As Barack Obama continues to consolidate a commanding lead, those who have made an unfortunate habit of spreading fear and panic have become increasingly reliant on the idea of "the Bradley Effect" unless impossible conditions AB & C are met. According to those who embrace this statistical bogeyman, people are actually lying to pollsters in order to conceal their racism (and consequently, McCain's actual level of support). Here's one recent example:
Obama needs to be 2% over the margin of error to counter the racism. Just as people are motivated by the attack they claim to dislike, there is also an underlying racism in the US. People will not admit to it when polled but will likely act on it in the voting booth.
As you can see, this commenter even went so far as to provide wholly fabricated numbers and statistical terminology. The confidence of the assertion is certainly impressive, but is there anything backing it up?
The Bradley and Wilder Effect in the 1980's
The conjecture is based on a handful of instances from the 1980's in which some polling companies inaccurately predicted the results of races in which an African-American candidate ran against a Euro-American candidate. Polling accuracy was not common among all such races, but as with many urban legends, a few instances are enough to "confirm the theory." At the time, the pollsters, eager to displace the blame for their poor performance, accused the voters of lying to them to conceal their hidden racist tendencies.
The problem with the conjecture, however, is that it is often based on a very small amount of actual data. In some cases, there was only one company doing polling, and the lack of crosstabs and exit polls does not allow us to determine whether the discrepancy resulted from lying respondents or faulty assumptions in the turnout model. Obviously, inaccurate polls are a common phenomenon, and when it does occur, it usually does so as a result of the pollster's failure to correctly weigh the results for party identification. Furthermore, such failures were more common in the 1980's, and since then, polling methodology has improved considerably.
The verdict then? As of the 1980's, it's inconclusive. The conjecture could be true, but with a lack of clear data and scientific confirmation, we can say nothing conclusive.
The Bradley Effect Today
Obviously, a lot has changed since the 1980's, especially in terms of the make-up of the electorate and prevailing views on race. Still, many refer to the Bradley Effect with strident confidence, as if it had only been further confirmed over time. However, a look at polls of recent races show that voters have no problem declaring their opposition to an African-American candidate, even when that means voting against their own party (a situation that would, presumably, increase their incentive to lie). Here are some examples from 2006:
2006 Ohio Gubernatorial Race
Final Results
Ted Strickland: 60.4%
Ken Blackwell: 36.8%
Final CNN Poll: 59-36
Final SUSA Poll: 62-32
Final Quinnipiac Poll: 59-32
2006 Pennsylvania Gubernatorial Race
Final Results
Ed Rendell: 60.4%
Lynn Swann: 39.6%
Final Rasmussen Poll: 56-36
Final Quinnipiac Poll: 57-38
2006 Tennessee Senatorial Race
Final Results
Bob Corker: 50.7%
Harold Ford Jr.: 48%
Final Rasmussen Poll: 51-47
SUSA Poll: 51-46
MSNBC Poll: 50-38
As you can see, the polls did not conceal any "hidden" racism. In fact, in some cases, they even underestimated the African-American candidate's level of support.
Clearly, racism was likely a hurdle for each of these candidates, but when African-Americans can win GOP primaries in the rust belt and come within 2% of winning in deep Appalachia, it is evident that it is not insurmountable.
Automated Polling
Another important factor that undermines the theory is that many prominent pollsters (including PPP, Rasmussen, and SurveyUSA) currently use computers to do their polling instead of live questioners. Given that the Bradley Effect is premised on the notion that people will lie to a live questioner, logic dictates that it would not apply in automated surveys in which respondents only have to press a button to register their response. As it was, the Bradley Effect was already based on dubious ground, but insisting that people will lie about their preferences to a computer is pushing the limits of pop psychology.
What About Undecideds?
A variation of the Bradley Effect states that it merely manifests among undecideds and that McCain will sweep to victory when all those racist "undecideds" finally get in the voting both. This conjecture is slightly more credible, though it may not be the case that respondents are lying or that it will be race that motivates their final decision. Of course, as demonstrated above, this has not been the case with recent elections, but given the much greater importance of the office of the presidency, it may very well be the case that McCain ends up winning undecideds. However, recent state polling indicates that it will not be enough for McCain. Even if he wins undecideds by an 8-2 margin, Obama will still slide over the 50% mark with ease in many of McCain's must-win "firewall" states. And if McCain wins as few as 60% of the remaining undecideds, we are looking at an electoral landslide for Obama.
We are winning. Yes, we are.
After nearly a decade of electoral disappointment, many Democrats seem to be having trouble getting around the notion that they are on the brink of clearly and decisively winning the presidential election. Beginning with the hijinx of the Florida recount in 2000, the assumed potency of GOP dirty tricks have been amplified to nearly mythical status, and as such, many simply assume that Democrats are somehow inherently incapable of actually winning. The many references to the Bradley Effect are a further reflection of this strange unwillingness to accept what the hard data is telling us.
But the reality is that, while volunteers still have a lot of work to do between now and election day, Barack Obama is kicking ass and according to 538's most recent projection, he has a 9-1 shot at winning. It's happening, folks. It really is. You don't have to deny it. You just have to keep working for it.