Will Obama Outperform the Polls Again?
by DHinMI
Sun Mar 02, 2008 at 03:32:39 PM PST
In less than 72 hours the polls will open in Texas, Ohio, Rhode Island and Vermont, and the polling suggests tight races. Pollster.com's average of Ohio polls is Hillary Clinton 49.8% and Barack Obama at 43.4%. Their average for Texas is Obama 47.5% and Clinton 44.7%. Polls are not predictors but snapshots of where the electorate is when the poll was taken. But as we are almost to election day, they should provide a good indication of what to expect Tuesday night.
But have the polls been accurate? Everyone knows about the consistent Obama lead in NH that didn't result in an Obama win, but a two point Clinton victory. But what about the polls since? An analysis of the poll results using the data at Pollster.com shows that Clinton has generally performed consistent with the public polls, while Obama has consistently performed much better than the average of the public polls. Because of the small number of polls in the sample, these results can't be viewed as meeting strict standards for empirical study. However, a few calculations suggest that there is no Wilder Effect (AKA as a Bradley effect). In fact, the variance runs in the opposite direction, with Obama's strength being under-represented in the public polling. If there's any problem related to race, it's probably that the turnout models are under-representing Africa-American turnou.
A few notes about the table below. I did not include caucus states, as it's difficult to translate polling to caucus results. South Carolina is included in this analysis, but it's the only state where John Edwards was still actively campaigning. For all subsequent states, I've only included polls taken after Edwards dropped out. If there were not at least six polls taken after Edwards' withdrawal—and for a few important states like CT and MA there were not—I did not include them in the analysis. Finally, regarding California, I excluded Zogby polls, not because they were wrong, but because they were such outliers from every other poll, and ended up being 26 points off the final result.
Again, one should use these figures as suggestive, not as data meeting rigorous social science methodology. Having said that, there are a few observations which these figures support. First, Obama's strength among African-American voters is probably being underestimated by pollsters. The states with the largest African-American populations, such as SC, AL and GA are where Obama has outperformed the polls by the biggest margins. Second, it appears that the withdrawal of Edwards has probably helped Obama. Since Edwards dropped out so close to Super Tuesday, there were many early votes for him that appear to have gone to Obama in the subsequent contests.
There's one other factor that I have not seen mentioned anywhere else: the role of AFSCME, EMILY's List and the American Federation of Teachers in supporting the Clinton campaign with targeted appeals to increase the turnout of women voters. As I will explain in my next post this evening, these three groups may very well have kept Clinton from losing in New Hampshire, and early in the campaign, where they expended great resources, Clinton generally either won or kept the contest close. But their ability to prop up Hillary Clinton's candidacy may be diminishing.
If the pattern seen in these figures continues on Tuesday, Obama will win Texas and possibly Ohio as well. And if Obama wins both Texas and Ohio, it will be a near certainty that he will be our presidential nominee.
http://www.dailykos.com/story/2008/3/2/16596/61053/380/467458