Polling predictions

Rove's brain won't call it for Bush.

By Mary Jacoby
Published November 2, 2004 3:25PM (UTC)
main article image

The chairman of the University of Missouri political science department, John Petrocik, is one of the country's premier analysts of voting patterns and polling methods. He is also a former Republican campaign consultant and -- most important -- an informal advisor to White House political chief Karl Rove. And what he has to say about Tuesday's election will do nothing to put Rove's mind at ease.

The outcome of Tuesday's voting, Petrocik told me in an election eve telephone interview, is virtually unknowable in advance. The polls are broken compasses right now, he said. He reached this conclusion only in the past few days, he said, but declined to say whether he had communicated his conclusion to the White House.


"A couple of days ago, I thought George Bush was more likely than not to be reelected, based on the public polling I had seen," said Petrocik, who was a consultant to George H.W. Bush and is the author of the award-winning 1976 book "The Changing American Voter." Now, he said, "I'm not sure any of us know."

Petrocik said his doubt is driven by the methodology pollsters use to determine who is a likely voter -- and thus who should be interviewed. "You can figure out who the nonvoters are, but you can't easily figure out who the likely voters are. And so pollsters can wind up identifying a lot of people as voters who turn out not to be voters," he said.

Poll respondents who don't indicate by their answers that they are likely to vote are not interviewed further, and their opinions are not registered. But the questions and methods used to screen out the likely nonvoters are treated by most pollsters as inviolable trade secrets. "Pollsters generate a set of propositions about who is most likely to vote, and they develop an algorithm to decide who they want to interview. And they ask us to take it on faith" that they've done it right, Petrocik said. Moreover, "weighting" polls to account for voter preferences that a pollster doesn't believe were adequately represented in a polling sample is another extremely subjective method, he said.


John Kerry partisans, meanwhile, should not invest too much hope in polls that show blocs of undecided voters in some battleground states, Petrocik warned. The conventional wisdom that undecided voters break for challengers because they represent change is not borne out by the data. "There's no evidence of that at all," Petrocik said. "In fact, the evidence that we have on that is 1) they [undecided voters] don't vote; 2) if they vote, they split 50-50 between the candidates; or 3) if there's a runaway candidate, they support the runaway."

Thus, the polls are as difficult to follow this year as toddlers in a moon bounce. In Florida, for example, where 27 electoral votes are at stake, the numbers are on a roller coaster. A CNN/USA Today/Gallup poll has Kerry up by three points among likely voters, a Strategic Vision poll has Bush up by three, and a Zogby International survey has the race tied.

Petrocik said he is not privy to the daily tracking polls the Bush campaign is taking in battleground states. But he said that in a year marked by high passion, huge numbers of newly registered voters and intense get-out-the-vote efforts, the election is about turnout, not poll numbers.


"I have no prediction," he said. "I can tell you different ways of doing it [crunching data] that show a Bush win. But do it another way, and it shows Kerry winning. I think the reasonable thing to say is that everyone promised us a close election, and we've gotten it."

Mary Jacoby

Mary Jacoby is Salon's Washington correspondent.

MORE FROM Mary Jacoby

Related Topics ------------------------------------------

2004 Elections Polling