4 ms·

Yeah, it's essentially the mean karma (subtracting 1 from each karma) divided by the standard deviation. Looks like you've re-invented the signal to noise ratio

Yeah, it's essentially the mean karma (subtracting 1 from each karma) divided by the standard deviation. Looks like you've re-invented the signal to noise ratio formula, or the reciprocal of the coefficient of variation formula (standard deviation divided by the mean), although they recommend not making the mean be zero. http://en.wikipedia.org/wiki/Signal-to-noise_ratio#Alternative_definition http://en.wikipedia.org/wiki/Signal-to-noise_ratio#Alternati... It's probably going to have problems. Wouldn't it punish someone who had mostly good comments and posts, and occasionally gets one with a huge amount of karma, versus someone who never got one with big values of karma.

tptacek 15y ago
Can you address that by just discarding outliers?
tansey 15y ago
>Looks like you've re-invented the signal to noise ratio formula Well, I didn't, Sharpe did. :) However, the important distinction is the notion of a risk-free rate of return. In this case, it's (loosely) the 1 upvote you automatically get for every comment; in finance, it's usually the return you get on US Treasuries (around 1%, thought right now it's effectively 0%). >Wouldn't it punish someone who had mostly good comments and posts, and occasionally gets one with a huge amount of karma, versus someone who never got one with big values of karma. Assuming all else is equal (meaning it's N comments of karma K vs. N+1 comments where the first N are of karma K and the N+1th comment is something huge relative to K)? No. This makes sense if you think about how the standard deviation is derived, also note that I am capping the minimum standard deviation at 1 so consistently hitting the same karma does not give you an infinite score. However, if both individuals have the same mean karma but one got it from consistently scoring around that mean, and another got it from having one huge upvoted comment and several smaller ones, then yes. But isn't that what we want?

tptacek 15y ago

Can you address that by just discarding outliers?

tansey 15y ago

>Looks like you've re-invented the signal to noise ratio formula Well, I didn't, Sharpe did. :) However, the important distinction is the notion of a risk-free rate of return. In this case, it's (loosely) the 1 upvote you automatically get for every comment; in finance, it's usually the return you get on US Treasuries (around 1%, thought right now it's effectively 0%). >Wouldn't it punish someone who had mostly good comments and posts, and occasionally gets one with a huge amount of karma, versus someone who never got one with big values of karma. Assuming all else is equal (meaning it's N comments of karma K vs. N+1 comments where the first N are of karma K and the N+1th comment is something huge relative to K)? No. This makes sense if you think about how the standard deviation is derived, also note that I am capping the minimum standard deviation at 1 so consistently hitting the same karma does not give you an infinite score. However, if both individuals have the same mean karma but one got it from consistently scoring around that mean, and another got it from having one huge upvoted comment and several smaller ones, then yes. But isn't that what we want?