With a couple of days to go, it seems that Obama will cling on to the presidency after a close shave against the 47% Flip-flop. Although the national vote is closer than close, the granularity of the electoral college is in Obama’s favour, and Romney just isn’t popular enough where it matters. Nate Silver has an elaborate blog on the NYT outlining the numbers that lead most neutrals to come to the same conclusion.
The question some people are asking is: “can Nate Silver predict the future?”
Well, if you’re going to predict something, let’s begin by saying the best thing to predict is the winner of a two horse race. Even an octopus can do that quite well. I’m not being flippant – the chance that you are right (50%) is the highest of any possible random prediction. It’s the closest you can get to knowing the future without loading the dice.
Nate Silver isn’t predicting the toss of a die. He’s got polls and history to weigh. He uses this information to generate a model of what is likely to happen. Then he can compute a number.
Several journalists, of national renown and otherwise, have dismissed producing a number as nonsense, and are prepared to put rather paltry amounts of money on it. Classic example of being right, but not for the right reasons. Not because you understand the problem.
Mark Coddington (amongst others) has come down squarely on Nate Silver’s side, talking up the rationality of what he does. But there is some clear blue water between, say, electrostatics and psephology. I’m very firmly against any kind of relativism in science. The force between two electrons separated by a metre of free space isn’t something that will depend on the granularity of the electoral college. There is only one answer.
Silver himself has given the following example: the Giants are up 24-21 in the fourth quarter – it’s not a toss up, they are favoured to win. This example is very simplistic, because it is only true if you consider a flat model for scoring points in the fourth quarter. You might choose to evaluate it differently- perhaps the Giants are almost always outscored by their opponents in the fourth quarter or so on. Then you might just as well expect them to lose, if you heard they aren’t too far ahead.
Simply put, what you predict depends on your model.
David Brooks has ill-advisedly used the word “magic” to describe how he understands this subtlety. He also more wisely decried fancy models as ways to predict complex behaviour, such as the stock market. I’ve listened a lot to David Brooks down the years (although I prefer EJ), and read many of his columns. He’s a reasonable guy, but he has an extraordinarily high opinion of his own intuition. This “political radar” is disparaged by Mark Coddington, and of course it’s all a bunch of post hoc ergo propter hoc rationalisation, and appeal to authority.
The laughable central plank of reasoning here seems to be that, because “Anyone can predict the outcome of the election – it doesn’t mean they know what they are talking about” (true, too true) therefore “Anyone that predicts the outcome of the election can’t possibly know what they are talking about” (not true at all). Brooks et al know who they want to win the election (well, Brooks would like the Republicans to win but perhaps Romney and Ryan to be raptured on the spot by a forgiving God). Sadly, because David Brooks and the rest of us are dazzled by a million different polls, it’s hard to stay sure of the outcome. Nate Silver uses the large number of polls to his advantage – they let him take advantage of regression to the mean.
But in a way, Brooks and the rest of the maths haters are completely right. You can’t be sure that your model is correct. Nate Silver knows his modelling is arguable, but with good reason, he thinks it quite bulletproof. It’s robust to wide, yet plausible, variations. Another way to say this, is that even if you change the model a lot, Barack still lives at the bottom of 16th St NW for another 4 years. I tend to agree with Nate Silver. I like his style too.
It’s nice to be able to say that everyone is right. You should be skeptical about models. But if the models are robust, they’ll give useful predictions. But the model is just that. It’s not the election itself. That result will have to wait for a few more days.