More on Bush and Public Opinion

Actually, it sounds as if seven out of seven predict that he’ll win the popular vote– if you get 49.9 percent of the popular vote, the other guy is not likely to get more than that, since surely more than 0.2 percent of the population will vote for third-party and independent candidates.

However, notice the big, big qualifier in the article: these models aren’t really trying to predict who will win! They’re trying to predict how much of the popular vote Bush will get, and they all say he’ll get within a few points of 50 percent. I could have told you that. The whole important difference is inside that margin of error. Kerry could win by a considerable electoral margin without perturbing these guys’ estimations of their models.

If you follow the links in the Chronicle article to the U Penn page on forecasting [http://morris.wharton.upenn.edu/forecast/Political/], some interesting comparisons can be made. The forecasters differ considerably from virtually all of the polling, which, almost without exception, continue to see a nip and tuck, too close to call election. Perhaps the Achilles heel of forecasting is precisely that it bases itself on past elections and their patterns, while this election seems to be, at a minimum, anomalous, and even possibly the harbinger of a new pattern with a highly polarized electorate. It is significant that the Democrats received no bounce from a near perfect convention, and that, I suspect, the Republicans will find themselves in a similar situation. That is unprecedented. And the entire election may be as well.

One of the odder journals of recent memory was an issue of PS from 2002 in which the forecasters tried to explain what happened in the 2000 election. The articles cried mea culpa, but tried to tweak the models to achieve post hoc accuracy.

I don’t claim expertise in this area, and I think there’s a lot to be gained from forecasting research. People are too snarky about this kind of work (as they are about political science in general). But my sense is that forecasters have always been too quick to try to correlate quantifiable indicators with projected vote totals. The numbers are probably not too reliable, because indicators have not always been measured comprehensively or well. In addition, such studies miss the election year narratives that seem to affect voter perceptions. Forecasts have trouble dealing with things like the excessive media scrutiny around the question of whether Al Gore put on makeup during the second debate. This was watercooler material, not fluctuations in the payroll survey.

But as I say, I’m not a specialist here. I’d like to know whether forecasters have included methods of factoring in voter perceptions. Given the disconnect between the APSA numbers and polling data, my guess is that they haven’t.