Accommodating Spatial Heterogeneity in Geographically Weighted Regression with Group Penalty.

Area of Science:

• Statistics
• Spatial Analysis
• Health Economics

Background:

• Analyzing multi-location survey data is increasingly common.
• Existing methods often fail to account for spatial heterogeneity, leading to suboptimal results.
• There is a need for methods that can perform variable selection and accommodate spatial variations.

Purpose of the Study:

• To develop a novel statistical approach, geographically weighted group lasso regression (GWGPL), for analyzing multi-location survey data.
• To effectively address spatial heterogeneity and perform shared variable selection across different geographic locations.
• To rigorously prove the selection and estimation properties of the proposed GWGPL method.

Main Methods:

• Development of the geographically weighted group lasso regression (GWGPL) model.
• Theoretical analysis to prove the selection and estimation properties of GWGPL.
• Application of GWGPL to Chinese social survey data ('One Thousand People, One Hundred Villages').

Main Results:

• GWGPL effectively accommodates spatial heterogeneity and performs shared variable selection.
• Simulation studies show GWGPL outperforms alternative methods in competitive advantages.
• Analysis of health survey data identified key variables associated with inpatient, outpatient, and self-treatment costs.

Conclusions:

• The proposed GWGPL approach offers a robust method for analyzing spatially heterogeneous data.
• GWGPL demonstrates superior performance in terms of grouping structure, coefficient estimation smoothness, and prediction accuracy.
• The identified variables provide insights into factors influencing healthcare costs in the studied population.