Skip to content

Nate Silver is no Carnac (but I read him anyway)

October 14, 2012 pm31 10:19 pm

Nate Silver predicts elections. Or claims to. Sort of.

He burst onto the scene in 2008, at He was a sports stats guy, a sabrematrician, and he thought he could use his skills to analyze poll results. He was good enough that he attracted some attention. A lot of attention. And on election day he had 49 states called correctly. Oh.

He commentated. The New York Times snatched him up. And now “Election Forecasts – Five Thirty Eight – Nate Silver’s Political Calulus” is updated on the Times website daily.

But he’s not very good at predicting. He is good at reading current polls – but not to tell us about the future. He excels in explaining what just happened.

Is that blip statistical noise? Trust Nate. How big was Romney’s debate bounce? Ask Nate. But will Romney do well at the next debate? Who knows?

The mechanics behind his work are simple. He thinks that pollsters are consistent, and that results are fairly consistent. So he tests the same poll with four pollsters, and his result tells him not about the state being polled, but about each of the pollster’s “lean.” He checks state polls against national polls, and sees how tightly they move together… or not. So the next time that state is polled, it tells Nate something about where the national race is, as well. And vice versa. Without a new poll in South Pennsyltucky, Nate intuits which way it has moved from the national polls, and the record of correlation between that state and the country. He easily doubles or triples the amount of information sucked out of each poll, and creates a much better-filled in picture of the state of the race, today, both nationally and in each state.

Which all means that Nate has a great handle on what would happen if the election were held today. But not so good for three weeks from now.

He also tries his hand at analysis of state politics. Not as impressive. He just doesn’t have the depth of understanding of history. Republican in 1940 does not mean the same thing as Republican today. Nate misses this. Gun ownership means different things in different parts of the country. In some states, the whiter places are more conservative, but not in all states. And bellweather counties don’t make sense in states where the electorate is changing. All salient details that escape Nate.

But I still read. His “today” snapshot is wonderful. And there’s no one I would go to first to find out what happened in the polls three days ago. Seriously.

3 Comments leave one →
  1. October 15, 2012 pm31 11:19 pm 11:19 pm

    My understanding is that Silver’s prediction is different than just a poll aggregator in at least two ways: 1) it incorporates economic data; 2) it takes into account predictable bounces (conventions). At least for the forecast. The now-cast is just an aggregator, I think.

    But no, he can’t predict the future. ;-)

    • October 16, 2012 pm31 11:04 pm 11:04 pm

      His economic data is pedestrian. Nothing special. And he does prefactor conventions.

      But what is really special is that he studies relationships between polls, so that he can take a Quinnipiac poll of New Hampshire, and use that data to shape the current state of the national race, and the shape of the race in at least several other states. Gallup moves up 2 points nationally… does that have impact in Pennsylvania? How much? I get how he’s set up the web of correlations… but I doubt I could execute it.

  2. October 28, 2012 pm31 10:05 pm 10:05 pm

    You might also check out Sam Wang:

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: