We created panel data for 47 prefectures between 2001 and 2014. The number of observations was 658. We limited our data coverage until 2014 because of data availability. Using the panel data, we tested the first hypothesis that the larger amount of spending by subnational governments was correlated with the lower suicide rates in their jurisdictions by estimating the following model:
[S]jt = α[E]jt + β[U]jt + λwjt + μjT + φt + ρj + εjt (1)
where the outcome variable [S]jt is a natural log of the suicide rate per 100,000 individuals in the year t in prefecture j. Suicide includes all deaths classified as X60-X84 under ICD-10. Considering the possibility that the effects of government expenditures varied by age and sex, we generated the suicide rates for six subpopulation groups: (1) males aged 20-39, (2) males aged 40-64, (3) males aged 65 and over, (4) females aged 20-39, (5) females aged 40-64, and (6) females aged 65 and over. The suicide data were calculated by using data from the Vital Statistics [26].
Our primary explanatory variable is [E]jt, which denotes the per-capita government expenditures in prefecture j in the year t. We used the sum of expenditures of the prefectural government in j and all municipal governments in j from the Annual Report of Local Public Finance [24]. Japan is divided into 47 prefectures, and the number of municipalities in each prefecture ranges from 15 to 179. Both prefectural and municipal governments can use their expenditures on social welfare, public health, employment-related issues, public works, education, and disaster reliefs. We used the total amount of spending, rather than spending specifically for social welfare and public health because other types of local government spending can affect people's well-being. For example, spending on infrastructures would produce job opportunities for the unemployed and may improve their economic and mental well-being. We transformed the per-capita amount, adjusted for inflation, into a natural log for estimation.
[U]jt in equation (1) refers to the percentage of unemployed people in prefecture j in year t. Building on recent research on the same topic [21,22,27,28], we used the unemployment rate as a measure of recession. The data were obtained from Statistics Japan [29].
Further, wjt refers to the socioeconomic characteristics of each prefecture in each year, all of which were likely to affect both the suicide rate and the government’s expenditures. Specifically, included in wjt are income per capita, fiscal strength index, population size, and percentages of the dependent population aged under 14 and 65 and over, in each prefecture and year. Income per capita, obtained from the System of National Accounts, was defined as the total amount of income in prefecture j in the year t divided by the population size [30]. The financial strength index measures the fiscal conditions of each prefecture each year. The index exceeds 1 if the amount of revenue coming from the prefecture’s financial sources exceeds the amount of fiscal demand and falls below 1 otherwise. This index is used to determine the amount of money transferred from the national to the local government. Because there were considerable year-to-year fluctuations, the values were averaged over the past three years. The data were obtained from the Annual Report of Local Public Finance [24]. The population size and the percentages of the dependent population were obtained from the Annual Resident Registers [30]. We used natural logs of the total population and income per capita in our regression analysis.
Finally, φt in equation (1) represents the year fixed effects, while ρi represents the prefecture fixed effects unique to each prefecture. The year fixed effects allowed us to control for the effects of annual socioeconomic and political changes at the national level, such as the effects of macroeconomic policies and business cycle that might affect the entire country. The prefecture fixed effects allowed us to control for the effects of time-invariant characteristics of the prefectures, such as the effects of culture related to suicide, and climate and geographic conditions. The inclusion of the prefecture fixed effects in equation (1) means that the model used variations in the level of government expenditures over time within each local government. We also added the prefecture-specific linear time trends, μjT, to the model to control for the effects of linear trends in the suicide rates unique to each prefecture.
To test the second hypothesis that the negative relationship between local government spending and the suicide rates was particularly robust during a more severe recession, we included the interaction term between [E]jt and [U]jt in the model as:
[S]jt = α[E]jt+β[U]jt +γ[E]jt*[U]jt + λwjt+μjT + φt + ρj + εjt . (2)
Because the marginal effect of local government expenditures was hypothesized to change as the unemployment rate increased, we calculated and plotted the marginal effect of [E]jt and its confidence interval at the different values of the unemployment rate.