastic Neighborhood CAR (SNCAR) model.
• Incorporated unknown parameters to define neighborhood selection probabilistically.
• Applied the SNCAR model to simulated and real-world spatial datasets.

Main Results:

• The SNCAR model demonstrates flexibility in estimating covariance structures.
• Accurate estimation was shown across various spatial covariance models.
• The model effectively handles data with complex spatial dependencies.

Conclusions:

• The SNCAR model offers a more adaptable approach to spatial modeling than traditional CAR models.
• Parameter-driven neighborhood selection enhances the model's ability to capture intricate spatial correlations.
• The proposed model shows promise for analyzing diverse spatial datasets, including environmental contamination data.