How to read poll results
Understanding polls is first step in interpreting them
Every day there seems to be a new poll indicating a shift in the lead or increasing momentum for one candidate or another.
And with none of the polls looking exactly alike, it seems impossible to know which polls are accurate.
Relax. CNN Polling Director Keating Holland answers some basic questions to make reading, and understanding, poll stories much easier. (Special Report: America Votes 2004, Poll Tracker)
Can a poll predict who will win an election?
Only a poll taken immediately before Election Day can accurately predict the outcome of an election.
Earlier polls might have some predictive power if nothing changes between the time the poll was taken and the election.
Of course, many things do happen: the primaries, conventions, debates and other events often change the minds of many people. (Why hold those events if they are not expected to change the minds of some voters?)
A good metaphor might be a scoreboard. If one team is winning in the first quarter or the third inning, you know that does not mean that the game is over. The score posted on the scoreboard just shows how the two teams stand at that moment, not what the final score will be.
Even in the final few seconds of a game, the scoreboard does not predict the final score; sports -- like politics -- are full of buzzer-beaters, Hail Mary passes and walk-off home runs.
How are people chosen to participate in polls?
All interviews are conducted by telephone. Since nearly every household in the United States has a telephone, this method gives nearly all Americans an equal chance of being selected to participate in our polls.
In a technique known as "random-digit dialing," a computer selects completely at random the phone numbers that our interviewers call. This method allows us to reach people with unlisted phone numbers and people who have moved recently, as well as those who are listed in the phone book.
Who is a likely voter and how do you know?
We know that not all Americans who are eligible to vote do so. And voters, as a group, can be different from nonvoters.
As a result, in the months before an election, pollsters try to identify those respondents who are most likely to vote. We usually refer to those people as "likely voters." Typically, they are registered to vote, they say they are definitely planning to do so, and they fit other criteria that years of polling have shown are likely predictors of voting behavior.
What does margin of error mean?
Random samples obviously aren't as accurate as interviewing the entire population. Fortunately, it is easy to measure the biggest possible difference between the results of most polls and the results you would get if you asked the same questions of all 200 million adult Americans.
That maximum difference is called the "sampling error" or "margin of error." Ninety-five percent of all polls are guaranteed to fall within that relatively narrow range.
If two candidates are within the margin of error, is it possible to say one candidate is leading the race?
Not exactly.
Let's say we do a poll in which Candidate A would win 56 percent of the vote if the election were held today.
If the sampling error were plus-or-minus three percentage points, it means there is some chance that Candidate A's support could be as high as 59 percent (56 plus 3) if we had asked all 200 million adult Americans. There is also some chance that the candidate's support could be as low as 53 percent (56 minus 3).
But there's a catch.
That simple math does not mean that the chances are equal that Candidate A's support is 53 percent, 54 percent, 55 percent, or 56 percent.
The likelihood is very low that her support is 53 percent, slightly higher that it is 54 percent, higher still that it is 55 percent, and highest of all that she would actually win 56 percent of the vote.
The same is true at the other end of the scale where the candidate might win more than 56 percent. And the same logic holds true for the candidate's competition.
What's an exit poll?
An exit poll is a special kind of poll conducted on election day.
It is not a typical telephone survey. Instead, specially trained interviewers are stationed at the exits of polling places and interview voters after they have cast their ballots.
The major advantage of this method is that we are absolutely certain that we have interviewed people who have actually voted.