In the absence of evidence, it's far safer to assume voters aren't rating consistently with one another. That's what ordinal data implies. It also doesn't suffer from this problem of perverse incentives by over/undervoting to get extra influence.
Under a mean system, when I can see the rating is currently a 4 and I think an app deserves to be a 3, I can strategically vote a 1 to get three times the voting power as if I voted honestly. Under a median system this doesn't work - my incentive is always honest. Some other nice properties of using the median as a primary sort: 1) If a majority of voters give an app a particular rating, the computed rating will always be within 0.5 points of that. 2) If a majority of voters give an app at least a particular rating, the computed rating will be at least 0.5 less than that. Neither of these are true in the current system - it's possible for a majority of voters to give an app 4 or higher and that app to be rated 3 (or worse!) You mentioned that the ratings currently look about right, which is justifiably a good reason to be conservative here. I think the median actually is the conservative choice - it's the correct measure when you're looking for wisdom of crowd phenomenon, it's the correct winner under most real-life voting systems (http://en.wikipedia.org/wiki/Median_voter_theorem), and it also happens to be about the same (in most cases) as the current system. I'll parse the data we have a bit and make a few examples of what ratings currently are, what they would be, and under what circumstances they might change. -- You received this bug notification because you are a member of Ubuntu Bugs, which is subscribed to Ubuntu. https://bugs.launchpad.net/bugs/894468 Title: Statistics algorithm for sorting ratings looks fishy To manage notifications about this bug go to: https://bugs.launchpad.net/ubuntu/+source/software-center/+bug/894468/+subscriptions -- ubuntu-bugs mailing list [email protected] https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
