http://www.stat.columbia.edu/~cook/movabletype/archives/2010/01/what_can_search.html?utm_source=feedburner&utm_medium=feed&utm_campaign=Feed%3A+StatisticalModelingCausalInferenceAndSocialScience+%28Statistical+Modeling%2C+Causal+Inference%2C+and+Social+Science%29&utm_content=Google+Reader

Not entirely dissimilar to using P2P search or availability data to predict sales.

http://arstechnica.com/media/news/2010/01/bittorrent-census-about-99-of-files-copyright-infringing.ars
The full study description:
http://www.freedom-to-tinker.com/blog/felten/census-files-available-bittorrent

"this section [movies/shows] was heavily biased towards recent movies, instead of being spread out evenly over a number of years."

The Ars conclusion, that "the more DRM there is on the legit versions of the content, the more popular it is on P2P" is confounded by the overwhelming greater desirability of content produced by the major labels, of course, but it does mean that small artists have little to lose from going DRM-free, or even freely-downloadable if what they want is traffic driven to their website.

Obvious limitations:
1. Small sample size
2. Not weighted by volume shared
3. Not an unbiased sample of the universe of pirated content

I wish the major health surveys would release small-area data so that we could do things like this. County is about the smallest we can get on a national scale, and only with mortality data.

In other news, just started clinics. Have no time for anything, but it's a lot of fun.