Comment author:jimmy
11 December 2009 07:00:01PM
1 point
[-]

Neither really, but there's no easy way to do it.

The mean has the problem that if a lot of people claim near ignorance (like I did), then that counts against Knox, when really, it doesn't mean anything.

The problem with the geometric mean is that it is biased towards the low end of the spectrum, so it depends on if the statement is negated or not. (GM(.001,.999)<4%)

The median is probably better than both, but it's still not the right way to do it.

Ideally you'd try to count up how much evidence each person saw and add those, but it is no easy task to estimate how correlated the evidence is (though it's probably a worthwhile subject to put thought into, since thats how you determine how much an additional persons belief is worth)

Even with a large degree of overlap, this is probably one of the cases where sharing beliefs should make everyones beliefs more extreme.

I'd sorta like to see what it'd look like on round two.

## Comments (260)

Best*7 points [-]I averaged up the answers given so far in this thread for Knox and got 35% mean, 20% median.

Is this an arithmetic mean or a geometric mean?

Which is the correct mean to use for averaging probabilities, anyway?

*3 points [-]The arithmetic mean of the log odds is pretty natural. It is 27%, but the median looks like 30% to me.

Neither really, but there's no easy way to do it.

The mean has the problem that if a lot of people claim near ignorance (like I did), then that counts against Knox, when really, it doesn't mean anything.

The problem with the geometric mean is that it is biased towards the low end of the spectrum, so it depends on if the statement is negated or not. (GM(.001,.999)<4%)

The median is probably better than both, but it's still not the right way to do it.

Ideally you'd try to count up how much evidence each person saw and add those, but it is no easy task to estimate how correlated the evidence is (though it's probably a worthwhile subject to put thought into, since thats how you determine how much an additional persons belief is worth)

Even with a large degree of overlap, this is probably one of the cases where sharing beliefs should make everyones beliefs

moreextreme.I'd sorta like to see what it'd look like on round two.

*0 points [-][comment deleted]