Anyone knows a reference for this bimodality Of 1-5 Ratings? This is not what I see in amazon. Votes are not normally distributed, sure, but they tend to cluster on the upper end of the scale. The mean for amazon votes (note that since the distribution is not normal, the mean is meaningless) is something like 3.4.

I can't find solid analysis of Amazon's rating system, but neither YouTube nor Netflix are bimodal:

* YouTube is all 5 stars (http://youtube-global.blogspot.com/2009/09/five-stars-domina...)

* Netflix: 3, 4, and 5 dominate (http://www.timelydevelopment.com/demos/NetflixPrize.aspx)

I think a closed forum with users who have established definitions for the difference between what each number of the scale implies would have a bimodal distribution. Sites like youtube ( http://youtube-global.blogspot.com/2009/09/five-stars-domina... ) and amazon ( http://portal.acm.org/citation.cfm?id=1562764.1562800&co... ) have more of a J-curve

Quite possibly a terrible example (+ hasty generalization, etc.), but here goes:

http://www.amazon.com/Structure-Interpretation-Computer-Prog...

I liked the way he applies moderately useful statistics to a real world application, mostly just as an example of how we all could be using the standard deviation along with the mean.