Earlier this week, the Washington Post’s Chris Cillizza tried to reconcile his own sense of Senate Minority Leader Mitch McConnell’s vulnerability, based on reporting, with the Post’s election prediction model, which gives McConnell a 97 percent almost-certain lock. It’s not just the models from (my friends) John Sides, Ben Highton, and Eric McGhee that give Democratic challenger Alison Lundergan Grimes almost no chance of winning; the Upshot's model gives McConnell an 88 percent chance.
Cillizza does an excellent job of explaining what’s going on, and why the models see less suspense than one might expect.
However, I want to push a bit on the percentages. As discussed before, the way to understand these models isn’t “if the election were held today,” nor is it quite “this is what will happen in November.” Instead, the projections show what would happen in November if everything remained as expected and if the models held. That is, if the president’s approval rating or other similar measures remain where they are; if the economy remains how it’s been (or how the models expect it to be) and if there’s nothing particular about these candidates or their campaigns that a model can’t account for.
I strongly suspect that for contests such as the Senate race in Kentucky, or perhaps Oregon for the Democrats, this means there’s more uncertainty than those imposing percentages would imply. That’s not a flaw of the models; it just reflects the limits of predictability. Not to be crass, but even just looking at the extremes, we’ve had at least a couple of candidates die at the last minute, and at least one dropped out after the nomination because of a scandal.
The closer we get to the election, the more that sort of uncertainty drops away. Of course, the closer we get to the election, the more we can simply look at a good polling average to figure out who is going to win.
As I’ve said, it should be possible for good local political knowledge to beat the models from this far out. The way to do that is to begin with what the models are telling us about the fundamentals, and then figure out how (and if) the specific circumstances of a particular election contest might make those assumptions iffy. Is there a challenger who doesn’t have top-rate credentials, but has a good chance of running a terrific campaign? Does one of the candidates have a reputations for ideological extremism (that the models wouldn’t pick up)?
Ignore what the models tell us, and what political scientists in general know about elections, and it’s easy to get carried away by a good ad, an ugly gaffe, or even a particularly enthusiastic campaign event. But no one believes that all Senate elections are perfectly predictable six months before the ballots are cast, and it’s important not to let specific numbers bully us into believing otherwise.
To contact the writer of this article: Jonathan Bernstein at email@example.com
To contact the editor responsible for the article: Max Berley at firstname.lastname@example.org