Chad Gaffield

Conceptualizing and Constructing the Canadian Century Research Infrastructure

The Canadian Century Research Infrastructure (CCRI) is an interdisciplinary, multi-institutional, and internationally linked initiative to enable research on the making of modern Canada. At the heart of the CCRI are microdatabases centered on the manuscript census enumerations for 1911, 1921, 1931, 1941, and 1951. This research infrastructure will be added to the results of other projects that cover the periods from 1852 to 1901 and to the Statistics Canada (STC) census microdatabases from 1971 to 2001. When completed in 2008, the CCRI will thus enable research to be made on the individuals, families, households, and communities that experienced the complex transformations of Canada since the mid-nineteenth century. By analyzing approaches to the epistemological issues involved in building the CCRI, the author seeks to advance scholarly debate by describing the research infrastructures distinguishing characteristics and explaining its various components that seek to both support and facilitate research projects. This overview provides the context for the three other articles in this theme issue of Historical Methods that focus on CCRIs sampling and census microdata management strategies as well as the initiatives georeferencing and contextual data systems.

INFERRING AND REVISING THEORIES WITH CONFIDENCE: ANALYZING BILINGUALISM IN THE 1901 CANADIAN CENSUS

This paper shows how machine learning can help in analyzing and understanding historical change. Using data from the Canadian census of 1901, we discover the influences on bilingualism in Canada at the beginning of the last century. The discovered theories partly agree with, and partly complement, the existing views of historians on this question. Our approach, based around a decision tree, not only infers theories directly from data, but also evaluates existing theories and revises them to improve their consistency with the data. One novel aspect of this work is the use of confidence intervals to determine which factors are both statistically and practically significant, and thus contribute appreciably to the overall accuracy of the theory. When inducing a decision tree directly from data, confidence interrvals determine when new tests should be added. If an existing theory is being evaluated, confidence intervals also determine when old tests should be replaced, or deleted, to improve the theory. Our aim is to minimize the changes made to an existing theory to accommodate the new data. To this end, we propose a semantic measure of similarity between trees and demonstrate how this can be used to limit the changes made.

Linearity, Nonlinearity, and the Competing Constructions of Social Hierarchy in Early Twentieth-Century Canada: The Question of Language in 1901

ince the 1960s, scholars have increasingly emphasized the ways in which routinely generated sources such as the census should be understood as creations of a “quantitative mentality” or “statistical mind.” However, this emphasis overlooks the fact that it was a particular type of thinking, namely, linearity, that in modem times characterized the features of those sources. Census forms were certainly designed by “statistical minds,” but, more important, they were crafted within a world view determined to impose linearity on understandings of change and the construction of social hierarchies.’ The linear thinking underlying documents such as the census strove to differentiate individuals in two ways: it put them into singular and exclusive categories (one and only one answer for each question), and it explicitly or implicitly ranked those categories hierarchically and connected them in one-directional ways. Behind each census question was the anticipation of singular responses that were more or less close to an “ideal.” In that sense, the census enumeration was not benignly counting and classifying social “realities” but was, in fact, constructing hierarchies in which individuals were singularly lined up from best to worst. For the most part, we, as historians, have proceeded to analyze those forms from our own linear perspectives. Indeed, much of the debate among quantitative historians since the 1960s has focused on ways of “overcoming” (or simply ignoring) multicollinearity, the confounding of independent variables, the artificiality of dichotomous variables, the challenge of categorical data, and many other difficulties in applying linear-based statistical models to historical evidence. Historians have tended to suppress their unease in analyzing sources such as the census at least partly because it has been seen as appropriate to analyze sources from the ~ perspective of their creation.* Thus, historians have backed away from emphasizing the extent to which historical evidence undermines the neat and tidy-and linear-assumptions of the creators of this evidence. Instead, scholars have, in recent years, increasingly called for new (linear) interpretations or have abandoned any hope of truly understanding the past in all its complexity. The following discussion focuses on the introduction of language questions into the 1901 Canadian census as a way of illustrating the value of analyzing the linear paradigm within which the census was created and of probing the elements of nonlinearity that characterize historical change and social configuration^.^ This discussion reports on three components of the relevant research strategy: (1) an analysis of why the language questions were introduced into the census; (2) a systematic study of the assumptions of the actual wording of the census questions; and (3) a close examination of the actual listings for the 231,909 individuals in the Canadian Families Project (CFP) national sample for whom “mother tongue” is reported, to discover the ways in which linear thinking (at various levels) was articulated and contested in the actual enumeration proces~ .~ In applying the insights of linear and nonlinear dynamics to research on the language questions of the 1901 census, three conclusions are reached: (1) the language questions were posed from the perspective of a linear paradigm; (2) the enumeration forms called for responses within a linear paradigm; and (3) the dynamics of language in turn-of-thecentury Canada must also be understood in terms of nonlinearity. Taken together, this evidence reveals how considerations of language led to competing constructions of social hierarchy. These competing social hierarchies do not fit easily into current scholarly interpretations of how officials tried to regulate society or how the state formed a pub-

The Automated Archivist: Interdisciplinarity and the Process of Historical Research

The basis of most historical research including social science history is quite unsystematic. This characteristic results from the ways in which researchers find and choose historical sources for examination. Despite claims to be systematic, historians still tend to identify relevant evidence in impressionistic ways. Many social science histories involve the rigorous study of a source happily discovered by chance. Of course, access to the past has never been easy. Researchers have always lamented a presumed lack of “essential” records. Nonetheless, the actual ways we discover existing evidence have received little attention despite the fact that this process is fraught with difficulties and hidden dangers especially for researchers of a social scientific bent. Do not the presuppositions of social science history extend to the identification of sources? How do we know when we have all the “relevant data” for a particular project? Can systematic data analysis be justifiably built upon unsystematic identification of sources?