The Pew Research Center has an interesting piece up on the "Reverse Bradley Effect" I've written about several times. It's by two academics at the University of Washington:

Analysis of primary counts and polling data from the early primaries, including those held before and on Super Tuesday (February 5), indicated that pre-election polls did indeed exaggerate support for Sen. Barack Obama in three states with relatively low black populations -- New Hampshire, California and Massachusetts. But the reverse was true in South Carolina, Alabama and Georgia, where blacks make up a larger bloc of voters.

As shown in the graph, the findings in South Carolina, Alabama and Georgia suggested to us the discovery of a new "reverse" Bradley effect, i.e., that in states with relatively large African American populations, pre-primary polls tended to underestimate support for Obama.

This is good stuff, as far as it goes. But I think the authors need a theory about why this might be happening, and maybe a little evidence in support of said theory. Fortunately, I've got that stuff right here.

First, the theory (see here and here):

Is it possible that some black voters would tell pollsters they support Hillary (or that they're undecided) because they don't want to sound like they're voting mainly out of racial solidarity, even though they actually intend to vote for Obama? If so, you could have a reverse Bradley effect, in which polling understated support for the black candidate in a primary with a large African American population (i.e., Obama in South Carolina). ...

I'd speculate that when African-Americans are in the presence of whites, the greater social fear is being considered a "race man" (or woman). Which means you'd expect some reluctance to express support for Obama when the interviewer is white. 

As for the evidence, see here [written the day of the SC primary]:

If Obama consistently did better among black voters in automated polls, which eliminate the "social discomfort" that might discourage them from telling (presumably white) interviewers they support him, we'd have evidence for this hypothesis.

So what do the polls say? They say I might be onto something:

In the three most recent automated polls in South Carolina (PPP, SurveyUSA, and Rasmussen), Obama takes 67, 73, and 68 percent of the black vote, while Hillary takes 13, 18, and 16. In the three most recent live-interviewer polls (Zogby, Mason-Dixon, and ARG), Obama takes 55, 59, and 61 percent of black voters, while Hillary takes 18, 25, and 25.

So, among black voters, that's an average lead of 69-16 for Obama in automated polls, but only 58-23 in live-interviewer polls--a huge difference (53-point lead in the former; 35-point lead in the latter). It's not exactly definitive--I'm only using three data points in each case, and there are other methodological differences between the polls--but it does strongly suggest that some black voters are reluctant to tell human pollsters they support Obama, but feel comfortable saying it to a machine.

And here [written the night of the SC primary, once the results were in]:

Don't sleep on the reverse Bradley effect. The South Carolina polls conducted by human beings way, way under-predicted Obama's black support. The polls conducted by machines only somewhat underpredicted it. I say that's pretty strong evidence that black voters were reluctant to tell human pollsters they supported Obama. To all you poli-sci grad-students out there looking for a dissertation topic--dissertate away! All I ask is that you cite me in your acknowledgements. And maybe name the phenomenon after me. That's it!

Message received, I guess. Er, sort of...

--Noam Scheiber