I've been toying with the idea of adapting the Elo rating system from chess to a social site's user score. The basic idea from chess is that you assume someone has an intrinsic skill in some arbitrary units (I think you start out around 1,000 and a 2400+ is senior master) and that we can only infer that skill from how they perform in tournaments. Furthermore, your skill in any particular game is assumed to be normally distributed about that intrinsic skill value. It is also assumed that your skill will only change slowly over time (we don't have to pin down a moving target).

So the mechanics of this system are basically that when you compete with someone that is much better than you and win your score should be adjusted upwards. Conversely, if you lose to someone worse than you, your score will be adjusted downward. Adjustments are done in batches, say after a player has finished a tournament (I use five voting events between readjustment).

My application to a user karma system would be analogous in that a "match" or "competition" would happen every time someone directly rates someone else (stories may be exempt). A high-karma user upvoting a low-karma user would increase the low-karma user's karma and decrease the high-karma user's karma. If there are two users with equal karma and one up or downvotes the other. the number of points exchanged will be minimal (the gain/loss scales with the difference in karma).

The idea is that your ideas are competing with the opinions of the high-ranked users. The assumption is that new users will have bad ideas and that high-karma users will downvote them. A downvote from a high-karma user is expected and doesn't cost the new user very much (they are expected to "lose" to them often).

I've sketched out some of the functions here: http://paste.lisp.org/display/76193

EDIT: grammar fix, typed this before coffee