Cross-posted to Overdetermined.net

Ben Smith on Politico has a new story today about research coming out of the University of Washington that suggests we may see a "Reverse Bradley Effect" this election cycle.

The Bradley Effect describes a situation where polls overrepresent the support a black candidate receives in the actual election.  It's named for Tom Bradley, a black Los Angeles mayor and 1982 California gubernatorial candidate.  Both pre-election polling and exit polling on election day showed Bradley to be the favorite to win the election, and some media outlets predicted a Bradley win.  Instead, once the votes were tallied, his white opponent George Deukmejian carried the race.

Instead, Anthony Greenwald and Bethany Anderson suggest that we may see a reverse effect this year, citing results from the Democratic primaries which suggested Bradley Effects in only three states (New Hampshire, California, and Rhode Island), but reverse effects where Obama exceeded polling predictions in 12 other states.  Greenwald and Anderson set as their threshold for identifying Bradley or Reverse-Bradley Effects a difference of more than 7% between predicted performance and actual performance.

Now, it's time for some intellectual honesty.

The premise of the original Bradley Effect was that voters couldn't be trusted to tell the truth to pollsters where race was an election issue.  Even exit polls misrepresented the shape of the Bradley / Deukmejian contest.

There has been a good deal of chatter in the media about a Reverse Bradley Effect, but up until now it's centered around things like increased black voter turnout.  That isn't a Reverse Bradley Effect - it says nothing about the truthfulness of polled voters.

Greenwald and Anderson, as well as North Carolina governor Mike Easley (appearing on Rachel Maddow), are now suggesting that voters may lie about their actual preferences.  This is the true Reverse Bradley Effect.  But can we really tell if it's happening?

Unfortunately, no.  Let me quote the relevant part of Ben Smith's blog, which is in turn quoting the Greenwald and Anderson's research:

Albertson and Greenwald believe the errors in the polls are being driven by social pressures that can operate when voters are contacted by telephone prior to an election. They said that polls from states in the Southeast predicted a large black vote for Obama and a much weaker white vote. They found that, in a few Southeast states, exit polls showed that both whites and blacks gave more votes to Obama than the pre-election polls had predicted.

Difference between pre-election polls and exit polls is not clear evidence of either a Bradley or Reverse Bradley Effect.   If exit polls agree with election results, voters are probably being honest with pollsters.

The issue here is undercounting, not prejudice.  Mark Blumenthal, on msnbc.com, discusses how this year represents a "perfect storm" for pollsters.  No one is sure of their turnout models this year, given the potential for an historic black turnout and worries that traditional polling strategies may be underrepresenting certain demographics (such as voters under 30).

I believe the latter problem, underrepresentation in polls, is being overstated.  Polling isn't pure random sampling, in most cases.  Numbers will be obtained from each demographic and projected as if that demographic represented a given percentage of the electorate.  So no matter how voters under 30 are polled, pollsters should be giving them roughly the same predictive weight.  The question then becomes, are the voters not being polled (e.g. voters who only use cell phones, which traditionally aren't called by polling groups) so qualitatively different from the voters being polled that the demographic is being misrepresented?  Certainly, there are qualitative differences, but the strength of those differences is an open question.  Even the youth who are polled favor Obama; I don't see why we should expect unpolled youth show significantly different trends.

The real problem with polls in this election season, and the one being misrepresented as a Reverse Bradley Effect, is that no one is quite sure what voter turnout will look like.  The enthusiasm gap between Democrats and Republicans is well-recorded.  Even if all the demographic indicators do match up with the Bush/Kerry race from 2004 (which is a weak assumption), will Democrats and Republicans show up in equal numbers?  The problem this year is that pre-election polls may not reflect election-day realities.  We have seen, so far, almost no indication that voters are lying to pollsters, one way or the other.