Topic: [Feature] Sorting by the Lower bound of Wilson score confidence interval for a Bernoulli parameter

Posted under Site Bug Reports & Feature Requests

Requested feature overview description.
http://www.evanmiller.org/how-not-to-sort-by-average-rating.html

It would be nice if we could sort by this algorithm, which balances the number of votes and ratio of votes to sort very effectively.

Why would it be useful?
Sorting by "upvotes minus downvotes" has issues, as a post with 1000 upvotes and 900 downvotes is placed above a post with 100 upvotes and 10 downvotes. A post that nearly half of all people dislike should be sorted below a post which 90% of people like.

It's also a bad idea to divide upvotes by total votes, since then a post with 1 upvote would be sorted above a post with 1000 upvotes and 1 downvote. It's likely that the 100% score for the first post isn't a reflection of how it should be sorted due to its small number of votes.

This sorting algorithm linked above finds a statistical confidence interval which describes what the actual score is likely to be after more people see it and vote, then takes the lower bound of it. A post with many votes has a smaller confidence interval around its current score, but a post with few votes has a wide confidence interval which tends to sort it below posts with more votes.

What part(s) of the site page(s) are affected?
Searching

  • 1