Why Nate Silver’s Gambling Streak Makes Me Trust Him More

America is filled with people who think its okay to lie, bullshit, or otherwise misrepresent the truth in order to advance the electoral prospects of a politician or the cause of a governing coalition. Let’s call them shills. Other people aren’t necessarily aware that they’re misrepresenting the truth, but their work is so shaped by what would advance the causes of a candidate or governing coalition that it’s often indistinguishable from the shills. We’ll call them hacks. In a better world, journalists would be sworn enemies of shills and hacks, and the best are. Unfortunately, the press, especially the political press, has more than its share of shills and hacks.

I’m not sure if you’ve been following the Nate Silver hullaballoo this past week or so, but the short version is this: he has used statistics to show that Obama had (at the time of the linked article) a 75% chance of winning, based on a model that weights different polls according to past accuracy, instead of using the very blunt national polls. Advanced stuff. But because those on the right don’t like the answer, they are crying, “foul!”

A) this isn’t surprising, given how the GOP tends to treat science the way a baby treats a diaper. But B) this is actually an example of how science works. (A and B being somewhat related)

Taken by itself, a 75% chance of winning isn’t testable for a single-outcome event like an election. You can win even if your probability is low — unlikely events happen all the time. (and 1 in 4, or the current 1 in 6, is not all that unlikely. Clearly, you can roll a 6 on a standard die, even though there’s only a 1 in 6 chance of doing so) In the long run we see truly unlikely events because there are a lot of events. But that’s what you need in science in order to test your model — lots of events, so that you may gather statistics. This is why it’s important that Silver not only uses his model(s) to predict the result of the presidential election. There are more results

In 2008, he correctly predicted the outcome of the presidential election in 49 of 50 states, and accurately identified the winner in all 35 Senate races.

So there’s more to it than simply picking an overall winner. When your model accurately predicts outcomes, you have confidence that the model is a good one. Just like in science. We aren’t satisfied if all we can do explain past events; it’s the predictive power, and the possibility that the model can be falsified, that is one of the things that sets science apart from any wannabes.