Ecological Inference and Electoral Analysis

Abstract

Ecological Inference and Electoral Analysis This paper examines the utility of a relatively new statistical method for studying the relationships between individuals' sociological characteristics and their voting behavior during periods when only aggregate data are available. Almost all of the propositions relating sociological factors and voting behavior are based on the post-I940 period when sample surveys began providing reliable individual-level data. Yet three decades constitute an extremely weak foundation for generalizations and, in order to have as large a data base as possible underlying our generalizations, we must extend our studies of individual voting behavior back into time. Moreover, it is clear that certain large-scale transformations in sociological groups' voting preferences take several decades to work themselves out. If we are to gain an adequate understanding of the dynamics of voting behavior over long stretches of time, we must be able to estimate the individual-level relationships from aggregate data. As Robinson has shown, we cannot directly infer individual relationships from aggregate ones. The fact that a certain set of geographical areas with a certain pattern of demographic and social characteristics behaves in a certain way cannot be construed as meaning that individuals in those areas possessing most or all of the various characteristics all behave in the same manner. Even though, for example, districts with a high proportion of black residents tend to return large Democratic majorities, there is still no airtight logic, no matter how great the suspicion, for inferring that blacks tend to vote Democratic. Fortunately, devices for circumventing the ecological fallacy have been developed.2 The one we will examine here is Goodman's technique, which uses ecological regression coefficients to estimate individual voting proportions.3 To understand how Goodman's technique