Thanks to the folks at the Monkey Cage for letting me post this week. I’ve enjoyed my time here. Today, I’m going to wrap up my discussion of Obama and the polls.
In my last two posts, I discussed the “Bradley effect” – the notion that number of white respondents will tell pollsters they are going to vote for a black candidate, then turn around and vote for his or her opponent on Election Day. I argued that there is no such thing as a uniform effect. Instead, we need to examine the motivations of voters in the context of particular election. In the 1989 New York City Mayoral election, those lifelong Democrats who supported incumbent mayor Ed Koch in the Democratic primary and did not want to vote for Dinkins could either say they were voting for Giuliani or abstain from giving a vote preference. In the racially charged atmosphere of the 1989 election, such voters who were sensitive to social concerns had an incentive to state they were “undecided” rather than reveal their true preferences. By 1993, with a poor record of performance behind him, these same voters could openly oppose Dinkins without fear of being cast as racist – they could simply say they were voting for the better candidate.
What can the past tell us about the future? Will Obama be plagued by a “Bradley effect” as some have suggested?
The key question to ask is “Does the 2008 Presidential election look more like the 1989 New York City mayoral election, or does it look more like the 1993 contest?” While current polls are not necessarily predictive – recall that it was around this time 20 years ago that Dukakis opened a 17-point lead over George Bush – nationally, the percentage of people who say they do not know how they will vote runs at about 10 percent, and some recent polls have even higher levels of “don’t knows.” These numbers are slightly higher than those found at this time in the 2004 campaign (though in 2008, unlike 2004, there is no incumbent on the ballot; in 2000 the number of “don’t knows” were considerably higher according to polling conducted for the National Annenberg Election Study). Are these “don’t knows” really hidden McCain votes that could decide the election?
I think not. In the early days of the primary season, to be sure, the polls gave some cause for concern. In New Hampshire, polls predicted an Obama win, but Clinton came out ahead on Election Day. New Hampshire’s place at the top of the primary calendar almost certainly helped set the perception that the polls were in for a rocky road. However, this result this seems to have been an aberration (and some have argued that their poor performance had little to do with race. No other contest repeated the pattern found in the Granite State. In fact, as Nate Silver at 538.com points out, across all the primary contests, Obama actually outperformed the polls. This “reverse Bradley effect” was strongest in states with large black populations, especially in the south. The reason for this discrepancy is unclear. Perhaps the likely voter screens underestimated black turnout. Or perhaps blacks are less likely than whites to reveal their candidate preferences.* In any event, it does not appear that the Democratic primary polls systematically underestimated Obama’s support.
If anything, the general election should look even less like the 1989 New York Mayoral election and more like the 2006 Tennessee Senate race. In the primaries, voters could not use partisanship as a guide or an explanation for their vote. The fact that the Bradley effect was almost entirely absent from the primary polling bodes well for the future predictive power of the polls. In the general election, with party as a meaningful guide, voters can cast their votes accordingly. The types of cross pressures found in the 1989 New York City mayoral election are entirely absent. Those voters who are uncomfortable with Obama can cast a vote for McCain on the grounds that they prefer the Republican in the contest.
There is, however, a potential exception to this rule. Interestingly, the one region of the country where Obama did not perform up to par with his polling numbers in the primaries was in the Northeast. In those eight states, Clinton, on average, did two percent better than expected (though take away New Hampshire and the effect is cut in half). These numbers are small, to be sure, but they are suggestive. Maybe voters in the northeast are more sensitive to the social concerns that shaped the polling results in the bi-racial contests of the 1980s. In other words, perhaps these polls are evidence of a small Bradley effect at work.
But even if the primary polls showed evidence of a small Bradley effect, the potential impact of such an effect in the general election should be small. First, as noted above, with the introduction of partisanship as a factor in the general election, the dynamics of the election change in ways that reduce of the costs of openly opposing the black candidate. Second, even if there was an effect comparable to the one found in the primaries, such an effect would not change the results. Polling suggests that the northeast will go heavily for Obama, as it did for Kerry in 2004. Thus, even if the fear of appearing racist causes some voters to misrepresent their voting intention to pollsters, Obama should remain comfortably ahead in the Northeast.
__________________________________________
*On this point, I am reminded of some analysis I did a while back in the 1999 Philadelphia mayoral election. There, John Street, a black Democrat, ran against Sam Katz, a white Republican. Like earlier polling in biracial contests, there were a great number of “undecided voters.” Unlike those other contests, however, large proportions of blacks, as well as whites, claimed that they did not know who they were going to vote for. My models indicated that the black undecideds would break for Street and the white undecideds would break for Katz. Since these two groups were approximately the same size, they effectively canceled each other out. Thus, the pre-election polls gave the right answer for the wrong reasons.