I talked to several people today who are asking if the NH pre-election polls got it wrong because of the “Bradley-Wilder” thesis: white voters express support for black candidates in polls, so as not to seem prejudiced to the interviewer, but vote for the white opponent on Election Day.
Pollsters Frank Newport (of Gallup), Gary Langer (of ABC), and Jon Cohen (of the Washington Post) have already discounted this thesis (here, here, here). Let me throw a little recent data at it.
For each of these black candidates (who were running against white opponents), I will report the final poll average (as reported by Pollster.com) and then their share of the vote. If this thesis is correct, then the first number will always be higher than the second number.
Barack Obama: 38.4 — 36.4 (see Matt Yglesias’ graph and scuttlebutt here)
Harold Ford: 45 — 48
Michael Steele: 45 — 44
Lynn Swann: 36 — 39
Deval Patrick: 57 — 56
Ken Blackwell: 37 — 36
Given the inevitable sampling error in polls, there is little consistent confirmation of the thesis.
I am not going to speculate on what might have produced the discrepancy between Clinton’s poll standing and actual vote share (note that such discrepancies didn’t exist on the Republican side). The pollsters cited above provide some initial thoughts. Ultimately, we need to get much more under the hood of the polls, as Charles Franklin notes, and that data isn’t yet available. But the racial identities and attitudes of voters do not seem an important factor.
[Addendum: Make sure and read the comments to this post, all of which are interesting. See also Mark Blumenthal’s comments and the Pew Study that he links to.]