Whither the Bradley Effect?

Heading into election 2008, most polls show Barack Obama with a small but significant lead over John McCain. But can the polls be trusted?

In recent weeks there has been increasing discussion of the 'Bradley effect'. This is the phenomenon that black candidates do better in polls than they do in actual election results. It is named after former Los Angeles mayor Tom Bradley, who had a significant lead in the polls in the race for California governor in 1982, yet lost to the Republican candidate.

Similar occurrences have been observed in the 1989 race for governor of Virginia and 1989 race for mayor of New York.

What explains this phenomenon? The standard explanation is that a certain percentage of voters lie to pollsters. They say that they will vote for the black candidate when they will actually vote for the white candidate.

This has been denounced as racism by many on the left. But this does not follow. Why would a racist claim to be voting for a minority candidate at all? Instead, it is political correctness. That is, the politically correct thing to do is to vote for the minority, so some voters will say that they are going to do so to avoid being considered racist, even though they have legitimate reasons for their vote.

On the other hand, others have denied that the Bradley effect is real, or that it still exists. Bradley's loss has been alternately blamed on high turnout of Armenian voters voting for his opponent or high turnout of rural voters opposing a gun control initiative on the ballot that year. Meanwhile, one recent study purports to show that the Bradley effect used to exist but no longer does, supposedly finding no evidence of it since 1996.

But there have been elections since then that fit the same pattern. In 2003, Republican Bobby Jindal (who is Indian), polled ahead going into the election for governor but lost after a late campaign by the democrat party including ads that darkened his skin and urged white voters to vote democrat before it was "too late". Jindal was elected governor of Louisiana in 2007.

In 2006, the Michigan Civil Rights Initiative (MCRI) appeared on the ballot in Michigan. This ballot proposition eliminated racial preferences in government hiring, contracting, and college admissions. Polls months before the election showed the MCRI with about 60% support. As the election approached and debate over the MCRI became increasingly heated, its support in polls continually declined, falling as low as 40% (with a large number of undecideds) the week before the election. These polls led the normally accurate Evans-Novak Political Report to predict that the MCRI would "fail by a large margin".

When election day came, the MCRI passed with 58% of the vote, roughly the level of support that it started with. This is all the more remarkable since the usual rule with ballot propositions is that most of the undecideds will vote no since change is riskier than the status quo. The only conclusion one can come to is that a significant number of voters were simply lying to the pollsters about their intentions.

So will there be a Bradley effect in the presidential election this year? There is no way to know for sure. Since it is a problem with polling, it is always possible that it could be corrected by different polling techniques.

But the polling techniques that are being employed raise serious questions. Pollsters more-or-less arbitrarily pick the percentages of Republicans, democrats, and independents to weight their samples. Many pollsters are assuming much higher turnout of the black and youth vote when weighting their samples. While higher black turnout appears to be all but certain, it is unclear to what extent there will be higher white turnout and to what extent this will counterbalance it. As for the youth vote, projections of much higher youth turnout in past elections have never been borne out.

It could be that Republican prospects are actually much better than what the polls suggest. Or this could be wishful thinking. We'll find out soon enough.