Race and polling

There have been numerous diaries about the disparity between Obama’s poll victory and his loss in the actual election. People are still talking about a “Bradley Effect,” and Thomas Edsall’s recent piece on Huffpo revisits it as an explanation for the disparity between polls and votes cast for Obama in New Hampshire.

Why beat this dead piñata? Because we’re heading into new primaries with new polling, and also because I keep seeing a dominant obviously false argument being made about this and am tired of it. But I also want to ask: how can we test to see if the Bradley hypothesis is true in future races?

[As a caveat, this is not an anti-Clinton diary. I think that most Clinton supporters believe there is still racism in the US, and support remedies to neutralize it. This diary is about how the pollsters could have overestimated Obama's polling strength by such a consistent margin. I’m not saying the reason people were surprised by Clinton’s strong showing was solely that Obama’s poll numbers reflected false reports that had to do with race (which is widely thought to have happened with Mayor of Los Angeles Tom Bradley’s 1982 campaign for California Governor).]

What is the false argument?

Here is Bill Schneider’s analysis offered on CNN:

To keep them honest, we compared those polls with the actual vote. The big discrepancy was in the Clinton vote. She got nine points more support than the polls predicted. Edwards' vote was close. Obama's vote was right on target, 37 percent in the polls, 37 percent in the primary. Obama got exactly what the polls predicted. No evidence of a racial effect. (h/t HoneyBearKelly, Gravastar)

A similar argument is part of yesterday’s diary here on this subject:

If you look at the polls, Obama got was averaging 38.7 percent (37, 42, 41, 39, 35, 38, 39) in 7 pre-New Hampshire polls. He got 36 percent when all of the votes were counted, which is well within the margin of error. In the polls Hillary Clinton averaged 29.3 percent (30, 29, 31, 30, 28, 28, 29) in 7 pre-New Hampshire polls. She got 39 percent, which is a 10 percent jump from what the polls showed she would get. So a better question would be why Hillary Clinton had a much better performance than the polling showed.

While I don't know the causes of the discrepancy, Schneider is clearly wrong. This is not evidence that there was no racial effect.

His argument is that since Obama got what he polled at, this shows there was no "Bradley effect." But this assumes we know all the undecideds broke for Hillary. That would be counterintuitive. Because of the coincidence of Obama polling at a number, and then getting that number, it is easy to assume these were the same people.

But it is also possible the same number of undecideds broke for Obama as people who said they would vote for an African-American candidate and then did not. So the actual total was the same as the one he polled at.

Now, it may well be that women undecideds overwhelmingly broke for Clinton, but that doesn’t mean that some men and women also told pollsters who called them at their home that they would vote for Obama, and that this number roughly matched the number of men and a few women who broke for Obama.

To be fair, this is not the only argument against a “Bradley effect”. Yglesias makes a slightly different argument:

The pollsters underestimated Clinton's level of support. People who were undecided as of the last round of polling seem to have gone overwhelmingly in her direction.
[also note the relevance of this to Wilder/Bradley effect speculations]

I can’t assess this myself, but it does seem to assume all the people who told pollsters they would vote for Obama did. In other words, it uses the coincidence of the same number to assume no Bradley effect, therefore no undecideds broke for Obama, therefore Hillary got all the undecideds. So it is sort of assuming what it is trying to prove. On the subject of undecides, ABC’s polling director Gary Langer writes:

Some folks are suggesting that "late deciders" made the difference - a common explanation for poor estimates. But the exit poll doesn't support the notion. Remove voters who decided on Tuesday and the New Hampshire exit poll result is Clinton +2 – exactly her actual margin. (Among those who decided "just today" it was Clinton +3.)

Of course, Langer also dismisses the “Bradley effect” as old news:

There have been previous races that misstated support for black candidates in biracial races. But most of those were long ago, and there have been plenty of polls in biracial races that were accurate. (For more on past problems with polls in biracial races, see this blog I wrote for Freakonomics last May.) And there was no overstatement of Obama in Iowa polls.

Well, of course Iowa was a caucus. But it is possible that, as black candidates become less anomalous, the Bradley effect is diminishing. Perhaps this would be true differentially – and less true in New Hampshire?  

The best argument against Bradley I’ve seen is from dloewe right here two days ago:

In a situation like last night in New Hampshire, voters had no reason to lie to pollsters. If someone wasn't willing to vote for Barack Obama because of his race, they could have said they were voting for Hillary. There would be no need to hide, no need to lie.

Even though I think there is a difference between running against Republican George Deukmejian (as Tom Bradley did in California) and Hillary Clinton, I’m not sure this completely explains it away. Here is one of the “Bradley effect” examples from Wikipedia:

In the 1988 Democratic presidential primary in Wisconsin, pre-election polls pegged black candidate Jesse Jackson — at the time, a legitimate challenger to white candidate and frontrunner Michael Dukakis — as likely to receive approximately one-third of the white vote. Ultimately, however, Jackson carried only about one quarter of that vote, with the discrepancy in the heavily white state contributing to a large margin of victory for Dukakis over the second-place Jackson.

So what I’m saying is yes, I accept many of the arguments for the last-minute surge by Clinton. But I have yet to see a convincing argument that race was not part of the reason for the discrepancy between Obama’s polling margin of victory and the actual results. And I guess what I’m really sick of is that idea that there has to be a single explanation of the polling discrepancies. I’ve heard so much of this: tears, dishes, debate, race, profit-taking. This is a complex phenomenon, and we keep getting involved in "my pet theory versus your pet theory" sort of debates.

So I guess I just want to say we should be looking at multiple factors and not ruling this one out on the basis of arguments like Schneider's if we want to read the polling data correctly in the future.

Future primaries will tell, I suppose. One good way to get a better handle on this might be to look at the polling on downticket races and compare them to the polling for the presidential primary. In other words, in those states where there are candidates of color elsewhere on the ticket, you would expect a similar drop in support between polling and the actual election in both presidential and downticket races.