01/14/2008 07:11 pm ET Updated May 25, 2011

Obama, Electability and the Recursive Bradley Effect

Hillary Clinton's poll-defying victory in last week's New Hampshire Democratic primary had pollsters, pundits and conspiracy theorists scrambling to explain the difference between Barack Obama's 8-point average lead in the preceding surveys, versus Clinton's 2-point victory on election night. Polls are often wrong, but rarely this wrong, and so not surprisingly, the post-election narrative was as much dominated by the unexpected nature of the results as the results themselves. Whereas Obama left Iowa with a surge of positive press, Clinton came away from New Hampshire with a gigantic question mark.

Over on Daily Kos, DemFromCT has an exhaustive roundup of the latest thinking on what went wrong (or what went right, depending on your perspective,) and while I tend to agree with the conclusion that multiple factors led to the pollsters' pratfall, I think there is one theory that deserves closer examination, not in spite of its lack of supporting evidence, but because of it. Of course, I'm talking about the supposed "Bradley Effect."

The Bradley Effect (also referred to as the "Wilder Effect") describes the observed phenomenon in which black candidates score significantly higher amongst white voters in public opinion polls than they ultimately do on election day. This is popularly represented as evidence of a degree of racism amongst white respondents, who apparently shy away from telling pollsters their true leanings, for fear of being perceived as racist. But as Pew Research Center president Andrew Kohut explains in the New York Times, the demographic underpinnings of the effect are actually much more subtle:

In 1989, as a Gallup pollster, I overestimated the support for David Dinkins in his first race for New York City mayor against Rudolph Giuliani; Mr. Dinkins was elected, but with a two percentage point margin of victory, not the 15 I had predicted. I concluded, eventually, that I got it wrong not so much because respondents were lying to our interviewers but because poorer, less well-educated voters were less likely to agree to answer our questions. That was a decisive factor in my miscall.

It is not so much that white voters generally lie to pollsters, Kohut argues, but that "poorer, less well-educated" white voters -- who we're told are less likely to support a black candidate -- tend to be under sampled in the typical survey. But I wonder if, in the context of a presidential primary, the Bradley Effect might actually insinuate itself into voter behavior in an even more subtle way, spinning questions about electability into a self-fulfilling prophecy? The most widely cited examples of the Bradley Effect come from general elections, but all things being equal, primary voters, particularly in our currently polarized environment, tend to be focused on selecting the nominee they believe to be most capable of winning in November. No doubt race has always been a dominant theme this election season, hence the big story coming out of Iowa being the unprecedented victory of our nation's first viable black presidential candidate. But if New Hampshire voters -- black and white alike -- remained unconvinced that our nation is ready to elect a black man to the White House, might they ultimately cast their ballot for a white candidate, despite their honestly stated intention to vote for Obama?

So, does the Bradley Effect at least partially explain the pollsters' flop in New Hampshire? Probably not... but that doesn't really matter, for the very discussion of the Bradley Effect has the potential to impact the behavior of Democratic voters in primaries down the line.

In reality, the much ballyhooed polling discrepancy involved Hillary Clinton's numbers only; Obama received pretty much exactly the same percentage of the vote on election night as the pre-election polls had predicted, so it's hard to argue that the polls oversampled Obama's support when he largely performed as expected. The data doesn't necessarily disprove a Bradley Effect, but it doesn't particularly support it either.

But it's too late for pundits to take back their speculation, and it is unlikely that the specter of the Bradley Effect won't continue to be raised in the days leading up to Nevada, South Carolina and beyond. On its surface the Bradley Effect, whatever its mechanism or evidence, appears to be a reasonable enough explanation for at least some of what we saw in New Hampshire, and if Democratic primary voters believe it to be true, it could influence their vote as well, not because they are racists, but because they perceive a substantial number of their fellow Americans to be racist. If Obama subsequently underperforms pre-election polls in other contests, "evidence" of the Bradley Effect builds, as does its place in the public narrative. What results is a self-catalyzing recursive process in which Democratic primary voters, focused on electability, transform unsupported speculation of a Bradley Effect into a reality, withholding their genuine support for Obama because they believe he cannot win. It's not racism per se that defeats Obama, but the perception of racism in others. (Which I suppose is racism, if only in a nuanced, institutional form.)

Of course, this is all just speculation. But speculation has an odd way of coming true, even when it's not.

[David Goldstein blogs on WA state politics at]