With an African-American candidate on the presidential ticket for the first time in history, there has been a lot of talk about “the Bradley effect” (or the Dinkins effect, or the Wilder effect). Obama’s numbers are inflated, the thinking goes, because a large number of white respondents will tell pollsters they are going to vote for a black candidate, then turn around and vote for his or her opponent on Election Day. In light of these dynamics, should we take the current polling numbers at face value?
A few weeks ago John described recent work by Dan Hopkins on this topic (a paper that I’ll return to later this week). But the accuracy of the polls is something I’ve thought about for a while. Ten years ago, I wrote a paper entitled “The Two Faces of Public Opinion” which advanced an argument consistent with the Bradley effect. I argued that public opinion polls may overstate support for racially liberal policies (in this case, government efforts to integrate schools). My analysis revealed that those individuals who harbor anti-integrationist sentiments, but worried about social concerns, were likely to hide their socially unacceptable opinions behind a “don’t know” response. In a similar vein, pre-election polls in electoral contests which involve candidates of different races offer a situation where individuals might be loath to express their true candidate preferences for fear of appearing racist. Some individuals who are apprehensive about voting for black candidates may declare themselves undecided rather than come out and say that they oppose a black candidate.
I examined data from a 1989 New York City Mayoral election. There, the black candidate David Dinkins held a fourteen- to eighteen-point advantage over his white opponent Rudolph Giuliani in polls taken only days before the election, but ended up winning the race by less than two percentage points. Correcting the polls using statistical techniques that accounted for the “don’t know” improved the predictive power of those polls. Clearly, some people who said they didn’t know how they were going to vote in fact did know – they just didn’t want to tell us.
Given these results, should we discount Obama’s polling performance accordingly? I think not, for reasons I will elaborate over the next couple days. Put simply, the popular interpretation of the Bradley effect is overly simplistic. Only in those circumstances where voters face significant cross pressures will they have incentive to dissemble to pollsters. In the New York City election, for example, Dinkins ran as a Democrat in an overwhelmingly Democratic city. In the current contest I do not think that voters will be subject to pressures great enough to fundamentally distort the polls.
Tomorrow, I’ll talk more about the history of polling in bi-racial elections.