Election Research

Demographics and Political Geography

While analyses of party organization and campaigning appeared from the latter part of the 19th century onward, the tradition of systematic quantitative studies of voting statistics began with André Siegfried in France, Harold Gosnell in the American Midwest, and V. O. Key in the American South from 1900 onward. Such analyses used demographic data—social and voting statistics, generally from the smallest constituency unit available—to make sophisticated inferences about individual voting behavior and its influence on both personal characteristics and political history.

This tradition of research, however, passed out of the mainstream under the impact of two developments. The first was the discovery of the ecological fallacy involved in inferring individual behavior from aggregate statistics. For example, the finding that constituencies in the United Kingdom with a large Black population voted disproportionately for the racist British National Party (BNP) does not imply that the BNP attracts non-White votes. On the contrary, it shows that Whites in such constituencies vote for it disproportionately, under the impact of high immigration. It is hard to tell from aggregate statistics what exactly is going on at the individual voter level. The second development relates to the use of mass survey techniques.

Survey-Based Research and Social Group Theory

In addition to difficulties involving the ecological fallacy, the problem for the statistical foundations of demographic research was compounded by the application of mass survey techniques to study individual voters directly. This approach was pioneered by Paul Lazarsfeld and his associates in Erie County, New York, for the U.S. presidential election of 1940, and replicated in Elmira, New York, in 1948. Not only did these studies demonstrate the feasibility of applying statistical techniques of sampling and quantitative analysis to voting, but they also demonstrated the way in which data collection and analysis had to be driven by theory to produce relevant conclusions or even to develop a questionnaire in the first place.

Coming from a market research background, Lazarsfeld envisaged voters as behaving like consumers choosing between tins of beans on the supermarket shelves. Relying solely on advertising for their information, brands promoted more in the media would sell better than ones promoted less. The 1940 questionnaire thus featured many questions on voter media exposure (with parallel content analyses of actual media coverage of candidates), with a few questions on social characteristics thrown in for classification purposes. To their surprise, the research team found few and weak correlations between media coverage and voting behavior but strong correlations with class, religion, and urban versus rural residence.

...