There's a great deal of evidence suggesting that the polls commissioned from Research 2000 by Daily Kos were either inappropriately cleaned up, or outright fabricated. To summarize crudely, there doesn't seem to be any random statistical noise in the data; there are no outliers, and there's an extremely odd tendency for the male and female crosstabs to both end in either an odd, or an even number; since these numbers vary pretty randomly, there's no reason that they should consistently be both odd or both even.
To his eternal credit, that evidence is being published by Kos himself. I wish more commentators were this forthright when they've been taken in by bad data.
I also wish he hadn't been taken in by bad data, including this infamous poll which proves that Republicans are, like, the awfulest, stupidest people in the entire known universe. Given the patterns that the investigators who published on Kos have found, it seems like that poll is probably bunk--concocted or massaged to fit the thesis of the person who commissioned it. (Just to be clear, I don't think Kos knew that this was happening. A customer who is told what he wants to hear is a customer who is happy to buy more polls.)
However, the fact that R2K has apparently been discredited will not prevent this poll from popping up, over and over, in the work of pundits and bloggers. The speed and openness with which Kos has exposed this problem is admirable. But we need to be even faster at sniffing out bad apples before they pollute the data barrel.
We want to hear what you think about this article. Submit a letter to the editor or write to firstname.lastname@example.org.