Was it massive polling error? Were women motivated by a double standard imposed on Clinton for showing emotion? Was it the Wilder / Bradley effect, where white voters lie about supporting African-American candidates to pollsters out of perceived social pressure? Was it something else? Twelve public polling organizations were in the field in New Hampshire entirely after the Iowa caucuses. One of these organizations concluded interviews on Saturday, January 5th. Seven concluded interviews on Sunday, January 6th. Four concluded interviews on Monday, January 7th.. The average of the final results from these pollsters is as follows:
Now, compare this to the results, with only one precinct outstanding
Clinton: 38.99% (+9.07, +30.3%)
Obama: 36.39% (-0.86, -2.3%)
Edwards: 16.91% (-2.01, -10.6%)
Richardson: 4.60% (-1.15, -20.0%)
While Obama and Richardson both saw their support drop from the final polling average, Edwards saw his drop as much as Obama and Richardson combined. It is possible that Edwards saw his numbers drop for a different reason than Obama or Richardson, or even that all three saw their numbers drop for different reasons. However, given Obama's numbers dropped the least, both in overall terms and in percentage terms, I am not convinced of a "Wilder effect here at all. Or, at least, I am not convinced that the "Wilder effect" was the only dynamic in play. It seems equally plausible that Edwards and Richardson saw their support drop much the way third-party support always drops from the polls to the final results. The lower the perceived chance a third-party candidate has the win, the larger their expected drop from the polls to the final results. It would appear that those voters broke toward Clinton.
Here is what I think happened, in chronological order:
First, Clinton probably had a superior absentee voter program, which gave her a small boost. Likely voters are not guaranteed voters, but those who have already voted are. Before they were rebalanced, the exit polls showed Obama narrowly ahead of Clinton, 39%-38%. Absentee voters were not included in the exit poll, and a successful and strong absentee voter program can indeed account for a 3-4% net swing, especially since Clinton held a commanding 48%--31% lead among voters who had their minds made up the longest. This is also, for example, is why Brian Bilbray outperformed Francine Busby in final polls in CA-50.
Second, the polls were somewhat wrong, probably due both to a very mild "Wilder effect" and to improper weighting of the electorate / measurement of likely voters. However, the polls don't have to be more than 1% wrong in order to make this scenario work (although the more wrong the polls were, the easier this scenario works). Given that Rasmussen, a polling firm that utilizes the automated, IVR methodology, showed the campaign to be a little closer than other pollsters who used live interviews, there probably was a mild "Wilder effect" of about two percent or so. IVR polls should eliminate the Wilder effect altogether, and so it is useful to look to them as a baseline when determining the presence of a Wilder effect.
Third, there was a break toward Clinton on Election Day itself, when no polls were taken. A survey of 2004 and 2000 polls taken between Iowa and New Hampshire shows there is a tendency for Iowa bounces to begin to recede after three to five days, meaning that by Election Day Clinton should have been pulling back on Obama anyway, with or without a sympathy vote. Exit polls back this up. Among voters who decided who to vote for in the last week, Obama led Clinton 43%-28%, probably due to a huge surge in the two days after Iowa. Among voters who decided in the three days before the election, Obama still led by a smaller amount, 37%--34%. Among those who made up their minds on Election Day itself, the bounce had faded entirely, and Clinton pulled into a 39%-36% lead. That accounts for at least another half of a percent. Of course, the sympathy vote probably didn't hurt.
Fourth, Edwards and Richardson supporters who favored Clinton as a second choice disproportionately broke away and choose Clinton, since the narrative implied both that she was the only other candidate who could win and that she needed help to do so. We all saw this, for example, in 2000, when Nader was a factor and Gore was in a position similar to Clinton. In the end, Nader underperformed his final polls by 1.2%, and Gore outperformed his final polls by 2.0%, providing Gore enough of a boost to win the popular vote. By way of contrast, Edwards and Richardson supporters who favored Obama as a second choice probably didn't think Obama needed any help. This could have added as much as 3% to Clinton's total.
Fifth, Clinton was assisted by the ballot order, probably to the tune of about 3%. Clinton was at the top of the ballot, and it is a well-known long and long-studied phenomenon in politics that placement at the top of the ballot provides a not insignificant edge to any given candidate.
This is a "perfect storm" scenario that requires no significant polling error, no significant Wilder effect, and no significant sympathy effect for Clinton. The top of the ballot, a superior absentee voter program, a naturally fading bounce, and an also somewhat typical "third party effect" from Edwards and Richardson supporters can, in and of itself, account for the ten-point swing from Obama to Clinton. However, that it requires all of these less than mysterious pieces to fall into place makes it somewhat dubious. The most likely scenario is that the four reasons I provide all played a role, but by less than the amounts I suggested here. Further, the three most common reasons being given for yesterday's shocker, Wilder effect, sympathy effect, and massive polling error, all also probably played a role, but less of a role than is being trumpeted by major news outlets. It was probably just a perfect storm for Clinton incorporating all seven rationales.
Still, this result should give real pause to anyone like me who believes pre-election final poll averages are almost always a very accurate predictor of final results. If a whole host of factors fall into place, clearly it is possible to bust up final averages by as much as 10%. Clinton got a perfect storm, drawing on as many as seven different factors. It will take a long time to sort this out with any certainty. In fact, certainty may never come. The future performance of polls in predicting final results will now be watched closer than ever. I guess they are "on notice" now.
