It assumes average posts would have the same amount of upvotes and downvotes. This is rarely true - people don't care to downvote boring comments, they downvote ones they don't agree with more often. This is why considering views might be better idea.

Interesting :-) I didn't know about his article, but nevertheless that's roughly how I've currently implemented the sorting algorithm in the discussion system that powers http://www.debiki.com/demo/-71cs1-demo-page-1 — I'm using the simpler Agresti-Coull interval though (http://en.wikipedia.org/wiki/Binomial_proportion_confidence_... ), not the Wilson interval. (And I've largely forgotten all about mathematics since University, 10 years ago.)