Two Essays in the Analysis of a Demographic System: France 1740-1909.

Abstract

This thesis consists of two interrelated essays based in time series analysis of French vital events for the period 1740-1909. Classical time series analysis decomposes the variance of a series into fluctuations, cycles and trend. We analyze two of these components--short run fluctuations and long cycles--in an application to historical French vital events. The first essay treats short run fluctuations. The second focuses on fertility and examines long cycles. Fertility, mortality and nuptiality form a simultaneous system which interacts with economic and meteorological conditions. In the short run we can take the demographic variables as endogenous, and economic and meteorological conditions as exogenous and thus formulate a closed system. Our principal indicator of economic conditions is the price of wheat, the principal food crop. While weather affects the outcome of the harvest, different aspects of weather affect the health and vital events of a population than affect the harvest. We therefore neglect the effects of weather on the harvest. Biometric models of fertility and empirical research on the biologically-based interrelations of fertility and mortality provide insight into the delays inherent in the birth interval and the repercussions of these delays in the demographic system. We combine the results from these biometric models with empirical research on the effects of nutrition on fertility and mortality, research in medical biometeorology and French historical demographic and economic research to determine the expected time path of effects. The resultant distributed lag system is then estimated. We find that economic/nutritional effects are more likely to be statistically significant in predicting the time path of vital events than are the demographic variables or the effects of meteorological conditions. Stochastic models of fertility predict one- or two-generation cycles in fertility, depending on the responsiveness of fertility to labor force size. St and ard time series techniques are inadequate to estimate the period of such long cycles from the short series that are typically available. When spectral analysis fails, Bogert, Healy and Tukey encourage us to examine spectral analysis applied to the log spectrum of the series of interest (spectral analysis). Observers have noted the presence of long cycles in historical French fertility, claiming these are due to fluctuations in labor force size. We impose simple restrictions on the fertility schedule and apply spectral analysis, then study the spectrum of the log spectrum as suggested by Bogert, Healy and Tukey. In this way we are able to demonstrate a two-generation cycle in French fertility, supporting hypotheses formulated in stochastic models of fertility.