1) The histograms (of rating distributions) that I’ve seen don’t support your hypothesis that many users rate items 3 stars. Why would apathetic/callow users bother rating at all?

2) A sophisticated system can normalize the ratings data by, eg, discounting ratings from apathetic users (who tend to rate everything the same way), or even disregarding 3-star ratings altogether.

Someone mentioned that most reviews at Amazon are positive. I think it makes sense: who would bother reviewing a bluegrass album unless they’re a bluegrass fan who owns this album? In which case, they probably like it. They may imagine that non-bluegrass fans will see their rating and be convinced to try it; therefore, it’s a review with an agenda.

If I want to discover new music, I find it more useful to find something I know that I like and look at the “people who bought X also bought Y” information. It’s also cool to look at the “favorite albums” lists of people who share my opinion of this particular album.

Read the article

Chris

Perhaps one could look at it in that if you are getting the majority of maybe’s for responses, what is truly not working was the question rather than the rating system. “Do you like broccoli steamed until just short of crisp?” is a much more complete question, and much more likely to garner you a yes or no response. Tailor your question with an eye to how much information you want in response, as it were.

I suspect many maybe’s could be “It depends” in disguise, since even a 5-star rating system only allows for degrees of yes or no, and many times the answer varies with the situation. I don’t like broccoli at all if you boil it the way my mother does, but it’s my favorite thing in a stir-fry. I liked the functionality of the feature as described, but as it didn’t actually work in my browser, how am I supposed to answer yes or no? Do I answer no and have you thing I didn’t care for the feature itself, or do I answer yes and have you totally miss the fact it didn’t work?