Nate Silver of 538 Signs With Penguin In Two Book Deal

observer.com

16 points·brm··4 comments

Hopefully he will reveal more technical details about his statistical methods.

When you're making predictions based off complex models like he is, transparency gives you more credibility. There's actually a pretty good description here: http://www.fivethirtyeight.com/2008/03/frequently-asked-ques...

Go through that page with a basic statistical package, plus maybe Wikipedia for reference and you can probably see most of what he did.

Well, I was hoping for more details than a brief answer to a FAQ.

See this earlier post:

http://news.ycombinator.com/item?id=356461

His baseball projections are based on a proprietary algorithm (PECOTA).

From wikipedia http://en.wikipedia.org/wiki/PECOTA :

detailed formulas are proprietary and have not been shared with the baseball research community.

Well, he lists all the variables in the regression he uses, tells you where he gets the polls, how he weights them, how he adjusts for timing, how he incorporates the regression results, etc. The FAQ is anything but brief -- there are LOTS of little links where he specifies all the details.

I agree it's not super well-documented where you can just plug-in numbers and go, but it seems like he provides the details that someone else would need to do something similar.

It depends on your goal. If you want to exactly replicate his work, you'd need access to the exact same data, and maybe he hasn't specified exactly what he uses to train the regression model. But if you just want to do something similar, the ingredients are all right there.

Also, re: the "proprietary" formula, it means that you don't know the specific parameters his model uses. But if you just want to learn how to do something similar (which is what it seems like is your goal), the information is all there. The methods aren't proprietary at all.

Reply on news.ycombinator.com