Friday, October 19, 2012

Nate Silver's problem

One of my friends likes Nate Silver's polling work since he analyzes polling like a baseball statistician.
Time for another sports analogy.
Silver continues to give high odds of President Obama winning the election - despite Gallup's recent polling and the trending of several states to Romney.
Part of the reason - he had Obama's chances so high before the first debate.
If you're trying to be statistical in your analysis, you can't have big drops in your numbers. It must be slow and gradual.
Take the national college football polls. Each year, the traditional powers and last year's winners top the preseason polls. If you aren't expected to win, it takes a long time to climb up the chart.
An unexpected unbeaten team finds it hard to pass the teams ahead of them, especially those with one loss. The pollsters like those team, and find it hard to drop them too much when they lose.
For a few years, I voted in the West Virginia prep basketball polls. I'd check the teams other voters liked, and try to keep an open mind about where to place teams. But as the season went on, it's hard to drop your top teams too far.
Silver had Obama's chances too high in September.
He treated Romney like he didn't have a chance. An underdog so far down, he's never make it to the top spot.
Romney's piling up some good numbers the last two weeks, but the inertia of past predictions make it hard for Silver to give him more than a 30 percent chance right now.
Silver's done well living off his 2008 poll analysis. He can write reasons why Gallup and Rasmussen are wrong to give Romney an edge, but a Romney win stomps all over his reputation.
Let the teams fight it out on the field. Romney's done pretty well the last two weeks, and I look forward to another strong performance Monday.

No comments: