2019 Volume 75 Issue 4 Pages 262-272
This paper introduces a two-stage linear modeling for coarse regional population distribution data on top of the group-theoretic spectrum analysis method by Onda et al.1) In the first step, the proposed method employs Lasso estimator to select a set of a small number of explanatory variables. In the second step, a hierarchical modeling-based variation partitioning combined with a significance test is introduced to rank the explanatory power of the selected variables. Employing actual population data of Stuttgart, Köln, Atlanta, and Dallas, we demonstrate effectiveness of the proposed method.