cjsoto on 17 July 2007
I don't take issue with the would-be debunkers' study design, but I DO disagree with their interpretation of the results. If 15.8% of the variability in game sales can be explained by variability in game ratings, that means the correlation between sales and ratings is r = .397. A correlation of this size lies somewhere between a medium-sized effect (one noticeable by a careful observer, about r = .30) and a large effect (one obvious to an observer, about r = .50; Cohen, 1988, 1992; http://en.wikipedia.org/wiki/Effect_size).
The bottom line: 15.8% might sound like a small amount of explained variability, but most statisticians would consider this evidence of a moderate-to-strong association between game ratings and game sales!
