Charlie Cook, guru of elections and master at predicting outcomes, gives a primer on reading polls before election day.
Verbatim:
First, don't focus on individual polls. With recent surveys from highly reputable firms showing everything from a 5-point lead for President Bush to a 4-point lead for Democratic presidential nominee John Kerry, the temptation is to focus upon the poll whose outcome matches the result you would like to see. A variation of this fallacy is to pay too much attention to the most recent poll. The news media tend to exacerbate these tendencies by hyping poll results that seem particularly newsworthy, that is, unlike most of the other recent polls.
Go, instead, with an average of the major national polls. A handy way to do that is to look at www.RealClearPolitics.com, which each day computes a seven-day moving average of these polls. Although the Web site includes a few polls that aren't among my favorites, it excludes those with the most dubious methodology, such as Internet polls. Currently, the average shows Bush leading Kerry, 47.8 percent to 45.8 percent, with Ralph Nader drawing 1.6 percent. Off-the-record assessments by top people in the Bush and Kerry camps are quite similar to RealClearPolitics's rolling average.
Second, don't focus on the point spread between the two candidates. Look at Bush's percentage. Given that in races with well-known incumbents, undecided voters tend to break overwhelmingly in favor of challengers, Bush's level of support is the most important number now available. Democratic pollsters Guy Molyneux of The American Prospect Online and Mark Blumenthal (aka mysterypollster.com) make this argument exceedingly well.
To win a two-way race, an incumbent needs to be approaching the 50 percent mark in the polls. In this race, Nader and other minor candidates will likely draw 2 to 3 percent of the overall vote. (Nader received 2.7 percent in 2000; my hunch is that he'll get only 2 percent this time.) So, Bush needs to be around 48 or 49 percent to win. He might be able to win with just 47 percent.
But below that, his chances of winning plunge.
Third, keep an eye on Bush's job-approval ratings, again looking at the average (currently 50 percent on RealClearPolitics.com). A rating below 50 percent spells trouble for an incumbent. The predictive value of the job-approval score is best in the summer; the trial heat becomes a better predictor in the final two months of the campaign.
Fourth, ignore the Electoral College. If someone wins the popular vote nationally by at least 1 percentage point, the electoral vote will almost certainly go the same way. But if the popular race is really, really close, there will be so many true battleground states that it will be futile to try to predict which way the Electoral College will go, particularly if one has to rely on state polls of doubtful quality. Even if one were privy to the high-quality survey research that the two campaigns and the largest "527" groups have, it would be impossible to know which way the handful of most critical states are going to go.
Does anyone think a poll could have forecast Bush's 537-vote victory in Florida four years ago, or Al Gore's 366-vote win in New Mexico? Trying to push the last five or six states into a "red" or "blue" column is a fool's errand.
While these are the gauges that the pros use, remember that none of them is perfect. Keep in mind, as Emory University's Alan Abramowitz has pointed out, that in the final week of the 2000 campaign, Bush led in 39 of 43 national polls, Gore led in two, and two were tied. In those surveys, Bush had an average lead of 3.6 percentage points, yet he went on to lose the popular vote by half a percentage point, so the national polls overstated Bush's strength by 4.1 percentage points.
While some of that discrepancy was due, no doubt, to the late-breaking news story about Bush's having been stopped 24 years earlier for driving under the influence, that doesn't account for the whole 4 points. In 2000, Democrats unquestionably had the superior get-out-the- vote operation. (This year, both sides have magnificent ground games.) And Bush's "compassionate conservative" message may well have left 4 million white evangelicals unmotivated and not voting.
But the point is, when a race like this year's presidential contest gets really close, nobody actually knows what's going to happen.
Back to top