I like the OkTrends blog; the analysis is more casual than scientific, of course, but the content is stuff that I was really interested to learn.

Also, as a more general point, I think that non-scientific but enormous back-of-the envelope calculations (like OkTrends) are a good supplement to tiny but scientifically designed studies. If the big&sloppy approach and the little&precise approach give completely different results, that's cause for further investigation. Pre-Internet, we didn't have the opportunity to do that much big&sloppy statistics, except very crude things like the census; but today, I think it might be useful to double-check the results of small experimental studies (e.g. in medicine) with self-reported stuff from an enormous population sample.