By Zachary Roth
The story of polls, the press, and Tuesday’s Wisconsin primary is not a pretty one, and there are no heroes.
In the smoking wreckage of the aftermath, it was painfully noticeable that pollsters and just about the entire campaign press failed to predict the late surge in Wisconsin that carried Sen. John Edwards to renewed prominence.
There was a precedent that reporters might have paid attention to here. In Iowa, polls taken even the day before the contest showed the race to be almost a four-way dead-heat, between Edwards, John Kerry, Howard Dean, and Dick Gephardt. Those polls didn’t come close to predicting Kerry’s margin of victory, particularly over Dean and Gephardt, who lost by 20 and 27 points respectively.
The pollsters have already expressed their own mea culpas over Wisconsin. As Jim Rutenberg of The New York Times reported yesterday, the head of Mitofsky Research — the exit polling service used by the networks and the Associated Press — on Wednesday posted the following note on a polling website:
“Yesterday exposed the biggest goofs in my memory.”
But Campaign Desk is more interested in the 2004 campaign press’s habit of making altogether too much of this consistently unreliable source. As Steve Seplow, a former campaign reporter and Knight Ridder Washington bureau chief, puts it, “What is more lazy than a reporter who relies on polls?”
In the days leading up to the Wisconsin vote, the press went astray in two different ways.
First, the final poll last week by the American Research Group (ARG) showed Kerry 53 percent, Edwards 16 percent, Dean 11 percent. That turned out to be shockingly inaccurate — the actual vote count was Kerry 40, Edwards 34, Dean 18. But that final ARG poll was conducted February 12 - five days before Wisconsinites went to the polls.
Why didn’t ARG poll after that? Dick Bennett, ARG’s president, told Campaign Desk that his news media clients —who he would identify only as a group of radio stations in Wisconsin and the Chicago area — decided not to pay for additional polling after February 12.
Those stations employ Bennett not only as a pollster — a provider of raw data — but also as a conventional political pundit who appears on their shows to interpret that data, and to speak more generally about the race as a whole.
According to Bennett, “the stations thought Dean was surging,” based not on Bennett’s research, but on anecdotal evidence, like the size of his crowds. That was a storyline the stations liked, since it suggested a dynamic, exciting race in which the much-maligned outsider might be staging a late comeback. So in the week before the race, everyone seemed to settle on the “surging Dean” theme.
Problem was, Bennett’s ARG polls weren’t picking that up. His February 12 survey, in addition to giving Kerry a 37-point lead, also showed that Dean’s unfavorable ratings were so high as to prevent him from challenging Kerry for the lead.
Having decided that Bennett’s polling didn’t fit the agreed-upon script, the stations stopped commissioning any more ARG surveys for the crucial final five days of the campaign. “When I didn’t find (a Dean surge), that’s no fun … cause what do you talk about?” said Bennett.
As Bennett sees it, the consortium of radio stations that made up ARG’s clients were cherry-picking poll results, choosing to commission only those surveys that they expected would show a Dean surge — not because they’re Deaniacs, but rather because of a desire to create and sustain what they saw as a compelling storyline.
Aside from missing the real story — Edwards’ surge in the race’s last few days, driven by a strong debate performance on Feb 15, and the endorsement the next day of the Milwaukee Journal-Sentinel, the state’s largest paper — the stations’ bias turned them from purveyors of news into creators of a story line that didn’t correspond to reality.
A more conventional misuse of polling data was displayed by most of the national media.
Here’s a representative lede from David Broder and Jim Vandehei’s Washington Post story on the morning of the vote: