The Monkey Cage

John makes an interesting point. If one looks at the absolute percentage that African American candidates secured in 2006, it does not look like there is a racial effect that undermines the value of polls when an African American is on the ballot.

But, there is another way to look at the numbers that might lead to a more disturbing conclusion.

According to pollster.com, Duvall had a 25-point lead over his Healey going into the election. In the end, he wins by 21 points. But, the percentage is 56% (one point shy of where he was polling). There seems to be a difference if one thinks about this as a percent of the electorate or as a gap between the viable opponents. What could account for this? I presume that some of Duvall's opponents declared themselves to be un-decided (or Mihos) supporters when polled.

In Maryland, the final margin according to Pollster was 4. In reality it was ten. Again, we see a similar pattern. If one looks at it as final percentage, it is about right. If one looks at it as margin, the polls underestimate the white candidate.

In Pennsylvania, Ohio, and Tennessee, the polls got the margin about right. In other words, the polls underestimate the white in two-thirds of the states with an African American Democrat on the ballot and got it right in the states with an African American Republican on the ballot.

What is going on here? My hunch is that we should be turning to Adam Berinsky (Silent Voices: Opinion Polls and Political Representation in America) and Tali Mendelberg (The Race Card) who argue that people are reluctant to express racially biased views publicly. The Democrat v. Republican difference might be attributable to the fact that racial equity is such a prominent part of the Democratic Party's agenda.

I have not gone back further than 2006, but historically races with an African American have suggested that the African American was going to win by a closer margin than the final numbers suggest. We saw this in a number of races that come to mind (Illinois Senate--1992, North Carolina Senate—1990; Virginia Governor--1989, New York Mayor--1989, Chicago Mayor--1983, and California Governor--1982).

Maybe things changed, but not enough to discount the chance that Obama's margin was not what was expected because of race. He went from a seven or eight point lead to a three-point loss. What accounts for this ten-point difference? I presume some of the difference (whether it is 2, 3, or 5 points is beyond me) occurred because of race.

My wild hunch is that he also lost some support because the polls were predicting that he had a comfortable lead. Strategic independents in New Hampshire may have decided that it was wiser to invest their vote in McCain (who was reported to be in a tight race) than in Obama. This could be explored if someone had included in an exit poll a question regarding what they expected the Obama and McCain margins to be. Finally, I expect he lost some support because of the debate and tears. In other words, I would not attribute the difference purely to the hidden racial effect that is not picked up by polls.

For the record, I am one of those people who asked John about the impact of race on the NH polls.

I also view myself as someone who studies institutions. In other words, take this posting with a grain of salt.

Though I am confident in your presentation of the share that the African-American candidates received, is the same true for the share for the white candidate? That is to say that didn't the white opponent in many of those races outperform the polls and thus, as Forrest suggests, isn't the difference in the margin really the important thing to consider? I haven't looked at the data, just simply thought I understand the Bradley-Wilder thesis differently.

I've gotten quite a bit of e-mail from friends about this and – at John’s invitation – I’m posting here. As Forrest notes, I wouldn't predict that people would lie about supporting Obama, then vote against him. Instead, I would predict that people would say they “didn’t know” if they were uneasy about appearing racist. News reports that the “undecideds” broke heavily for Clinton seem to support this conjecture.

Of course, this can’t explain the entire gap. As John notes, there have been some recent races involving black candidates where the polls were essentially correct. As an aside, I think that it is important to place elections in context. The 1989 NYC mayoral election is the case I’m most familiar with, both because I’ve done some analysis of the pre-election polling data, and because I have some qualitative evidence. In that election the preferred candidate of older Jewish Democrats (read: my mother, my father, and their friends), Ed Koch, lost the Democratic primary. Considering that some of these people had never in their life voted for a Republican candidate, a vote for Giuliani in the general election could be seen as nothing but a vote against the black candidate, Dinkins. It’s interesting here to compare the 1989 election to the next mayoral contest. Dinkins ran against Giuliani again in 1993. In that election the pre-election polls were very accurate. One explanation for the discrepancy in the performance of the polls between 1989 and 1993 is that in 1989, Democratic voters could not openly oppose Dinkins without appearing to be racist. By 1993, however, they could oppose Dinkins because in the intervening 4 years he had established a poor record of performance. In short, I think that it’s a mistake to make any general claims about what happens when black candidates run against white candidates without considering the motivations of voters. In the 2005 MA Healey/Duvall contest, for example, Healey was an extraordinarily weak candidate. That changed the polling dynamics as well.

My guess is that there was some combination of racial dissembling and faulty voter screens. As Charles Frankin notes, we won’t know anything until (if ever) the pollsters share their raw data. Some interesting issues to consider as we move forward.

Hi, John. You might take a look at Yair's thoughts here. His thoughts are similar to Adam's but with some specific conjectures.

Obama did, in fact, win NH. The hand count shows this. Diebold's machines got it wrong.

Some media (Fox, NBC) have been much quicker to catch onto this fact than others (CNN). I hope this will affect the weight people give to what each of those media say in the future.