# What is an alternative to Wilson score interval to calculating the 'best' rated items?

It would appear that Reddit uses the Wilson score interval to calculate the 'best' comment in the stack.

An explanation can be found here: http://amix.dk/blog/post/19588

However, this only seems to apply if you have a binary rating system (e.g. upvotes/downvotes). If instead, you have a singular rating system (like Facebook "Likes"), this method won't work as p = n ( p being the # of positive votes, n being the # of votes).

One method I can imagine that may work is set n to be the # of votes total in the system, or the # of votes on the maximally rated item.

Is this approach flawed?

-
What is your mathematical definition of "not flawed"? –  Aryabhata Mar 30 '12 at 6:23

## 1 Answer

The entire problem that the Wilson score solves only comes up because there are upvotes and downvotes and it's not immediately clear how $4$ up, $2$ down compares with $5$ up, $3$ down. If you have only positive votes, the problem doesn't arise, and it's obvious that the only way to select the 'best' comment is to select the one with the most upvotes, since there's no other information and thus no basis to prefer one with fewer upvotes to one with more upvotes. (This is on the assumption that you don't have additional information like the timestamps of the upvotes, but trying to make use of that would be a completely different problem from the one solved by the Wilson score.)

-