p(benefit) is an awful way of looking at these medications. Yes, if you pull a random medication off the shelf and try it on someone at random, of course you'll

p(benefit) is an awful way of looking at these medications. Yes, if you pull a random medication off the shelf and try it on someone at random, of course you'll have a terrible result! Go ahead, walk into the pharmacy and try a random medication off the wall! I dare you! But if you find the right one it is life-changing. And there are enough of them out there that each one only needs to be a hair better than random for it to be unlikely that none of them work. Same math that powers techniques like random forests. I don't know what the numbers are, but if you look at p(any medication working) instead of p(this medication working), it's damn sure not sugar-pill statistics.