The Methodology and Performance of Election Day Polls

Abstract

Methodological details are presented of a survey research technique known as exit or election day polling. Using this technique, major American news organizations collect and analyze voting and attitude data from samples of persons who have just cast ballots. On the basis of the 1980 elections, differences in polling strategies and performance of the exit poll method are examined. How election day survey data are used by journalists is discussed. Mark R. Levy is an Associate Professor, College of Journalism, University of Maryland. Some of the data utilized in this article were made available by the InterUniversity Consortium for Political and Social Research. Neither the collectors of that data nor the Consortium bear any responsibility for the analyses presented here. The author would like to thank those exit pollsters who so graciously shared their time and secrets. Public Opinion Quarterly Vol. 47:54-67 ? 1983 by the Trustees of Columbia University Published by Elsevier Science Publishing Co., Inc 0033-362X/83/0047-54/$2.50 This content downloaded from 157.55.39.249 on Wed, 03 Aug 2016 05:45:48 UTC All use subject to http://about.jstor.org/terms ELECTION DAY POLLS 55 and, to a lesser degree, academic students of the election process, had relied primarily on preelection surveys and vote returns from selected precincts to interpret the vote (Bicker, 1978; Bohn, 1980). But many election analysts, within and outside the media, believed those two methods to be seriously flawed. Preelection surveys, for instance, suffered from the classic screening'' problem of how to identify likely voters (Crespi, 1977), and more important, because of publication deadlines, they often required early field work closeout and were thus unlikely to capture late shifts in voter sentiment. Precinct-level vote data too were recognized as having substantial journalistic and scientific limitations. First, election returns contained no explicit information about voter attitudes and perceptions, although journalists often inferred them anyway. Second, so-called tag or analytical precincts were often selected through purposive sampling of election districts with disproportionate concentrations of voters having shared demographic characteristics (race, religion, income, etc.) or with strong partisan voting histories. See, for example, Levy and Kramer (1972). Aware of the ecological fallacy (Robinson, 1950), concerned that ghettoized voters did not represent all voters sharing a given demographic attribute, and faced with increasing difficulty in locating homogeneous precincts, media election analysts turned to election day polling. By 1980, the three television networks, the Associa-ted Press, The New York Times, and The Los Angeles Times were all surveying voters as they left the polls.2 This article outlines the methodology of election day polling and compares the different approaches taken in four major polls.3 Examples and data are drawn from the 1980 presidential primaries in New Hampshire, Florida, and California, and from the November 4 general election. This report focuses exclusively on the presidential campaign, although election day polls using virtually the same methods have been and continue to be routinely conducted in state and local contests as well. The elections selected for analysis here were chosen because the author believed that these contests represent the range of challenges faced by the profession. New Hampshire, for instance, is a small, 2 Although six news organizations were involved in election day polling, there were actually only three national election day surveys in November 1980, since CBS News conducted its polls in association with The New York Times, and NBC News teamed up with AP. The L.A. Times polled only during the presidential primaries. 3In addition to these four election day polls, Teichner Associates, Inc. of Princeton, N.J. has conducted approximately 20 exit polls since 1979 for broadcast clients interested in local and statewide elections in major television markets nationwide. This content downloaded from 157.55.39.249 on Wed, 03 Aug 2016 05:45:48 UTC All use subject to http://about.jstor.org/terms