Statistical Recycling of Documentary Information: Estimating Regional Variations in a Pre-Censal Population

Abstract

In the process of studying the French Revolution, the authors were faced with the problem of how to use rather poor pre-censal data. Not only were there many different sources available, but there is little information on how they were collected, or how reliable they were. Rather than abandoning hope, or arbitrarily choosing one or another of the sources, the authors factor analyzed them, and found that there had been two main methods of collecting provincial population and estimates in eighteenth-century France. The censuses loading heavily on the most important factor turned out to be those based on counts of births. Using factor loadings to adjust each of the provincial population estimates according to the source will permit the authors to use this data with considerable confidence. The authors use this methodological discussion in order to examine broader issues of data reliability and the utility of quantitative sources in social historical work in general. I. THE SCIENTIFIC USE OF FLAWED DATA For a number of years, we have been engaged in a study of the French Revolution of 1789, using a quantitative methodology which, at bottom, is the search for systematic regional covariations. We seek to study the correlations among three classes of variables, all measures of variations among regions of France: conditions of social life (broadly defined) in the old regime, the openly voiced demands and grievances of the French public at the beginning of the great upheaval (expressed in the famous cahiers de doleances produced by tens of thousands of assemblies across the country), and proand counterrevolutionary acts by various groups during the revolutionary period.' Our research is addressed to such questions as: How did the politics of the French in 1789 vary across the map of France? What aspects of the social, economic, and governmental structure are associated with different sorts of demands? And how are these variations in political views correlated with variations in revolutionary participation?2 In order to perform such studies, we must, of course, standardize many of our measures by population.3 For example, if we wish to study the correlation between the incidence of a certain tax and some other variable, such as hostility to the government, we often must convert the tax to some per capita measure, and for *We are particularly grateful for the detailed reading, the critical comments, research advice and actual data series generously provided by Jacques Dupaquier. Of course, we remain responsible for all errors.