The 'Bradley Effect'

Polls are often wrong, yet Hillary Clinton's Lazarus-like resuscitation in the New Hampshire Democratic primary left political wonks slightly more confused than usual. Pre-election polls showed Barack Obama comfortably ahead with an average margin of more than 8 points, yet Clinton managed 2-point victory on election night.

Was the miscalling of the result due to the 'Bradley Effect' -- the phenomenon of white voters telling pollsters they are undecided and then breaking in large numbers toward the white candidate?

The short answer is probably not, although the Bradley Effect is the most provocative diagnosis and may be invoked by the media if Obama underperforms pre-election polls in forthcoming contests. New Hampshire was likely more a story of serious polling methodology lacunae that failed to account for the likely voting habits of 'undecided' voters. 

Background

The phenomenon derives its name from Tom Bradley, the former mayor of Los Angeles, who in 1982 looked very likely to become the first black governor of California. Bradley boasted a double-digit lead over his white rival, George Deukmejian, but ended up losing. Similar voter behaviour seemed to be at work in other high-profile races:

• The 1989 race for Governor of Virginia between black Democratic candidate Douglas Wilder and white Republican candidate Marshall Coleman; Wilder eventually scraped home by less than half of one percent, despite pre-election polls giving him an average lead of nearly nine percent.
• Pollsters gave Rudy Giuliani almost no chance in the 1989 New York mayoral race against David Dinkins; Dinkins ended up winning by a nose, with 50% of the vote to Giuliani's 48%.

Racism?

Many believe that instances of voters telling pollsters that they are undecided, but then choosing white candidates in the privacy of the voting booth, is evidence of White America's latent racism.

• In each of the above instances, pre-election polls were accurate in their prediction of the black candidate's vote, but greatly underestimated the vote for the white candidates.
• However, Eugene Robinson points out in the Washington Post that the phenomenon has been absent in other recent high-profile races in which black and white candidates competed.

The reason for the discrepancy in New Hampshire may still be a cultural one -- yet it is more a story of some voters' attitude to surveys rather than their inherent bigotry. Andrew Kohut points out in the New York Times that poorer, less well-educated white people, are less inclined to respond to surveys. They may not like being polled, but they like voting. And they tend to have more unfavourable views of blacks than respondents who do the interviews.

Pollsters failed to factor in the likely voting habits of this great chunk of 'undecided' voters, most of who ended up voting for Clinton; Obama received almost exactly the same percentage of the vote on election day as the soothsayers had predicted.  As David Goldstein writes in the Huffington Post, the data does not necessarily disprove a Bradley Effect, but it does not particularly support it either. It is more a case of commentators mistaking the discrepancy as evidence of racially motivated flip-flopping, rather than pollsters screening out voters as unlikely to turn out -- who actually did go to the polls and vote for Clinton.

The coming weeks will give the media more opportunities to measure pre-election polls against actual results, including in states with more racial diversity than New Hampshire. And the ranks of Bradley Effect-spotters will swell if dramatic gaps between pre-primary polling and the final result remain unexplained.