@paulg Works only with infinite pools of applicants. If First Round funded 100% of elite college students, could still get same results
-
-
-
@terzicigor In practice not. In practice there is a huge supply of people who went to elite colleges yet aren't very good.
End of conversation
New conversation -
-
-
@paulg Did you come up with it? It is brilliant and I need to know who to credit -
@tommyjensen I didn't learn about it from someone else, but I would be surprised if no one else had thought of it.
End of conversation
New conversation -
-
-
@paulg Problem: If one group *is* inherently better than another, you would expect better perf from that group even under fair selection -
@paulg So this test works for groups you expect to be roughly equal (men/women?) but not for those you expect to be different (colleges) -
@jasoncrawford Yes, that's right. Though in my experience alumni of different colleges don't vary as much as the colleges' brands imply.
End of conversation
New conversation -
-
-
@paulg After some thought, I don't think this method is robust against an active adversary seeking to hide bias against select subgroups. -
@paulg I think you'd have to consider the sub-group distributions and make sure that they are at least the same shape. -
@g33kbeast Do you mean precondition (c) or something else? -
@paulg Even if (c) is true, someone trying to hide bias can mess with selection process. Simpson's paradox in reverse.
End of conversation
New conversation -
-
-
@paulg if women are discriminated against on exits/next rounds, it lowers the bias measure in FRC example -
@inazarova@paulg Or any other discrimination along the way. The original idea is crisp and clever, but fails for this reason, IMO. -
@davidhelgason@inazarova@paulg W/o further discrimination performance would be even stronger. So, w/discrimination bias measure is higher? -
@davidhelgason@inazarova@paulg agreed, thanks.
End of conversation
New conversation -
-
-
@paulg super clever. Said another way, reversion to the mean should be present if sample not biased, and if it's not there may be a reasonThanks. Twitter will use this to make your timeline better. UndoUndo
-
-
-
@paulg Sama data also imply Investor bias *against* youth (< 25 years old) despite Conventional Wisdom being the opposite.Thanks. Twitter will use this to make your timeline better. UndoUndo
-
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.