Artur Davis and the Reverse Bradley Effect

Most observers have concluded that the number one challenge that U.S. Rep. Artur Davis faces as he runs for governor next year is persuading ...
by | March 19, 2009

Davis Most observers have concluded that the number one challenge that U.S. Rep. Artur Davis faces as he runs for governor next year is persuading Alabama voters to cast their votes for a black man. Though that's a serious issue for him, Davis first will grapple with what you might call his number two challenge: persuading Alabama Democrats that, come November, he'll be able to overcome that number one challenge.

Davis is likely to face Lt. Gov. Jim Folsom Jr., who is white, in the Democratic primary. If Democrats believe that Davis has no chance to win a general election, they'll be much more likely to support Folsom. As a result, the perception that Alabama voters won't support an African-American could do nearly as much harm to his campaign as the reality.

In this regard, Davis received some bad news in November, when exit polling showed Barack Obama only taking 10% of the white vote in the state. That was his lowest percentage among whites in any state in the country. In West Virginia, a state that's alleged racism became a subject of debate during the Democratic primaries, white voters were four times as likely to support Obama.

In that context, I was intrigued by this sentence from a Politico story about George Wallace's daughter endorsing Davis:

According to CNN exit poll data - which Davis vigorously disputes - 98 percent of black voters supported Obama, while only 10 percent of white voters did.

Since the dawn of polling, politicians have selectively disputed surveys that produced results they didn't like. In this case, though, Davis appears to be right.

When I asked the Davis campaign for an explanation, they pointed me to a little-noticed December column in the Tuscaloosa News. The column quotes Natalie Davis, a Birmingham-Southern political scientist, as saying that based on a regression analysis she conducted, Obama received 16% to 17% of the white vote in Alabama, not 10%.

I spoke with Natalie Davis, who confirmed that finding to me. I'm not statistically proficient enough to explain in technical terms what she did, but I can summarize.

She, of course, knew how many votes Obama received in each county. She also knew the racial makeup of each county's pool of registered voters, since Alabama makes that information public.

With that data, it's pretty easy to create a model of what Obama's vote totals would be in each county if he really received 98% of the black vote and 10% of the white vote, as the exit poll reported. If Obama's vote totals don't generally match, you can be pretty sure the exit poll was off.

I tried my hand at doing something similar by looking at three rural Alabama counties: Blount, Cullman and Walker. It's a good bet that Obama was strongest among white voters in Alabama's bigger cities and college towns. So, if he really only won 10% of the white vote statewide, I'd expect him to have received less than 10% of it in these three counties.

When looking at these counties, I assumed that voters of all races turned out at equal rates. I also assumed that Obama received 98% of the black vote (that's what the exit poll indicated) and 67% of the "other" vote, which is pretty much what he got from voters nationwide in this category.

If those assumptions are right, Obama must have received 12.2% of the white vote in Blount, 15% in Cullman and 18.3% in Walker to reach his actual vote totals. If he did that, it seems very unlikely that he only received 10% of the white vote statewide.

Of course, my assumption that registered voters showed up at the polls at equal rates regardless of race is probably wrong -- African-Americans turned out in greater numbers. These counties have very small black populations, however, (about 1% of registered voters in Blount and Cullman and 6% in Walker), meaning even if blacks were extra motivated to vote, my results wouldn't change very much.

I'm probably oversimplifying the topic a little, but there's at least some strong evidence that Natalie Davis is right. Why, then, would the exit poll have gotten the results wrong? She has a provocative answer and you'll see that I didn't pick Blount, Cullman and Walker at random.

"If you had a white interviewer in a county like Blunt, or Cullman or Walker and you're white and come out of the polling place, it would be hard for you to say you voted for Obama," she told me. "I have no data to support that, that's just my sense of where rural, white Alabamans are."

What she's describing is known as the Reverse Bradley Effect. It's a theory that came up at various points during the campaign, especially when Obama repeatedly outperformed the polls in Democratic primaries in the South.

I'm at least somewhat skeptical that the Reverse Bradley Effect is what was going on in Alabama. All polling is subject to sampling error and the possibility of error increases when you're talking about a single sub-group such as white voters. Plus, exit polling always faces a host of unique methodological challenges. So, the exit poll could have underestimated Obama's white support in Alabama for completely mundane reasons.

Also worth noting: Exit pollsters try to keep answers anonymous. Respondents fill out a written questionnaire, so it's not like voters are being asked to say out loud, in front of their neighbors, who they supported.

What would suggest a Reverse Bradley Effect is if throughout the South exit polls had underestimated Obama's white support. Natalie Davis says she's interested in conducting similar research in other Southern states.

But, to return to Artur Davis' campaign for governor, do Natalie Davis' findings really boost his chances? Davis, the political scientist, says she has been a supporter of Davis the politician (she's an active Democrat, who even ran for U.S. Senate in 1996). But, she's not optimistic that he can win.

She points out that Democrats typically need 38% or so of the white vote to win in Alabama. Artur Davis would have to do vastly better than Barack Obama with white voters to win -- otherwise, Obama wouldn't have lost Alabama by 20 points.

Still, I think it's a fairly big deal for congressman Davis if Obama took 17% of the white vote, instead of 10%. In 2004, John Kerry only won 19% of the white vote in Alabama, according to exit polling. Though 2008 was a more Democratic year nationally, Alabama has been trending Republican over time. So, if Obama took 17% or so that would suggest that his race didn't cost him many votes.

Of course, that Democratic presidential candidates regardless of race have had so much trouble winning whites in Alabama is, at best, a mixed blessing for Davis. Democrats still tend to do better when running for other offices in Alabama, but, nonetheless, you might say that Davis' party affiliation is the third major challenge he faces.

Josh Goodman
Josh Goodman  |  Former Staff Writer

More from Politics