Wednesday, January 09, 2008

Are Voters Yoyos?

Sticking with the theme about crowds and getting information about them, the press reported yesterday’s results in New Hampshire as a surprise on the Democratic side --- with Hillary Clinton making a comeback over Barack Obama. A poll taken after the Iowa Caucuses had Obama with a 9 point lead. But Clinton actually won. So this is news. Of course, if that poll were wrong, if the Iowa results were not a good predictor of what would happen in New Hampshire because the demographics are different, then there may not be much news here. That’s what I’m trying to think through in this post.

First, let’s look at a prediction that did pan out. The Governor of New Hampshire said there’d be big turnout, upward of a half millions votes. CNN reports the results with 96% of the precincts reporting. My count from that is a little more than 505,000 and if districts were uniform in the number of votes there are about 20,000 yet to be reported. Very good Governor, you got it right.

Next, let’s look at a really dippy prediction made by CNN on TV. During the 9 to 10 PM hour in the Midwest, Clinton had a 2% to 4% lead, a pattern, that had been holding up most of the evening, but CNN refused to make a prediction because two college towns, Durham (University of New Hampshire) and Hanover (Dartmouth) had yet to report and it seemed that Obama was quite popular on College Campuses. (CNN ultimately declared Clinton the winner soon after the AP had done so.) Indeed, those towns went heavily to Obama, but the volume wasn’t high enough to make a big difference overall. Indeed, the Durham volume seems particularly low given that they have 14,000 students. But there is a simple enough explanation for it. They are on break now and don’t resume till after Martin Luther King Day. So most of the students aren’t on campus (and even if they were they very well might be registered in their home district rather than in Durham). Duh. (Dartmouth does have a Winter term so many of the students are there, but they have many out of state students and the total student population is much smaller than at UNH.)

If you compare the Entrance Poll from Iowa results to the Exit Poll from New Hampshire results and focus on the gender items that are on the top, you’ll notice first that the split overall between male and female is near the same across the states, that the Obama numbers are also remarkably the same across the states, that Clinton’s increased success in New Hampshire came somewhat at the consequence of Edwards and Richardson, but she also seems to have picked up the votes that had gone to Biden or Dodd in Iowa.

There is no compelling reason to assume that if Biden and Dodd hadn’t dropped out and if the New Hamshire primary had been on the same date as the Iowa Caucuses that the splits in the results would have been the same across the two states – the results on the Republican side indicate that clearly – but for the sake of argument, let’s assume it as I implicitly did in the previous paragraph. Then, in accounting for the fact the Biden and Dodd did drop out of the race, a pollster to get accuracy in New Hampshire would have to over sample former Biden and Dodd supporters to find out who they’d turn to next. (It makes sense to me that these people were undecided till the last minute.) Similarly they should over sample both Kucinich and Richardson supporters, perhaps not to the same degree but to some extent, because their supporters might come to realize that a vote for their previously preferred candidate would have little to no effect on the ultimate outcome. (That Biden and Dodd did gather some votes in New Hampshire after they had dropped out of the race indicates that the trickle of supporters they kept were responding to some other prerogative than to affect selection of the Democratic candidate.) One can explain almost half of those who reported they were undecided day of based on this sort of candidate switching. This idea got essentially no coverage in any of the news and analysis.

The other type of switching, this was heavily reported, occurred among independents who opted to change the party they’d vote for. I haven’t seen the breakdown of these folks ex post. The wisdom before the fact was that if these folks opted to vote Democratic that favored Obama. And on this point I can’t help but feel that the poll suggesting Obama had a 9% lead in New Hampshire might have driven some of these people to vote Republican – their vote would matter more there and they didn’t want to see Huckabee or Romney win. There is not much to be done about prior polling affecting actual voting behavior. It’s going to happen. But if the prior polling is itself inaccurate for the reasons I’ve suggested above, that is disturbing --- it implies the reporting of the polls creates a regression toward the mean consequence in and of itself.

Now let me turn to the idea that Obama and Edwards supporters – mostly women – switched to Clinton over the last couple of days. This is the “comeback” that is being reported today. The comeback is attributed primarily to how Clinton performed in the debate Saturday night, particularly at the point where her “likeability” was brought up and in her getting teary eyed on Monday. (If we actually knew the second place choices of the Biden et. al. supporters, that those changed to Clinton yesterday would also be part of the comeback.) Another possible source of the comeback was an op-ed piece by Gloria Steinem that makes for a compelling read (whether you agree with her or not) on why it’s harder to be a woman presidential candidate than to be a black presidential candidate and that difference plays out the most in how emotional/controlled the candidate appears – our gender biases run that deep. We voters, who want to have it both ways and seek out the best attributes each gender has to offer in one package, might then change our view of the candidates when one aspect of the persona, previously veiled, becomes more overt. This is the source of the comeback, or so it seems. It’s what I termed the yoyo, viewed from the voters’ perspective rather than the candidate’s.

This morning, Maureen Dowd took on Hillary, on the grounds that both the Saturday night and Monday expressions were completely controlled and planned beforehand in the same manner that Bill used to bite his lip while delivering an address, and Gloria, on the grounds that can’t we just evaluate the candidates on where they stand on the issues, but in the process gave tacit endorsement to the comeback hypothesis. Dowd is a good writer and it’s interesting to see her indictments of Democrats rather than Republicans. But she’s no statistician. (I’m not either, but I’m probably a better social scientist than she is.)

And there is a problem with the comeback hypothesis. It simply might not be true. It might simply be that Clinton actually had more support in New Hampshire ahead of time than she had in Iowa and vice versa for Edwards. That and the rest of what I’ve got above could explain the entire outcome, in which case the comeback is a post hoc rationalization only. (If you haven’t read Stephen Jay Gould on the Streak of Streaks, you should. We humans are into post hoc rationalizations.)

What’s the problem maintaining the comeback hypothesis if it is not true? First, the polling is likely to continue to be misleading, at least until it settles down to a two candidate race. Second, the candidates themselves won’t understand how to wage their campaigns most effectively. My prediction (economists are notorious for believing that future economic phenomena can’t be predicted beyond the trends that are already known at present) is that the Richardson support will ebb and then the Edwards support will likewise wane. To win Clinton and Obama will need to woo the supporters of these other candidates without alienating their current base. How should that be done? Third, it’s also a disservice to the voters. There is no question now that the vast majority of the electorate is dissatisfied with the Bush administration. Voters are entitled to make their judgments on the candidates based on a reasonable guess as to how they’d actually serve if they were elected president. For the election in 2000, one can make a good case that the campaign did not help voters understand much if at all what the Bush presidency would be like. An embrace of the comeback hypothesis, if it is wrong, encourages voters to repeat the same mistake for the election in 2008.

Paul Krugman, whom I’ve taken on from time to time, encourages voters to look at politicians on the issues and to ignore the rest. On this I have to agree with him. It’s the best way to judge. There may not be a huge difference between the leading Democratic candidates when considered this way and this way only. So it might be hard to select the most preferred. And procrastination is another thing that we humans are into. But once a choice has been made in our minds, I for one don’t believe we continue to bounce around among the alternatives. And I certainly don’t want to give the candidates more fodder to take W.C. Fields as their role model. We’ve been down that road too often.

No comments: