144
H. Matsuura
8.2.6 Data Analysis: Empirical Strategy
In this chapter, I estimate the following regression.
Migration Rate m
= β 1 ∗ High RC I m + β 2 ∗ Low RC I m + Z γ
+ θ p + ε m
(1)
where the dependent variable is the in- and out-migration rate (%) as of 1 January
2013 in municipality m in prefecture p. Explanatory variables include a dummy
variable for high RCI, a dummy variable for low RCI as well as Z, which is a set
of other control variables that includes logged per capita income, logged number
of schools per square kilometre and logged population density. θ p is the prefecture
fixed effects.
In addition to the specification above, I test whether high (low) resilience affects
differently in municipalities affected and not affected by an earthquake. I estimate
the following regression.
Migration Rate m
= δ 1 ∗ (D I S m ∗ High RC I m ) + δ 2 ∗ (D I S m
Low RC I m ) + δ 3 ∗ D I S m + β 1 ∗ High RC I m + β 2
∗ Low RC I m + Z γ + θ p + ε m
(2)
where DIS is a dummy variable for earthquake-affected municipalities (at least one
earthquake-related death). I am particularly interested in the sign of δ 1 and δ 2 , which
are the coefficient of the interaction of dummy variables for earthquake-affected
municipalities with high RCI and low RCI, respectively.
8.3 Results
8.3.1 Level of Resilience Capacity Index in Japan
Table 8.1 shows summary statistics of the inputs for each of the 12 indicators used to
calculate the RCI. The single variable with the most missing observations was voter
participation rate because there were fewer candidates than seats available, thus no
election was held in these municipalities. Income equality, regional affordability and
homeownership data also contain a relatively large number of missing observations
since the Housing and Land Survey of 2013 excludes towns and villages with a
population of 15,000 or less. The average out-migration rate per Japanese population
H. Matsuura
8.2.6 Data Analysis: Empirical Strategy
In this chapter, I estimate the following regression.
Migration Rate m
= β 1 ∗ High RC I m + β 2 ∗ Low RC I m + Z γ
+ θ p + ε m
(1)
where the dependent variable is the in- and out-migration rate (%) as of 1 January
2013 in municipality m in prefecture p. Explanatory variables include a dummy
variable for high RCI, a dummy variable for low RCI as well as Z, which is a set
of other control variables that includes logged per capita income, logged number
of schools per square kilometre and logged population density. θ p is the prefecture
fixed effects.
In addition to the specification above, I test whether high (low) resilience affects
differently in municipalities affected and not affected by an earthquake. I estimate
the following regression.
Migration Rate m
= δ 1 ∗ (D I S m ∗ High RC I m ) + δ 2 ∗ (D I S m
Low RC I m ) + δ 3 ∗ D I S m + β 1 ∗ High RC I m + β 2
∗ Low RC I m + Z γ + θ p + ε m
(2)
where DIS is a dummy variable for earthquake-affected municipalities (at least one
earthquake-related death). I am particularly interested in the sign of δ 1 and δ 2 , which
are the coefficient of the interaction of dummy variables for earthquake-affected
municipalities with high RCI and low RCI, respectively.
8.3 Results
8.3.1 Level of Resilience Capacity Index in Japan
Table 8.1 shows summary statistics of the inputs for each of the 12 indicators used to
calculate the RCI. The single variable with the most missing observations was voter
participation rate because there were fewer candidates than seats available, thus no
election was held in these municipalities. Income equality, regional affordability and
homeownership data also contain a relatively large number of missing observations
since the Housing and Land Survey of 2013 excludes towns and villages with a
population of 15,000 or less. The average out-migration rate per Japanese population