We analyzed migration from large cities to rural areas in Japan. The purpose of this study is to solve the problem of multicollinearity, which has been a problem when using regression models such as gravity models, and to analyze as many socioeconomic variables of cities as possible.
This study showed the following two points. First, by using the elastic net regression for the analysis with the gravity model, it is possible to select significant variables while simultaneously dealing with strongly correlated variables. Furthermore, the basic variables of the gravity model, which were population size and interregional distance, were also selected, and the absolute values of the regression coefficients were larger than the other variables. Second, when all 39 explanatory variables used in previous studies were submitted to the elastic net regression, 21 variables were selected as being associated with the migration.
Further analyses using more variables, including panel data analysis, are needed to clarify the factors that influence migration from large cities to rural areas.