I agree that often “EA is unusually bad” isn’t grounded in data (often instead in first person accounts, although I’d note that if we assume reporting infrastructure won’t capture things well, that’s actually where most of the signal lives). I’m pretty skeptical that we should conclude from this data that EA isn’t unusually bad.
Others have made similar points but just at a high level, my concerns when comparing to the other data (e.g. Title IX) would be that we are biasing both the numerator and denominator of the reporting rate in a way that biases us to think we are doing better than we are.
Numerator: There are lots of different reporting mechanisms and places and so we shouldn’t expect CEA CH to capture all the cases as it is one of many parallel endpoints. Something like Title IX sits at the end of a mandatory reporting funnel. If I tell my RA that my professor harassed me, they are mandatory reporters of that. The same isn’t true for if I tell someone at org X that someone at org Y harassed me (it is true within a single workplace but because EA is quite fractured, this isn’t a mandatory reporting funnel to CEA CH).
Denominator: For a great deal of “EAs” what this means is just engaging online and donating to GiveWell. The Title IX denominator is people who are co-located on the same campus. The exposure-per-person is likely wildly different to other contexts.
I basically think errors here aren’t random so your number here is a reasonable lower bound but I’m not sure it provides insight into the overall rate.
I agree that often “EA is unusually bad” isn’t grounded in data (often instead in first person accounts, although I’d note that if we assume reporting infrastructure won’t capture things well, that’s actually where most of the signal lives). I’m pretty skeptical that we should conclude from this data that EA isn’t unusually bad.
Others have made similar points but just at a high level, my concerns when comparing to the other data (e.g. Title IX) would be that we are biasing both the numerator and denominator of the reporting rate in a way that biases us to think we are doing better than we are.
Numerator: There are lots of different reporting mechanisms and places and so we shouldn’t expect CEA CH to capture all the cases as it is one of many parallel endpoints. Something like Title IX sits at the end of a mandatory reporting funnel. If I tell my RA that my professor harassed me, they are mandatory reporters of that. The same isn’t true for if I tell someone at org X that someone at org Y harassed me (it is true within a single workplace but because EA is quite fractured, this isn’t a mandatory reporting funnel to CEA CH).
Denominator: For a great deal of “EAs” what this means is just engaging online and donating to GiveWell. The Title IX denominator is people who are co-located on the same campus. The exposure-per-person is likely wildly different to other contexts.
I basically think errors here aren’t random so your number here is a reasonable lower bound but I’m not sure it provides insight into the overall rate.
Can we move to a position where we are more uncertain on EA’s badness? And where we discuss it in those more uncertain terms?
On the second point, what is a more accurate estimate of the EA rate, in your view?