Here is how I actually characterize Andrew Kohut’s argument. I quote from my article:
Kohut's argument goes as follows: Clinton did much better in the final count than Obama among poorer, less educated voters. These voters "have more unfavorable views of blacks" than wealthier, more educated voters. Kohut doesn't accuse these voters of lying. Instead, he argues that the voters who have unfavorable views of blacks tend to be underrepresented in polling samples, because they "do not respond" to pollsters--thus accounting for the inaccurate readings of support for Clinton and Obama. (my italics)
In other words, I say specifically that he does not think these voters lied.
Kohut says I relied on “incomplete” data. Of course I did. When I wrote my piece, the University of New Hampshire survey was the only one I could find that offered cross-tabulated results. But in relying on the Gallup poll, which he was kind enough to send me, Kohut relied on even less conclusive data. The University of New Hampshire poll was taken on Saturday and Sunday. It showed Obama nine percent ahead of Clinton--only one percent above the average of the pre-election polls--and Obama’s own total of 39 percent was only four percent off what his actual total would be.
The Gallup poll included Friday, the day after the Iowa caucus, when Obama’s totals were most likely to be affected by an immediate boost from Iowa. (This happens invariably after a competitive Iowa caucus--for instance, in 1976, 1980, 1984, 1988, 1996, and 2000.) And if you compare it with previous polls before the Iowa caucus, the boost seems to have come even more among lower than among higher income voters. These voters had previously been one of the bulwarks of Clinton’s vote in New Hampshire, as measured, for instance, by the December 23 University of New Hampshire survey. The Gallup poll not only showed Obama doing unusually well among these voters, but also showed him overall with 41 percent total, and 13 percent ahead of Clinton, and Edwards with 19 percent and closing in on Clinton. So if you were going to ask why this poll didn’t predict outcome, you would look mostly at how early it was taken, and how much it was affected by the Iowa results. If you want to discover why the average pre-election poll didn’t predict the outcome, you would be much better off using the UNH poll as your standard.
How, then, to measure the results? The UNH poll didn’t publish a measure of household income, but it did include education, which Kohut himself cited as a measure of which group might hold an unfavorable view of blacks. I would contend it is a much better measure of social outlook than household income, because it is less correlated with age and marital status and would not group, say, laborers with elementary school teachers. On the basis of this measure, The UNH survey shows that the greatest increase in Clinton’s vote was in more rather than less educated voters--which contradicts Kohut’s thesis.
But let’s step aside a minute and assume that the Gallup Poll does have something to tell us. Looking at it with fresh eyes what would one notice? Not changes in the vote among income groups, but the shift in the candidates’ support among women. The Gallup Poll shows Obama with 36 percent of the female vote, Clinton with 34 percent, and John Edwards with a whopping 22 percent. The exit polls show Clinton with 46 percent, Obama with 34 percent, and Edwards with 15 percent. What is striking is that while Clinton gains 12 percent, Obama loses only two and Edwards loses seven percent.
So what the Gallup Poll may tell us is that many women voters, offended by Edwards’ and perhaps also Obama’s (“you’re likeable enough”) behavior in the Saturday evening debate and by Edwards’ reaction to Clinton’s tearing up, abandoned them for Clinton. Edwards’ loss of votes among female voters would not have registered on the Gallup Poll, most of which was based on surveys during Friday and Saturday. So if you are looking at the Gallup Poll, the obvious conclusion would be that what shifted was women’s support for Edwards.
I am not pressing that as a final conclusion, but I am pointing out that if Kohut had wanted to take the Gallup Poll as his model pre-election poll, he had a much better hypothesis for what happened staring him in the face. He might have also considered the vote among the young and old. There is an equally dramatic shift from Gallup to the exit polls in the over-50 vote from both Obama and Edwards to Clinton. This reflects, I suspect, older women and probably not older women with an unfavorable view of blacks.
Putting these polls together, I would say that what happened is that on Friday and on Saturday prior to the debate among the candidates, Obama and Edwards benefited across the board, but particularly among downscale voters, from the Iowa boost. By Sunday, when the UNH and other surveys were still polling, that boost was wearing off; and in the UNH survey you could see white voters without college degrees returning to Clinton. What would happen over the next two days is a large shift--evidenced in a comparison of the UNH survey with exit polls, of college-educated women away from Edwards and to lesser extent Obama into Clinton’s column. I think that is what happened in New Hampshire. And the reason it didn’t show up in pre-election polls had nothing to do with the Bradley effect.
John B. Judis, Senior Editor
The New Republic