Tuesday, May 20, 2008

My take on Poblano

Recently there has been some interesting coverage from Mark Blumenthal and criticism from Dick Bennett of ARG about a blogger named Poblano, who posts at the excellent fivethirtyeight.com.

Here's the crux of the story: Poblano has a non polling statistical model he uses to forecast what's going to happen in primaries, and it's been remarkably accurate in predicting the margins by which Clinton or Obama have won recent primaries. He also projects turnout, where his predictions have not been as accurate. But no one else is putting out accurate data based turnout estimates that I've seen so I don't see that as a particular strike against him.

Poblano's accuracy seems to be making some pollsters insecure, but I don't see it that way. When polling companies release numbers on the horse race in various states, there's not much point to it other than to call the election right. If Poblano's models continue to be more accurate than a lot of the polling out there, then good for him! It's us pollsters' problem if someone else spending a heck of a lot less money finds a better way to do it than us.

Check out Poblano's projection for Oregon. I hope he's wrong since we have Obama winning the state by a lot more :) Nonetheless, he's doing very interesting and quality work and I encourage anyone who wants to read a good site to know what's going on in the Presidential race to check out fivethirtyeight.com.


Anonymous said...

Thanks for the link. I checked it out.

Sorry, but this is not quality modeling. These models are not able to accurately reproduce any of the primary votes.

Your polling is much more interesting. Thanks for your good work.

Anonymous said...

Exactly. There are no hold out samples.

Look at: http://wmbriggs.com/blog/2008/02/25/example-of-how-easy-it-is-to-mislead-yourself-stepwise-regression/

Anonymous said...

Poblano crashed and burned in Kentucky.

Keep polling, please!

Web Statistics