I don’t have a good model of what this topping out will look like. My intuition is that there’s quite a bit of variance in the top 0.1% though I agree that the case is weaker for a normal distribution. My reasoning for why “student goodness” is probably not a normal distribution is partially because if you care about multiple relatively independent factors (say smarts, conscientiousness, and general niceness) in a multiplicative way, and the individual factors are normally or log-normally distributed, your resulting distribution is a log-normal.
One funny hypothesis that someone like Bryan could give is something like “oh, the top EA students are all libertarian (ie, it’s the same picture).
I could maybe buy this, based on the (US-biased, Bay Area-biased) student groups I interact with, and especially if I factor in probably some pro-libertarian bias in Caplan’s judgements of students.
By topping out I just meant that Bryan’s impression of students I saturated at an upper bounded for the best libertarian and EA students they met. This could be either be because the students that they met actually did have a very high and similar performance, or because their model/test for assessing how good people are had the best students scoring very highly (so their test was not well calibrated to show differences in top students).
I think this was probably a bit of an off hand remark, and Stefan is right in it being very weak evidence of the performance of top libertarian/EA students.
I personally agree with all of your comment and agree that the underlying distribution seems unlikely to be normal.
I don’t have a good model of what this topping out will look like. My intuition is that there’s quite a bit of variance in the top 0.1% though I agree that the case is weaker for a normal distribution. My reasoning for why “student goodness” is probably not a normal distribution is partially because if you care about multiple relatively independent factors (say smarts, conscientiousness, and general niceness) in a multiplicative way, and the individual factors are normally or log-normally distributed, your resulting distribution is a log-normal.
One funny hypothesis that someone like Bryan could give is something like “oh, the top EA students are all libertarian (ie, it’s the same picture).
I could maybe buy this, based on the (US-biased, Bay Area-biased) student groups I interact with, and especially if I factor in probably some pro-libertarian bias in Caplan’s judgements of students.
By topping out I just meant that Bryan’s impression of students I saturated at an upper bounded for the best libertarian and EA students they met. This could be either be because the students that they met actually did have a very high and similar performance, or because their model/test for assessing how good people are had the best students scoring very highly (so their test was not well calibrated to show differences in top students).
I think this was probably a bit of an off hand remark, and Stefan is right in it being very weak evidence of the performance of top libertarian/EA students.
I personally agree with all of your comment and agree that the underlying distribution seems unlikely to be normal.