The Case for Caution: This System is Dangerously Flawed

Konner, Joan

On any election day, interested citizens want to know one thing--who won. News organizations are in the business of getting accurate results to !heir audience as quickly as possible. but counting actual votes takes hours, and sometimes days. Beginning in the early 1960s, news organizations developed ways of projecting the outcome of races in order to speed the process of reporting before votes were actually counted. They began to develop methods and systems of modeling and polling that could indicate, statistically, the likely winner in any given race. The motive was to give the audience what it wanted--the faster, the better. Of course, the news media intended to make their projections as accurate as possible. In time, polling and analysis became increasingly sophisticated. Results from preelection samples, along with extrapolations from precinct models, exit polls. and partial election returns, were combined into what I will refer to as the networks' election-day polling and projection system. This system provided the basis for making election projections faster and better, meaning with fewer mistakes. Year after year, the systems were improved, spurred on by competition among the news organizations to be the first to report outcomes to their audiences. Not incidentally, the highly competitive polling and projection business grew increasingly costly. In 1990. the first network pool for exit polling and projections, Voter Research and Surveys (VRS), was formed with the intention to meet the increasing costs and share expenses. Cost sharing made it possible for the networks to provide the greatest sweep of polling. Without the pool, the networks would have had to restrict their reach and coverage because of budget limitations. Of course, in journalistic terms, pooling meant the information would be less re liable. While the networks could, by combining resources, undertake larger polling operations and more sophisticated modeling that could reduce the risk of certain types of error, the vulnerability of the networks to any errors that did result was increased. When data are wrong, with only one source of information, there is no opportunity for correction. Nevertheless, financial considerations trumped reliability--and the best practices of journalism. By election day 2000, after several permutations, a comprehensive polling and projection system was in place backed by a consortium of five television networks and the Associated Press. Its purpose was to collect and disseminate polling data and voting information by which news organizations could make their independent calls, maintaining an element of competition among them. In thinking about this system in its entirety, we must consider not only the Voter News Service (VNS), the reconstituted consortium operation, but also the analysis and reporting operations of the separate networks as well. The system was economical. and it was fast. But was it accurate? The answer: not as accurate or as reliable as it was intended, promised, or needed to be, especially when it came to calling a very close race. We learned that answer on election night 2000. At the core of the reporting problem were two mistaken projections in one state, Florida, which turned out to be key to the outcome of the national election. The television networks and other news outlets twice projected the winner and twice recalled those projections. News executives, particularly television news executives, as well as editors, correspondents, and producers themselves described election-night coverage as a "debacle," a "disaster," and a "fiasco." Something had gone wrong--terribly wrong--in the polling and projection system. It is not the purpose of this article to ferret out the exact sources of the errors on that night. The experiences of election night 2000 do, however, serve as a useful lens through which to examine the overall efficacy of the system that was in place. It is my contention that this system is too fraught with the potential for error for news organizations to rely on its projections in the way that they have in the recent past.


  • thumnail for JK-Public-Opinion-Quarterly.pdf JK-Public-Opinion-Quarterly.pdf application/pdf 1.93 MB Download File

Also Published In

Public Opinion Quarterly

More About This Work

Academic Units
Published Here
November 9, 2015