Exit Polling in the 21st Century: A Perspective from the USA

　　On 15 June 2012, during the 65th Annual Conference of WAPOR (World Association of Public Opinion Research) held in Hong Kong, a special panel was hosted by WAPOR’s sister organization AAPOR (American Association of Public Opinion Research) on “Exit Polling in the 21st Century: A Perspective from the USA”. Four presentations were made, three of them are summarized in this article, the fourth will be published next time.

“Why Accurate Exit Polls are Important to Democracies”
　　■ Michael W. Traugott, University of Michigan, USA

　　This presentation discusses exit polls from a normative perspective. Election is a way to legitimate the transfer of power and control to the government. By casting their votes on one candidate or party, voters also transmit their views to the leaders. Therefore, exit polls not only describe or explain the patterns of voting, but also people’s expectation for the newly elected government, thereby affecting its policies.

　　In the United States, all exit polls are conducted by the media, who consider elections attractive because 1) they involve conflict and visible figures, 2) they occur on a schedule, and 3) they have a clear resolution on election day. Because of these, news organizations can allocate their resources efficiently to cover elections.

　　The American system is characterized by the competition between televisions and newspapers, but the nature of competition has changed over time. They compete primarily for segmented audiences, for revenues generated by advertisement, and also for peer recognition for outstanding election coverage. Over the last decade, economic pressures are growing in the news business.

　　In USA, there are series of simultaneous local elections, but not national ones. Each election is governed by a local jurisdiction, be it a state, a county or a city, and there are local rules and regulations governing voter eligibility and voting procedures. A debate is currently taking place in USA on how voting can be simplified. The Democratic Party wants to simplify the rules but the Republican Party opposes. The Americans do not elect their president by summation of all the votes cast across the nation, they use a system of Electoral College whereby electors appointed by each state formally elect the President and Vice President. The closing time of polls differs across time zones, and one concern about exit polls is their relatively early declaration of a winner before all votes are casted.

The timeline of American exit polling is as follows:
1967: Start of exit polling in selected states, each run by a　network
1990: Establishment of the Voter Research Service (VRS)
1994: Establishment of the Voter News Service (VNS)
2003: Establishment of the National Election Pool (NEP)
2008: Demise of the Los Angeles Times exit poll

　　The primary function of exit polls in America is to project the outcomes in individual states, and then to explain the patterns of voter support demographically and attitudinally. They are not used to validate the vote as in some other places. The news organization will archive the data and make it publicly available after some time, thereby becoming an important source of continual analysis.

　　Research shows that the Americans are generally interested in polls both as a form of communication and entertainment, but are poorly informed about their methods. They care about exit polls and projections from the perspective of freedom and privacy. Because the US Constitution guarantees the freedom of the press, exit polls cannot be regulated. Instead, the news organizations adopt a gentleman agreement, and do not release any exit poll result until election closes at a particular state.

　　　　　　　　　　　　＊　　 　＊　　　 ＊　　　 ＊


“Exit Polling 101: An Overview to Current Methodological Approaches”
　　■ Paul J. Lavrakas, AAPOR, USA (presented in absentia by Trevor Tompson)

　　This presentation summarizes how exit polls are planned and conducted in recent elections in USA. Pre-election telephone polling starts about 10 days before the elections. Both landline and mobile RDD (random digit dialing) frames are used, and some researchers are exploring the use of registration-based frames. The data would be used in some computer modeling.

　　Prior to the election information is gathered from all sources, including pre-election polling results, past election outcomes within relevant geo-political areas, and expert knowledge. These information is used to set the “priors”, that means priming the election day computer models.

　　Probability sampling is used to select precincts and voters. This involves multistage sampling of voting precincts within each geo-political area being surveyed, and a random systematic sample of voters within each sampled precinct, say, selecting every fifth voters coming out of an exit. There are two types of data collection, one from initially designated sample that cooperates (the successful interviews), and one about initially designated sample that refuses to cooperate (the interviewer records on the refusal sheet the approximate age and gender of the target).

　　Interviewers are carefully selected and trained prior to the election day. The draft questionnaire is usually long, but will be shortened after pilot testing. Data confidentiality is ensured, and the whole process is to capture enough data to predict the election outcomes, and to understand the “mandate” of the election.

　　Another information that goes into the formula of prediction along with the exit poll data is the returning of “real” votes at the precinct level. Researchers actually hire many thousand people to get the vote results across the entire geo-political area being measured.

　　Reliable and secure systems are used to process the data. Analyses are conducted throughout the day, and stringent measures are used to guard against unauthorized early dissemination of the incoming data.

　　Early forecasting are based on data from priors, early voter surveys, and data from exit polls. Extremely conservative decision rules are used for calling an election outcome before any real vote is in. Estimates of within-precinct bias compared to real votes are constantly made. For races too close to call when voting ends, real vote counts are continually updated, so is the outcome prediction.

　　　　　　　　　　　　＊　　 　＊　　　 ＊　　　 ＊


“The Future of Exit Polling”
　　■ Trevor N. Tompson, NORC at the
　　University of Chicago, USA

　　Increasing emphasis is placed on making voting more convenient. New technologies will permit different modes of voting, including early voting, voting by mail, convenient voting centers (where people can go anywhere outside their home town to cast their votes), and the question is “Will Internet voting be far away?”

　　As society gets more diverse and complex, exit polls become more complicated in terms of translation and language. Race and ethnicity of interviewers is another concern in the area of cultural sensitivities.

　　Regulatory challenge is always an issue, especially after 2000 when a wrong call of winner was made. Governments are becoming more active in regulating surveys of all kinds, including exit polls. People are more concerned about privacy, so there will be more restrictions on access to the polling places, more privacy regulations and publication bans. Public distrust is getting worse year after year. Response rates for opinion surveys are declining.

　　New technologies for exit polling include hand-held devices and internet surveys. The same applies to actual voting, and paper ballots is giving way to voting machines, touch screens, and so on. Researchers are already adopting internet panels at a lower cost, rather than using telephone survey. Registration-based sampling will be the next thing that USA will spend a lot of money in exploring. The registration list at precinct and state levels are available to the public, so research institutes can use such sampling frame to develop the techniques. Address-based sampling is another innovation that one can use to improve the accuracy of the survey. Mobile devices and social media are also used for election day surveys now.

　　There used to be only the news media doing the polls on election day, now more exit polls are being run by academic institutes at state level, and interest and political groups are doing their own election day data collection. Rumor has it some are doing it for commercial ventures. It is problematic if we have only one source of election day data, but too many sources for too many different purposes and quality levels also creates confusing results.

　　Financial crisis is facing traditional media sponsors of exit polls. With new methodologies and technologies at a lower cost, the gulf between traditional and new methodologies has increased, and trusted techniques may have to be abandoned.

　　Exit polls will not go away in the United States, but they will change and new models of sponsorship will evolve. There will be more competition from other models, especially “exit polls” built on internet panels, and new methods that emerge will be a hybrid of old and new approaches.