PRINCETON, NJ -- For the third presidential election in a row, Gallup, in conjunction with CNN and USA Today, is conducting a continuous presidential tracking poll. The tracking began on Labor Day and will continue through the weekend before the November 7 election.
There is nothing mysterious about a tracking poll. In the simplest terms, it is survey research conducted continuously over an extended period of time.
Tracking for the most part uses the same methodological techniques as Gallup's normal polling. The interviewing time frame for the typical national Gallup poll can extend from two to four nights. For tracking polls, interviewing is conducted every night, seven days a week, in a continuing cycle. The results are never reported for a single night, however. Several nights of interviewing -- at the current time, three -- are averaged and reported on a rolling basis every day. Each day we report a new three-night average. The results thus represent a continuously changing portrait of where the American public stands on the issues with which the questions deal.
The rationale for continuous interviewing is to measure change. Discrete polls conducted every two or three weeks certainly can show movement in the population, but may miss important developments that in the heat of a fall presidential campaign can sometimes occur on a daily basis.
Indeed, during the fall of an election year, we expect substantial volatility in the vote intentions and attitudes of the public. This should not be surprising. The candidates themselves spend millions of dollars on television and radio advertising, and the news media focus heavily on covering the essentially continuous campaigning of the candidates. Voters begin to tune in more and more to the election as the days go by. We would fully expect that at least some voters would change their minds from day to day or from week to week during this time period -- in response to this avalanche of input. If they didn't, there would be little use in having campaigns, personal appearances, television and radio commercials or the debates.
A tracking poll allows the survey researcher to develop a better picture of how the public is being affected by campaign events. Tracking allows us to provide a quicker measure of the public's reaction to major events than would be possible with a less regular polling schedule.
There are several reasons why Gallup and other polls conduct frequent research to closely follow a presidential campaign.
First, there will inevitably be attempts to characterize the presidential race regardless of whether scientific, neutral data are available. Without neutral polls, journalists and others end up relying on leaked polls from the campaigns, imprecise and unscientific soundings such as Internet polls, pundit observations, or interviews with a few men or women "on the street."
Second, the presidential campaign has interest to Americans (and foreign observers) in and of itself. Americans are not only voters, but also observers. Many of us enjoy knowing how the fortunes of the candidates are changing, what the reaction of the public is to various campaign initiatives and more generally, how the race is shaping up. To know how the candidates are faring is of as much interest to many Americans as it is to hard-core political campaign consultants and reporters.
Third, polling is informative. Keeping track of the issues that concern voters allows both journalists and candidates to maintain closer contact with the people the candidates seek to represent.
Fourth, the candidates and party leaders have access to their own private polls as the campaign progresses, and without publicly released, neutral polls, journalists and the public are left with a poor understanding of what might be driving campaign decisions.
Fifth, public polls can help level the playing field in an election, by giving all candidates equal access to information instead of just those who can pay for it.
It is important to remember that the results of pre-election polls are not intended to predict how the election will turn out (with the exception of the very last poll conducted the weekend before the election, which usually is a good predictor of the actual election results). Instead, polls are conducted to indicate who would win "if the election were held today." We expect changes as the campaigns unfold. Indeed, Jimmy Carter led Ronald Reagan in fall polls conducted in 1980, but Reagan went on to win by 10 points. Carter also led Gerald Ford in 1976 by significant amounts, but the race ended up being very close. Bill Clinton was ahead of George Bush by large margins in 1992, but ended up beating him by only six points.
The Gallup tracking program now underway is based on methodological procedures that Gallup has developed over many years -- including the tracking programs that monitored the presidential elections of 1992 and 1996.
At this point, the Gallup tracking poll includes interviews with a random probability sample of 400 national adults every night. The same complex sampling methodology used for normal Gallup polls is employed in developing the samples used on the nightly basis. The objective is to provide every 18+ adult in the United States an equal probability of falling into any given night's sample.
Reports of the results of the interviewing, as noted, are based on the latest three days of interviewing, with a total of approximately 1,200 national adults as the base. Each three-day sample, as is standard practice for Gallup polls, is weighted to conform to U.S. Census Bureau population estimates for gender, age, region, education and race. The data are also weighted to account for the number of phone lines used in respondents' households to receive calls, thereby correcting for any bias introduced by oversampling households that use more than one phone line.
Once a household is reached, a random selection procedure is used to choose a specific adult respondent living within that household. If the randomly selected respondent is not home, a callback is arranged. Up to five calls are made to each sampled household in order to complete an interview. Additionally, sampled phone numbers that are not answered or are busy are also called back , for a total of up to five calls, before the number is replaced with a new sampled number. These callbacks are made on different nights and at different times in order to attempt to contact the household and complete the interview.
Regional quotas are set each day to ensure that every sample is geographically representative across the continental United States and that day-to-day differences in results can not be attributed to differences in the geographic distribution of respondents.
Once the sample of 1,200 adults is available, Gallup employs a number of steps to winnow those interviewed down to the sample of those we consider to be most likely to vote. This procedure provides us with the best estimate of those most likely to vote in a scenario in which -- in 1996 - about 50% of the voting-age population in this country voted. We monitor internal indicators of turnout, and adjust our turnout estimates as needed as the election draws closer. The models we use are self-weighting, and allow for the inevitable changes in who is most likely to vote. We do not control for nor weight for party identification because partisanship is a variable that can change from day to day and week to week.
The ballot question ("If the election were held today, would you vote for … ") is always the first substantive question asked in our tracking polls. We push undecideds; that is, we ask those who do not make an initial choice of a candidate which way they lean. We are currently reading four candidates' names to respondents -- Gore, Bush, Buchanan and Nader, along with their vice-presidential running mates and parties. We allow respondents to volunteer the names of any other candidates for whom they would like to vote.
We post the results of interviewing every day on our Web site in two forms. First, we report the latest three-day average. Second, in order to provide a somewhat longer-term overview, we post the latest six-day rolling average, which, by its very nature, helps monitor more lasting change in the attitudes of the electorate. We also post the full trend line of our tracking results, along with demographic and geographic breakouts and substantive analyses of what is happening.