Reclaiming independence in spatial-clustering datasets: A series of data-driven spatial weights matrices

Area of Science:

• Spatial statistics
• Epidemiology
• Environmental science

Background:

• Traditional spatial models use weights matrices (W) based on the first law of geography.
• Spatial dependence often exhibits clustering or geographic discontinuity in epidemiological and environmental data.
• First-law-of-geography-based W may lead to inaccurate estimations and reduced statistical power in clustered spatial data.

Purpose of the Study:

• To propose data-driven weights matrices (DDWs) that incorporate spatial clustering.
• To evaluate the performance of DDWs in spatial autoregressive and conditional autoregressive (CAR) models.
• To compare DDWs with existing methods for disease mapping with clustered spatial dependence.

Main Methods:

• Developed data-driven weights matrices (DDWs) based on scan statistic identified spatial patterns.
• Applied DDWs to spatial autoregressive and Leroux-prior-based conditional autoregressive (CAR) models.
• Evaluated model performance using simulations and case studies, comparing with classic W and density-based clustering models.

Main Results:

• DDWs demonstrated considerably better performance than classic W in datasets with clustered spatial dependence.
• DDWs incorporated into CAR models showed significant advantages, particularly for common diseases.
• DDWs effectively address spatial discontinuity and clustering in spatial analysis.

Conclusions:

• Data-driven weights matrices (DDWs) offer a superior approach for spatial modeling with clustered or discontinuous spatial dependence.
• DDWs enhance accuracy and statistical power in epidemiological and environmental studies.
• The proposed DDWs provide a valuable tool for disease mapping and risk estimation in complex spatial patterns.