Lots of CSCC people have been reading and talking about Nate Silver's election polling website FiveThirtyEight.com. If you haven't seen it, it's worth a look. (538 is the total number of electoral votes.)
Nate Silver first got major attention on the night of the Indiana and North Carolina primaries when his projections managed to capture the movement toward Obama that all the other polling had missed. In recent weeks, Nate Silver has had a lot of exposure in the Main Stream Media, and he is, as he puts it, "enjoying his fifteen minutes." Silver is careful to note that he's an Obama supporter, but he also has a stake in making unbiased prediction, unlike many of the polls that he interprets based on their track records.
Subscribe to:
Post Comments (Atom)
Posts
3 comments:
Let me add a bit to Loren's post. There are a few polling compilation sites, most notably Pollster.com and Electoral-vote.com, in addition to Five Thirty-Eight. What Nate at 538 has over the others is a more sophisticated projection model. As one example: he can update probabilities of winning particular states based on changes in national polling results (or in polling other states). He has demographic data linking the states, so a shift towards Obama in a poll in Tennessee would result in an increase in Obama's projected chance of winning Kentucky as well, but won't affect, say, Oregon so much.
This is important, because state polling is sporadic and a given state hasn't necessarily been polled recently enough to reflect real trends.
By having these kinds of features, he was ahead of the other sites in showing the dynamics of the race: first the "Celebrity" ad was eroding slightly Obama's lead, then the conventions and Palin pick give McCain a big jump, and now we're back to a huge Obama lead. The other poll compilation sites were lagging a week or so behind Nate's in picking up the effects.
I visit 538 once a day to get his daily update in the projections, to get what I think is the most accurate current snapshot of the race possible.
I can't speak to his technique, but he also seems to be using multiple simulations. I find the output much easier to grasp as it tells you how often various outcomes might occur. Doe she discuss what are the inputs and relationships in his model?
Jordi: Yeah, check out the "FAQ" (there's a button in the upper menu bar). He gives a pretty lengthy discussion of the variables that go into the regression and how he infers correlations among states based on demographics and other things....
Post a Comment