Since late January, a diarist over at DailyKos has been getting more and more attention. He uses the handle “Poblano”, and he has been predicting primary election outcomes a few days before the actual voting with astonishing detail.
If he was just a dude getting lucky with predictions, that’d be one thing. “Poblano” actually showed his math. He’d take voter demographics from prior elections, take crosstabs from polling outfits and weight them based on performance during the primary season, and pour them into an arcane “statistical regression analysis model” and have it spit out his results.
He got mentioned in National Journal and has been a frequent contributor to Al Giordano’s The Field blog. Poblano even started his own blog at http://www.fivethirtyeight.com.
Today “Poblano” pulled back the curtain. Turns out he’s Nate Silver, perhaps the most respected baseball statistical analyst working that field. He’s managing editor of Baseball Prospectus, and the developer of a performance prediction tool called PECOTA that every MLB team uses. Basically he took the same statistical predictive methods he’s been using for baseball for five years and put them on the election.
Poblano was pretty much the only guy out there saying on May 4th that Indiana was going to be close–he had Obama losing by something like 1.74 percentage points. He cited a “15 most similar districts” model that he was using to predict the suburbs around Indiana, and at the time I thought “Christ, that sounds like VORP or PECOTA.” Should’ve known, huh?
If Nate isn’t already rich enough off his baseball gig, I have a feeling that a couple of major news outlets will try to enlist his services…and perhaps even a campaign.
He has a sequence of charts in the upper right hand corner detailing the current data from his model…
which has it pretty fucking close. What’s the margin of error on this? His Obama v McCain projection has McCain up by 3 delegates.
He has one showing it very close, but throughout his blog (kind of buried, to be honest) he’ll point out that his predictives require a lot of polling data, and polling data is remarkably fickle. His current charts say that if it was Obama/McCain on Tuesday, it’d be incredibly close. They don’t claim to represent what that same matchup will look like after the conventions and debates and September/October campaigning.
If the actual numbers are even for Obama vs McCain, as his site shows now, then we might as well just assume McSame will win, as it’s guaranteed there’ll be enough electoral fraud to push it over the edge.
His predictions are based on past results, so that’s already factored in. If it comes to that, we’ll just have to look to the Guardian Council^W^WSupreme Court to appoint a President.
