Laurence Simon, in the midst of a rant about a Houston affiliate’s attempts to pump up its ratings, exposes the bad statistics peddled by AC Nielsen:

The samples are ridulously small and unstable, absolutely abysmal when it comes to tracking democraphics other that Yuppie Whitey because of tweeser-sized samples, the numbers are cooked twice, and then the exceptions and loopholes in reporting the results are downright shameful.

The Nielsen sample might be reliable as a national sample, but it’s absolutely hideous once you try to subdivide it for over 300 local markets. Even as a national sample, they should be intellectually honest enough to attach a giant confidence interval (which, based on reports of their sample size, is probably on the order of several ratings points at 95% confidence) to their estimates. It’s a miracle their numbers correspond at all with reality.