One pollster's take on Wisconsin

Pollster Dick Bennett thinks he's caught sight of a significant wave that will crest in today's Wisconsin primary. Either that, or he's going to catch a significant amount of grief on Wednesday.

We contacted Bennett, president of the New Hampshire-based American Research Group, because his organization's most recent surveys had left our head spinning.

A poll conducted Friday and Saturday of likely voters in Wisconsin's Democratic presidential contest had caused a buzz because, in contrast to other recent surveys that had shown Barack Obama slightly ahead, it gave Hillary Clinton a 6-percentage-point lead (49%-43%).

Bennett's questioners were back at it on Sunday and Monday, and the results were dramatically different: Obama led by 10 points (52%-42%).

The margin of error ...

for both polls was plus-or-minus four points.

In discussing his results, Bennett acknowledged that this campaign season has frequently given him and other pollsters pause. Like many, for instance, he misfired in gauging this year's New Hampshire primary -- his final poll in his home state had Obama up by nine points; Clinton won by just shy of three.

But Bennett expressed his confidence in his final sounding in Wisconsin, saying his questioners had identified a clear trend toward Obama among the type of voters Clinton has been counting on: Lower- and middle-income whites.

The shift started among men in this group, many of whom were the target of John Edwards' populist message. Initially uncomfortable with Obama, they warmed to him as the short Wisconsin campaign intensified, Bennett said. And at that point, their allegiance became strong: "They don't like Clinton anymore after they decide on Obama," he said.

The big turnabout in the last few days, according to Bennett, occurred among women in this demographic. Simply put, many who had been for Clinton changed their minds. "They're happy with either one," Bennett said. But they opted to support the candidate they see as having momentum: Obama.

You can read the pollster's complete analysis here. And in a few hours, we'll see whether he was on, or off, the mark.

-- Don Frederick

Comments

[Luke]

This pollster deserves some kudos for sure. Not just for being right, but also for actually putting forth his seemingly contradictory poll results with some confidence. After New Hampshire and California, it seems lots of pollsters would rather sit this one out.

[Ronald Loui]

Whenever I see an ARG poll, I just take 5 points from Clinton and raise Obama by 5. ARG has been consistently off with this bias. But it is true that the last ARG poll had Obama ahead of Clinton in Texas, so I'm not saying it's deliberate bias. There was actually a discussion at pollster or rcp talking about how their method was actually no different from other pollers'.

Of about a dozen polls I just looked at, ARG predicts the spread in a state with an average of 11 points of error in 14 states. That's not too good. The best poller had a much lower average error (on spread) of 7.6 points in 13 states. That's Survey USA. (This uses only final polls if they were within a month of the actual vote.) Then comes Research2000 at 8.3 in 7 states, Rasmussen at 8.7 in 17 states, Zogby comes in at 8.8 in 9 states, then MSNBC at 9.2 in 11 states.

The only people worse than ARG are CNN, with a whopping 18 points of error in 5 states. Why not just pull the numbers out of your pants? The local Universities are worse, at 18.5 in 7 states, though this is often an earlier polling number, and Obama keeps creeping up on Clinton in the last weeks. Don't be too smug, though, as the next worse, and a very close egregious ignominious third place, is the local media, with 15.4 points of error on average, in 13 states.