Statistical analysisFirst, for descriptive observations and screening purposes, distributions of the selected variables were studied using histograms. Log-transformation was performed for the skewed variables of population density and unemployment rate. Then, correlation coefficients between DFLE65s of men and women and independent variables were calculated. For the correlation analyses, Pearson's product moment correlation coefficient or Spearman's rank correlation coefficient were used for the independent variables of normal or non-normal distribution, respectively. An independent variable with one absolute value of the correlation coefficient of 0.20 or more was selected as a factor that was potentially associated with the DFLE65s. The variables that were insignificant or irrelevant with regard to DFLE65 were excluded (e.g., the percentage of paved roads or electric energy consumption). Further, the linearity of each association with independent variables and DFLE65s were checked by analyzing scatter charts, and its heteroscedasticity was assessed by observing a residual plot. Finally, multivariate linear regression analyses were performed using the 24 selected variables. The proportion of older people (aged 65 years or more) and the population density (log-transformed) were included in the regression models as potential confounding factors. These data were included because it could theoretically affect other explanatory variables although the correlation between the two indicators and DFLE65s was not significantly strong. In fact, many explanatory variables were associated with them (e.g., Pearson's correlation coefficients of the number of public health nurses (PHNs) per 100,000 population were -0.81 for log-transformed population density and 0.80 for the proportion of older people).In Pearson's correlation analysis and regression analysis, Aomori prefecture data was excluded as an outlier because its DFLE65s for both men (14.0 years) and women (17.3 years) were lower than that of the average of 47 prefectures (mean standard deviation [SD] = 15.2 0.4 in men and 18.5 0.4 in women) by three SDs. In addition, residual analyses showed that the inclusion of Aomori prefecture data in the regression analysis deteriorated the linearity and equality of variance of the regression between DFLE65s and explanatory variables.Self-reported health status was determined based on the question, "What do you think of your current health status?" in a Comprehensive Survey of the Living Conditions of People on Health and Welfare. 21 The following five alternatives were provided as answers: good, relatively good, normal, relatively bad, and bad. In our study, the proportion of good self-reported health that was obtained by combining the proportion of "good" and "relatively good" self-reported health was used for analysis.All the variables used contained no missing data. All p values are two tailed. All data analyses were performed using the SAS