A different (and to my mind immediately convincing) presentation of gwern’s data: plot on imgur.
This shows, for each of four age brackets and each of two EA-ness-es, a histogram of reported charitable giving. The histogram bins are divided at integer values of gwern’s “CharityLog” (so the first one is people who gave 0, then exponentially increasing donations for successive bins). The age brackets are <20, 20-30, 30-40, 40-70. I excluded rows for which CharityLog, Age or EffectiveAltruism wasn’t defined, or for whom the age was 70.
In all four age ranges, (1) the self-reported EAs were less likely to report giving nothing and (2) the self-reported EAs’ histogram was substantially rightward of the self-reported non-EAs’.
I haven’t attempted any sort of formal analysis (significance tests, etc.) on this, because gwern already did some and because it’s immediately obvious from the graphs (plus the fact that the numbers aren’t tiny) that the difference here is significant and not small.
(Tiny caveat to last claim: one might, if desperate, seize upon the fact that in the youngest age group the single largest reported donation was from a non-EA and suggest that the right tail of the distribution might be fatter for non-EAs, which could make them more generous overall. Maaaybe. If anyone thinks that hypothesis worth taking seriously, they’re free to analyse the data and see what they find. I know which way I’d bet.)
I remark that this shows that for none of these combinations of age and EA-ness was the median reported donation zero, which is hard to square with su3su2su1′s claim. (Perhaps there are more zeros among people whose age and/or EA-ness couldn’t be determined? Perhaps “didn’t say” is being counted as “zero” in some contexts? Or, though I hesitate to suggest such things, perhaps su3su2su1 said something incorrect?)
I’m not sure where su3su2su1 got his median number of 0; Yvain’s 2014 and 2013 survey writeups never looked at donations split by EA, IIRC, so I assume he must’ve gotten it from somewhere else, maybe his own work, in which case perhaps it was a coding error or his tool made questionable choices like treating NAs as 0s or something like that.
I don’t think su3su2su1 made any claim involving donations split by EA-ness; wasn’t it just that the median donation reported in the LW survey[1] (as a whole) was zero? (The underlying assumption presumably being that LW is a Wretched Hive of Scum and Effective Altruism, so that if LW users generally aren’t generous then EAs must likewise not be generous.)
[1] Which LW survey? No idea, of course. It seems fairly clear that su3su2su1 was prioritizing for rhetorical effect over accurate analysis anyway.
A different (and to my mind immediately convincing) presentation of gwern’s data: plot on imgur.
This shows, for each of four age brackets and each of two EA-ness-es, a histogram of reported charitable giving. The histogram bins are divided at integer values of gwern’s “CharityLog” (so the first one is people who gave 0, then exponentially increasing donations for successive bins). The age brackets are <20, 20-30, 30-40, 40-70. I excluded rows for which CharityLog, Age or EffectiveAltruism wasn’t defined, or for whom the age was 70.
In all four age ranges, (1) the self-reported EAs were less likely to report giving nothing and (2) the self-reported EAs’ histogram was substantially rightward of the self-reported non-EAs’.
I haven’t attempted any sort of formal analysis (significance tests, etc.) on this, because gwern already did some and because it’s immediately obvious from the graphs (plus the fact that the numbers aren’t tiny) that the difference here is significant and not small.
(Tiny caveat to last claim: one might, if desperate, seize upon the fact that in the youngest age group the single largest reported donation was from a non-EA and suggest that the right tail of the distribution might be fatter for non-EAs, which could make them more generous overall. Maaaybe. If anyone thinks that hypothesis worth taking seriously, they’re free to analyse the data and see what they find. I know which way I’d bet.)
R code (note: I have no idea what I’m doing):
[EDITED to add: oh, I see I can embed an image in a comment. Here we go:]
I remark that this shows that for none of these combinations of age and EA-ness was the median reported donation zero, which is hard to square with su3su2su1′s claim. (Perhaps there are more zeros among people whose age and/or EA-ness couldn’t be determined? Perhaps “didn’t say” is being counted as “zero” in some contexts? Or, though I hesitate to suggest such things, perhaps su3su2su1 said something incorrect?)
I’m not sure where su3su2su1 got his median number of 0; Yvain’s 2014 and 2013 survey writeups never looked at donations split by EA, IIRC, so I assume he must’ve gotten it from somewhere else, maybe his own work, in which case perhaps it was a coding error or his tool made questionable choices like treating NAs as 0s or something like that.
I don’t think su3su2su1 made any claim involving donations split by EA-ness; wasn’t it just that the median donation reported in the LW survey[1] (as a whole) was zero? (The underlying assumption presumably being that LW is a Wretched Hive of Scum and Effective Altruism, so that if LW users generally aren’t generous then EAs must likewise not be generous.)
[1] Which LW survey? No idea, of course. It seems fairly clear that su3su2su1 was prioritizing for rhetorical effect over accurate analysis anyway.
Of course he did. It’s literally the first line of my post:
Wow. I totally can’t read. You are of course completely right. Sorry about that.