出生力の地域格差の要因分析 : 非定常性を考慮した地理的加重回帰法による検証

Our study focused on regional differences in fertility from the viewpoint of the spatial effect on fertility behavior, and re-examined previous research by using regression analyses that take account of spatial autocorrelation. More specifically, we applied geographically weighted regression to assess heterogeneity of the relationship between regional fertility rates and their covariates. Our analytical samples are 2311 towns and villages in Japan based on 2005 administrative boundaries. We used total fertility rate calculated based on vital statistics (Bayesian estimates) in 2005 as a dependent variable. Independent variables include socio-economic condition, female labor force participation, political measures on child care, and household structure that come from a database based on census. Our result suggests that residuals of the global model using ordinary least squares show strong spatial autocorrelation, meaning that statistical inference may be unreliable. Based on the result from this global regression analysis, we attempted to examine spatial variations in the coefficients by estimating geographically weighted regression model. The result suggests that most of coefficients for covariates have statistically significant geographical variations, and in some regions, sign shifts in the opposite direction from what it is in the global model. We conclude that fertility response to external forces may vary across regions because of their historical and geographical settings, and results of the global model may not be appropriate to uniformly apply for each region. Our result also suggests that policy measure should be flexibly carried out reflecting unique regional conditions.