TICKET-SPLITTING: AGGREGATE MEASURES VS. ACTUAL BALLOTS ALAN R. GITELSON, Loyola University of Chicago and PATRICIA BAYER RICHARD, Ohio University

For Some Decades now, political scientists have investigated the influence of various motivational, behavioral, and institutional variables on voting behavior. More recently, the phenomenon of ticket-splitting has captured the attention not only of political scientists but of political practitioners and journalists as well. Ticket-splitting refers to the behavior of individual voters; it generally is taken to mean the act of casting votes for candidates representing more than one political party on a single ballot. The study of split-ticket voting, however, has been confined to the use of indirect measures, primarily survey and aggregate data, because of the unavailability of actual ballots. This study, however, utilizes actual ballots.

The researchers examine several aggregate measures of ticket splitting which have been proposed and/or employed in the literature. We then evaluate the accuracy of these aggregate measures by comparing the estimates of ticket-splitting with actual ticket-splitting in one community, ascertained from its 1972 and 1976 election ballots. Finally, we assess the hazards of drawing conclusions about ticket-splitting based on aggregate measures of the phenomenon.
A BRIEF REVIEW OF THE LITERATURE
For a variety of reasons (e.g., their availability, their avoidance of many of the measurement problems of survey research), social scientists frequently use date on geographic units such as states, counties, or census tracts to assess individual behavior (Shively 1969; 1183). At the same time, they have directed attention to the logical and methodological problems inherent in drawing inferences from one level of analysis to another (Robinson 1950). Others (Alker 1965; Duncan, Cuzzort, and Duncan 1961) have expanded on the problems arising from inter level inferences. In essence, this "ecological fallacy" consists of inputing correlations on the individual level from correlations between the same variables based on groups or aggregations of individuals as the units of analysis.
Studies of split-ticket voting run the risk of inaccurate inferences when they rely on aggregate election data (Gitelson 1978). Burnham (1965) and Rusk (1970), for example, use the difference in partisan vote proportions for different office contests to make statements about individual ticket splitting. Yet Cowart claims (1974: 110-12) that this indicator is necessarily a conservative measure of the true level of ticket-splitting, always to the actual level. Others, for example DeVries and Tarrance (1972) and Phillips (1975), use split partisan outcomes for two elected offices to discuss levels of individual ticket-splitting. Both these indicators, difference in partisan proportions and split outcomes, may misestimate the number and proportion of split tickets as well as give rise to incorrect comparisons between communities or elections with regard to levels of ticket-splitting. They bear no necessary relationship to the actual number of split ballots. The data and discussion which follow raise questions about the degree of distortion arising from two aggregate measures of ticket splitting.
AGGREGATE MEASURES OF TICKET-SPLITTING
Two aggregate measures for determining levels of ticket-splitting are evaluated by the authors: the Burnham measure and the Duncan and Davis method of setting limits.
DOWNLOADHERE