Demographic, economic, and disaster data for Italy (1820–1962) and Japan (1671–1965) are used. Data from many countries were initially gathered, but Italy and Japan were chosen for the final analysis based on the following factors: the availability of region-level data, the prevalence of natural disasters, and the need for a country each from Europe and Asia.3
The choice of two natural disaster-prone countries, one from Europe and one from Asia, enables one to build upon Jones’ (1981
) theory that differences in fertility behavior in Asia and Europe were a result of differences in the prevalence of natural disasters—not differences in culture, society, history, or politics—and, therefore, that any other society subjected to the volatility of the Asian environment would have responded in a similar manner. The data sources used for Italy and Japan are detailed in Appendixes 1
Demographic data were compiled from various print sources or extracted from the Princeton European Fertility Project online demographic data set (Treadway 1980
). These variables include the following measures of fertility: crude birth rates, the index of total fertility (If), the index of marital fertility (Ig), and the index of non-marital fertility (Ih).4
They also include crude death rates. The print sources for Italy’s demographic data are Livi-Bacci (1977
) and del Panta (1979
). The print sources for Japan’s demographic data are Hanley and Yamamura (1977
); Morris and Smith (1985
); Smith (1977
); Jannetta and Preston (1998
Annual real wage data were from Jeffrey Williamson: the Italian wage index (1900 = 100) was taken from Williamson (1995
) while the Japanese wage index (1934–1936 = 100) was provided in digital form. This wage data was used to calculate, for each year, the variance of the detrended wage over the past 20 years prior to and including that year in the base case.5
Whenever possible, the variables for both the level and the variance of the wage were averaged over the same years over which observations for the demographic dependent variable spanned in any given regression.6
All averages were centered on the floor of the midpoint of the years covered in the average.
Daily natural disaster data on the number and magnitudes of earthquakes, tsunamis, and volcanos were taken from the National Geophysical Data Center web site (Dunbar et al. 1999
; Lockridge 1999
; Whiteside 1999
). In order to measure the severity of the earthquake, either the Modified Mercalli Intensity Scale of 1931, which takes on integer values from 1 (least intense) to 12 (most intense); or the geophysical magnitude, where each increase in magnitude represents a 10 fold increase surface wave amplitude,7
is used. The magnitude of a tsunami is defined as the log of twice the maximum runup height of the wave; this value is incremented by 2, so that the minimum value is 1 instead of –1.8
For the volcano magnitude, the volcano explosivity index is incremented by 1, so that it takes on integer values from 1 (tephra volume = 1E4, column height <0.1 km above crater) to 9 (tephra volume ≥1E12, column height >25 km above sea level). For all types of disaster, a magnitude or intensity of 0 denotes no disaster.
From the raw natural disaster data, variables are constructed for each year in the data set for (1) the number of each type of disaster over the past 20 years prior to and including that year, and for (2) the sum of the geophysical magnitudes of all occurrences of earthquakes, tsunamis, and volcanos over the past 20 years prior to and including that year. For example, for the 1880 observation, environmental shocks are aggregated over 1861–1880. When multiple observations existed for any given disaster, an average of the magnitudes reported was taken.
In the base case, both the measure of environmental shock and the measure of short-run economic volatility consider events that have occurred over the previous 20 years. This is because individuals who are making choices about marriage, childbearing, and potential tradeoffs with their careers do so in their early twenties,9
and, therefore, have experienced the previous two decades of environmental disasters and economic instability. As a robustness check, the previous 10 years is also used to measure short-run instability.10
The demographic, economic, and natural disaster data are used to construct a panel data set for Italy and a panel data set for Japan.
For the Italian panel, the unit of observation is a region. There are 18 regions in Italy; following del Panta (1979
), these regions can be grouped into five geographical areas: Northwest, Northeast, Center, South, and Islands. The panel covers all 18 regions and spans the years 1820–1962. While measures of fertility were available at the region level, the measures of mortality compiled were at the level of the geographical area.11
Thus, for each geographical area, the area’s values for the crude death rates were used for all regions in the area. The same national Italian wage data, which reflect mainly the wages in Northern Italy, were applied to all regions. Volcano data were matched to the region in which the volcano was located. All other natural disaster data were matched to the region based on the location(s) named in the source, or, when the location name could not be located in any region, on the latitudes and longitudes given. Disasters that occur in areas larger than a region are assumed to strike all regions in that area; disasters that occur in areas smaller than a region are assumed to affect the entire region.
In the panel data set for Italy, the fertility variables generally each span 3 years per observation. For example, for the fertility indices, there are up to 11 observations of each fertility index for each region: (1) 1862–1866, (2) 1870–1872, (3) 1880–1882, (4) 1890–1892, (5) 1900–1902, (6) 1910–1912, (7) 1921–1926, (8) 1930–1932, (9) 1935–1937, (10) 1950–1952, and (11) 1960–1962. Of the 18 regions, 14 regions have observations during 1862–1866, and 16 regions have observations during 1870–1912; all the 18 regions have observations from 1921–1962. For the crude birth rate data, there are several more observations over other 3-year intervals for some of the regions, so the crude birth rate data extends back to 1820.
Because the fertility data spans mostly 3 years per observation, the natural disaster variables, which represent the sum of either the number or intensity of natural disasters over the last 10 or 20 years, are averaged over 3 years. Similarly, the real wage and the variance of the wage (which is the variance in the detrended wage over the past 10 or 20 years, divided by 1000) are averaged over 3 years as well. Because the wage data begin in 1871, and because at least 10 prior years of wage data are needed to form the variance of the wage, the regressions that include the variance of the wage span the years 1880–1962. Geographical area-level crude death rates span 4–5 years per observation. All the variables spanning multiple years are centered on the floor of the midpoint of the years covered in the average. For example, for each region in Italy, a birth rate that spans the three years 1880–1882 is coupled with the average of the natural disaster variable for the three 20-year ranges 1861–1880, 1862–1881, and 1863–1882, and the observation is centered on the floor of the midpoint of the years covered in the average, which in this case is the year 1881, the midpoint of 1880–1882. See Appendix 1
The Japan panel consists of data from 13 mura (villages), one shi (city) and one han (domain), and spans the years 1671–1965. For lack of a better term, the level at which an observation applies—whether it be a village, city or domain—is called a “place.” Observations are available for different years for each place, and each unit of observation covers from 1 to 41 years depending on the place. For example, for the domain, Morioka, there are 15 1-year observations: one every 10 years from 1680 to 1790, and also the years 1803, 1828, and 1840. The observations for the other places cover different years. Each of the places can be matched to one of the 11 regions in Japan. Regressions include either place or region fixed effects.
The same national Japanese wage data were applied to all places; the data were constructed by Jeffrey Williamson from pre-1930 wage data for Kyoto and Kamikawarabayashi (near Osaka) and post-1930 wage data for Tokyo. The wage data span the years 1727–1938. Volcano data were matched to the prefecture in which the volcano was located. All other natural disaster data were matched to the prefecture based on the location(s) named in the source, or, when the location name could not be located in any prefecture, on the latitudes and longitudes given. Disasters that occur in areas larger than a prefecture are assumed to strike all prefectures in that area; disasters that occur in areas smaller than a prefecture are assumed to affect the entire prefecture. All the places are assumed to be affected by the disasters matched to the prefecture(s) in which they are located.
For the panel data set for Japan, crude birth rates span 1–41 years per observation, with an average span of approximately 10 years. Consequently, the natural disaster variables, which represent the sum of either the number or intensity of natural disasters over the last 20 years, are averaged over 10 years. Similarly, the real wage and the variance of the wage (which is the variance in the detrended wage over the past 10 years, divided by 1000) are averaged over 10 years as well. Because the wage data span the years 1727–1938, and because at least 10 prior years of wage data are needed to construct the variance of the wage variable, the regressions that include the variance of the wage span the years 1737–1935. Crude death rates span the same number of years as the crude birth rate for each observation. All variables spanning multiple years are centered on the floor of the midpoint of the years covered in the average. See Appendix 2
According to the summary statistics in Table , the mean crude birth rate for the Italian regions over 1820–1962 was 30.31 births per 1000 people, while the mean crude birth rate for Japan over 1671–1965 was 28.79 births per 1000 people. On average, in any given 20-year period, there were 4.98 earthquakes and 0.69 volcano eruptions in the Italy regions over 1820–1962; further there were 0.65 earthquakes, 0.35 tsunamis, and 0.65 volcano eruptions in Japan over 1671–1965.