Wednesday, August 13, 2008

My only political comment and it's not political at all

Flat out, politics will not be a topic of discussion on this blog, but there is one blog that I want to mention for having an unusually rational look at the daily polling results and predicting the Electoral College Results that decide the election, not nationwide polling.

The site is FiveThirtyEight.com. It can be a little confusing at first to look at, but here is the idea in a nutshell. Various groups publish polling data for each state, resulting in an average and a standard deviation, from which the distribution can be worked out. Using these input distributions, 10,000 Monte Carlo simulations are run for all 50 states with the output being another distribution. As of today, 8/13/08, Obama wins 65.4% of the simulations with the most likely outcome being 299.2 to 238.8, and the most likely outcome of the popular vote being 49.7% to 47.7%. The results change daily (or less frequently if there are no new polls for a given day). Watching the site over the last few weeks, the numbers reported above haven't been changing much: Obama is getting about 300 Electoral votes, winning about 60 - 65% of the simulations and getting about a couple of percentage points more than McCain in the popular vote. There are far more details to get into (older polls are weighted for less), but this is it in a nutshell.

While Monte Carlo simulations are just a tool that can be misused just like any other tool (scientific or otherwise - ever seen a bolt on a bike rounded off by a kid using pliers?), this is a far more rational approach that just looking at the "average" reported by a poll. However, this is the first national election for which the technique has been used, so the connection to "reality" has not be firmly established. I am certainly looking forward to seeing how well this model does in the actual election. And I have to expect that the "reality" will be used to refine the model for future elections.

The operator of the site is Nate Silver, a baseball statistician of renown. He is openly an Obama supporter, but believes that this effort is party-neutral.

No comments: