Covariates
Our event-history analysis model has four general groups of covariates: (1) local rice prices; (2) household landholding net of number of kin household members; (3) time periods; and (4) the current age and residing village of each index individual, which are always included in the model as control variables.
In the analyses of adult and old-age mortality as well as in the analyses of out-migration of native males and females, marital status of each index individual is also controlled for.As mentioned at the outset of the chapter, we use annual rice prices in the local market of Aizu to measure short-term economic stress. In our earlier study of mortality in the two villages (Tsuya and Kurosu 2000a), we examined the effects of different rice price series—prices in Aizu and in the central market of Osaka—using different specifications: raw prices, de-trended prices based on moving averages, and prices de-trended using the Hodrick-Prescott filter.24 We found that fluctuations in agricultural output in the two northeastern villages were best measured by raw prices (with or without logarithm), rather than de-trended prices. This study therefore uses the log of raw rice price. To further examine the nature of effects of rice price variations, we also test the effects of prices time-lagged by up to five years. Because of strong auto-correlation among annual rice prices, our model includes rice price for only one year at a time.
Household landholding (mochidak^a in 4%//) is a continuous variable measuring the total productive capacity of the land held by the household. This variable is introduced primarily to measure the amount of household economic resources, though this variable may also be a proxy for the wealth and socio-economic status of household.25 The demographic effects of landholding are estimated, net of number of kin household members.
Total household size, including servants and non-kin, was found in one of our earlier studies to be highly multicolinear with landholding since, given the under-mechanized agricultural technologies of the time, large landholding often required several helping hands (Tsuya, Kurosu, and Nakazato 1997). Theoretically, it is also appropriate to control for the number of co-resident kin, rather than total household size, as non-kin and servants are thought to have had much less access, if any, to the resources and wealth of the households they served.Our model also takes into account the demographic effects of different time periods within the entire 154-year period (1716-1870) under consideration. As explained in the previous section, there are considerable temporal differences in local economic development, policy contexts, and environmental conditions in the domain and in northeastern Japan as a whole during the eighteenth and nineteenth centuries. First, the commercialization of agriculture based on smallscale family
farming with high labour intensity began on a large scale in the middle of the eighteenth century (Hayami 1985: 91—2). At around the turn of the century, the domain government reversed its official policy and began encouraging production of cash crops, resultingin proto-industrialization in many agrarian villages in the domain (Nagata, Kurosu, and Hayami 1998). Finally, in the late 1830s (after the Tempo famine), the rigid social structure and stratification of the Tokugawa regime started to crumble, leading to the period called bakumatsu, the last years of the Tokugawa period. Therefore, we divide the 154-year period into four sub-periods: before 1760, 1760-99, 1800-39, and 1840-70.
Current age of an index individual is a covariate the effect of which we need to control, because demographic behaviour is a function of age. This variable consists of 5-year age groups, except for ages 1-4, at which mortality is thought to be high. Because the data from the two different villages are pooled, a dichotomous variable indicating whether index individuals were residing in Shimomoriya or not is also included in the model. Marital status is included in our models for adult and old-age mortality as well as in the models for out-migration of native males and females. This covariate consists of five categories: never married, currently married, widowed, divorced, and marital status unknown.
5.