John / Mark,
Yeah, that was an exceptionally weak piece of work.
As you have both said, the analysis was at the level of states rather than individuals. According to the paper, the reason is that the researchers hand anonymous data that was basically the number of subscribers to a specific network of online pornography sites by Zip Code. What’s great about such a dataset, rather than poll responses, is that this provides a measurement of actual behavior, rather than stated behavior, and people tend to lie about sex — a lot.
What was so lame about the analysis (at least upon a first quick read) is that the researchers did analysis at the state level when they had the hard-to-get porn usage data down at Zip Codes. Voting data is easy to get at the county level, as is good broadband penetration data (the FCC provides this data at the Zip Code level, but there has been a long-running debate about accuracy at that level of granularity). You can also get all of the key demographic variables used in the analysis at the county level. There is an argument not to do the analysis by Zip Code because of debatable broadband penetration rates, but why didn’t they at least do this at the county level? It would not have eliminated the “not at the individual level” problem, but would be a lot better than state-level analysis. I did this kind of work just to get a blog post right; I don’t get why they weren’t willing to do a little more work for a published research paper.