2022 年 11 巻 Supplement 号 p. PP05_p16-PP05_p19
In this study, we use a generalized lasso to fit spatially varying coefficient models to the case of predictor variables with both numerical and categorical scales. We constructed the generalized lasso model with two 𝐿1 penalties: one was to link some categories within one categorical predictor, and the other was to link a corresponding categorical predictor between adjacent regions. Then, we applied the proposed method to a province-wise analysis of house sales price data on Java Island, Indonesia, to the original data without concatenation between categories for each categorical predictor. An optimal model is obtained in which categories and regions with the same effect are pooled.
