空間的影響の分解：州レベルの健康アウトカムにおける方法論的実証と再分析

While spatial autoregressive (SAR) models are increasingly used in population-level psychological studies, researchers often overlook the crucial step of parsing effects into direct, indirect and total impacts, a standard practice in spatial econometrics. In this paper, we demonstrate the necessity of this practice by re-analyzing Gruda et al.'s (2024) U.S. Dark-Triad and health dataset with heteroskedasticity-robust SAR models and full impact decomposition, revealing significant changes. The previously observed direct protective effect of state-level narcissism on hypertension mortality disappeared when accounting for interstate spillovers. Conversely, the association with lower cancer prevalence and depression strengthened. Several health-behaviour findings reversed direction, indicating naïve regressions conflated within- and between-state effects. Machiavellianism and psychopathy coefficients also shifted. These results demonstrate that spatial spillovers can dilute, negate or reverse local effects, cautioning against policy inferences based solely on direct estimates.