Indicator Selection: Construction of Geographic Factor Analysis Model

This chapter focuses on indicator selection and constructs the GeoFA model to address the spatial defects of the traditional FA model in TSUE research. It first analyzes the core logic of the traditional FA model and its prominent spatial blind spots, namely neglecting spatial dependence and spatial heterogeneity. Then, it elaborates on the theoretical connotation of geospatial effects and the theoretical challenges posed by spatial non-stationarity to traditional models. On this basis, two technical paths for the GeoFA model construction are proposed: one based on the geographically weighted correlation coefficient matrix and the other on the geographically weighted spatial similarity matrix. Their mathematical expressions and weight matrix construction methods are also provided. Finally, multi-model comparative experiments are conducted to evaluate the two proposed paths. The GeoFA model based on the geographically weighted correlation coefficient matrix is selected as the optimal model. This model provides scientific and technical support for indicator dimensionality reduction in TSUE evaluation.