I'm watching this Nate Silver versus Strategic Vision thing with some interest. While I've been critical of Nate before, his quantitative goodness relative to broader media crappiness means I'd be quite happy to see him rise in stature. Helping to bring down a fraudulent pollster is the kind of thing that could be a real feather in his cap, so I'm hoping he's right.
On the actual issue at stake, I'd like to see more simulations that are comparable to Strategic Vision's polls. Nate has graphs up comparing the semi-randomness of the final digits in 2008 general election polling with the nonrandomness of the final digits in Strategic Vision's polling on everything. Those are useful, but something more than a pairwise comparison would probably help him make his case. It's hard to get an intuitive sense of how big an outlier Strategic Vision's results are in terms of the Benford's Law kinds of stuff if we only have one thing to compare them to.
Post a Comment