While I’m sympathetic to the usual criticisms of donation matching, I don’t think your critique applies very well to the quoted sentence as written. Instead, I think its validity comes on the matched donor end (GiveWell’s 2011 critique likewise seems to imply that matching works out okay for the matching donor.)
On the matching donor end, they really do get $1 of direct impact plus influence over someone else’s $1 of impact. Unless one thinks that influence is worth $0, then it seems reasonable to characterize the combined output as “worth” more than $1. I would prefer the use of increase rather than multiply in the quoted sentence, because the reader may associate multiply with integers more readily than they would associate increase. As a whole, I think the original post fairly conveys to matching donors what outputs they are getting for their money. Thus, I find the stylized example—in which Alice and Bob are both being promised $2 of impact—to not track the original post very well.
I also think it’s generally OK to calculate ($1 + influence kicker) for the matcher and ($1) for the matched. A strict requirement that all impact credit must equal no more than $2 is not consistent with socially accepted understandings, and there is an degree of flexibility that is non-misleading in a social context.
[numbers/details made up] Suppose I gave GiveWell $1 to run non-matching based ads, and people donated $10 as a result, which GiveWell then regranted to AMF. Both use specially-designated donations to cover administrative overhead. The prorata share of GiveWell’s admin overhead is fifty cents, and the pro rata share of AMF’s is 25 cents. Moreover, GiveWell and AMF employees indirectly contributed fifty cents by accepting less than fair market value for their work. So the total cost of the “$10 of impact” is actually $12.25. We would not expect GiveWell to tell donors that their $1 was worth only 81.6 cents of impact. Nor would we assign zero impact to me, the admin-overhead donors, or the employees. There’s some “inflation,” but I don’t think anyone has been deceived in this hypo. More generally, as long as the charity is transparent about what the matching donor is getting, and doesn’t claim they deserve all of the impact of the matched donation, I don’t see an obvious problem as to the matching donor.
(FWIW I find “matching / matched / matcher” confusing, and will stick to Thom’s “standard donor / bonus donor” terminology. Sorry if I’ve responded to you assuming you meant one of these things when you actually meant the other.)
Unless one thinks that influence is worth $0, then it seems reasonable to characterize the combined output as “worth” more than $1.
To the extent that influence over the $1 has any value, it’s value that is taken from the bonus fund donor. By donating to the bonus fund instead of donating directly to charity, you lose influence over where the money goes. There’s still a potential for value creation here if the influence is worth more in the hands of the standard donor than the bonus donor, but I would if anything expect it to go the other way, that standard donors on average tend to make worse choices than bonus donors.
A strict requirement that all impact credit must equal no more than $2 is not consistent with socially accepted understandings, and there is an degree of flexibility that is non-misleading in a social context.
I disagree, I think we should indeed be strict with this requirement. What’s socially accepted outside of EA circles isn’t a good guide here IMO – after all, almost the whole reason EA exists is because the broader social consensus on how to do good has some pretty serious flaws.
I think the reason to be strict with this requirement is because failure to do so can lead to people choosing worse actions over better ones – e.g. if you’re telling your two $1 donors that they’re responsible for $1.20 and $1 of impact respectively, then they ought to be willing to pay up to $0.20 of payment processing fees between them to make it happen, but that would ultimately lead to less money going to charity.
(To be clear, when I say “strict” I mean only “any deviation from this is a reasoning error”, not “any such errors are unacceptable” or whatever. Mistakes are OK, but this is a mistake.)
I think it’s relevant and useful that I noticed the “influence bonus” that the standard donor gets comes from the bonus donor losing influence.
I didn’t notice this at first! But I thought to myself “ok, Jason’s argument that the standard donor gets more than $1 of value sounds right, but I know there’s only $2 value in the inputs. Where is the extra value coming from? If it’s being created, how is it being created?”
This kind of question really only makes sense if you stand by the “$2 in, $2 out” kind of thinking, so I think this is a good example of why that’s a good principle that led me to clarify my thinking and notice something new.
If all the effective charities have similar levels of effectiveness, then the bonus donor probably doesn’t care about having influence within that subset?
No, the standard donor is meant to be a different audience, someone who isn’t seeking to maximise effectiveness—that’s why the post asks EAs not to be standard donors.
My use of non-”standard” terminology seems to have caused confusion; if I understand yours correctly, I think it is the bonus donor who usually gets more than $1 of value. The bonus donor surrenders control over which of six super-effective charities receives his $X. In exchange, the bonus donor gains influence over whether $2X from the standard donor goes to a standard charity or a super-effective one. Given EA beliefs about how starkly more cost-effective its preferred charities are over the median charity, the bonus donor likely views this control-for-influence trade as a net positive unless they have a strong preference among the six super-effective charities. It’s the standard donor who is getting the short end of the stick in my estimation.
However, I think “$2 in, $2 out” obscures that mutually beneficial trades of control/influence can occur. The existence of mutually beneficial trades is, after all, fundamental to economics. When we are discussing trades in an economic marketplace, we assign value based on how each trade participant values what they gave up and what they got. If I buy a widget at $1 but would have been willing to pay up to $1.25 for it, we would say that the trade has generated 25 cents of consumer surplus. We wouldn’t say that the widget must be assigned the same value on both ends of the transaction, and thus all trades are zero-sum and surplus cannot exist. Applying this model back to matching, it would seem consistent of us to calculate aggregate donor value based on the honestly held preferences of each donor and the true economic substance of the transaction.
So it’s possible for a trade to result in more utils under the preferences of both donors and those of the facilitator. Maybe the most extreme example would be a case in which I was planning to donate $1 to Politician Red, and you were planning to donate $1 to Politician Blue, but we both surrendered full control over our donation in exchange for influencing the other to donate the $1 to AMF instead. Or less dramatically, suppose I slightly favor effective animal-welfare charities over effective health ones, while my brother is relatively indifferent between effective and his favored non-effective health charities (but strongly prefers either to any animal-welfare charity). We can compromise on the effective health charity, rather than splitting 50⁄50 between our first choices.
On those trades, it seems really incongruous to say that there was $2 of impact before, and yet only $2 after. In the me/brother example, you could say that there was only $1.10 of impact pre-trade ($1 for me, 10 cents for my brother) and $1.90 post-trade (95 cents for each dollar spent, since we stipulated that the animal-welfare charity was slightly more impactful). I would roughly allocate that impact as $1.40 for me / 50 cents for my brother. Why? It seems illogical to assign less impact for the donor’s own dollar than they would have attained prior to a mutually beneficial trade, and “split the surplus” seems like a reasonable default rule. But the end result is that the trade (which is equivalent in effect to a “match”) has indeed increased the impact attributable to both my brother and to me! So, the claim that one can often increase impact by trading control/influence with a donor who would otherwise make suboptimal choices would evaluate as true. And I think that’s the core claim that donation-matching orgs are often making, at least to EAs?
While I’m sympathetic to the usual criticisms of donation matching, I don’t think your critique applies very well to the quoted sentence as written. Instead, I think its validity comes on the matched donor end (GiveWell’s 2011 critique likewise seems to imply that matching works out okay for the matching donor.)
On the matching donor end, they really do get $1 of direct impact plus influence over someone else’s $1 of impact. Unless one thinks that influence is worth $0, then it seems reasonable to characterize the combined output as “worth” more than $1. I would prefer the use of increase rather than multiply in the quoted sentence, because the reader may associate multiply with integers more readily than they would associate increase. As a whole, I think the original post fairly conveys to matching donors what outputs they are getting for their money. Thus, I find the stylized example—in which Alice and Bob are both being promised $2 of impact—to not track the original post very well.
I also think it’s generally OK to calculate ($1 + influence kicker) for the matcher and ($1) for the matched. A strict requirement that all impact credit must equal no more than $2 is not consistent with socially accepted understandings, and there is an degree of flexibility that is non-misleading in a social context.
[numbers/details made up] Suppose I gave GiveWell $1 to run non-matching based ads, and people donated $10 as a result, which GiveWell then regranted to AMF. Both use specially-designated donations to cover administrative overhead. The prorata share of GiveWell’s admin overhead is fifty cents, and the pro rata share of AMF’s is 25 cents. Moreover, GiveWell and AMF employees indirectly contributed fifty cents by accepting less than fair market value for their work. So the total cost of the “$10 of impact” is actually $12.25. We would not expect GiveWell to tell donors that their $1 was worth only 81.6 cents of impact. Nor would we assign zero impact to me, the admin-overhead donors, or the employees. There’s some “inflation,” but I don’t think anyone has been deceived in this hypo. More generally, as long as the charity is transparent about what the matching donor is getting, and doesn’t claim they deserve all of the impact of the matched donation, I don’t see an obvious problem as to the matching donor.
(FWIW I find “matching / matched / matcher” confusing, and will stick to Thom’s “standard donor / bonus donor” terminology. Sorry if I’ve responded to you assuming you meant one of these things when you actually meant the other.)
To the extent that influence over the $1 has any value, it’s value that is taken from the bonus fund donor. By donating to the bonus fund instead of donating directly to charity, you lose influence over where the money goes. There’s still a potential for value creation here if the influence is worth more in the hands of the standard donor than the bonus donor, but I would if anything expect it to go the other way, that standard donors on average tend to make worse choices than bonus donors.
I disagree, I think we should indeed be strict with this requirement. What’s socially accepted outside of EA circles isn’t a good guide here IMO – after all, almost the whole reason EA exists is because the broader social consensus on how to do good has some pretty serious flaws.
I think the reason to be strict with this requirement is because failure to do so can lead to people choosing worse actions over better ones – e.g. if you’re telling your two $1 donors that they’re responsible for $1.20 and $1 of impact respectively, then they ought to be willing to pay up to $0.20 of payment processing fees between them to make it happen, but that would ultimately lead to less money going to charity.
(To be clear, when I say “strict” I mean only “any deviation from this is a reasoning error”, not “any such errors are unacceptable” or whatever. Mistakes are OK, but this is a mistake.)
oh, I also want to add that:
I think it’s relevant and useful that I noticed the “influence bonus” that the standard donor gets comes from the bonus donor losing influence.
I didn’t notice this at first! But I thought to myself “ok, Jason’s argument that the standard donor gets more than $1 of value sounds right, but I know there’s only $2 value in the inputs. Where is the extra value coming from? If it’s being created, how is it being created?”
This kind of question really only makes sense if you stand by the “$2 in, $2 out” kind of thinking, so I think this is a good example of why that’s a good principle that led me to clarify my thinking and notice something new.
If all the effective charities have similar levels of effectiveness, then the bonus donor probably doesn’t care about having influence within that subset?
Yes, but then the standard donor doesn’t care about having the influence either, right?
No, the standard donor is meant to be a different audience, someone who isn’t seeking to maximise effectiveness—that’s why the post asks EAs not to be standard donors.
My use of non-”standard” terminology seems to have caused confusion; if I understand yours correctly, I think it is the bonus donor who usually gets more than $1 of value. The bonus donor surrenders control over which of six super-effective charities receives his $X. In exchange, the bonus donor gains influence over whether $2X from the standard donor goes to a standard charity or a super-effective one. Given EA beliefs about how starkly more cost-effective its preferred charities are over the median charity, the bonus donor likely views this control-for-influence trade as a net positive unless they have a strong preference among the six super-effective charities. It’s the standard donor who is getting the short end of the stick in my estimation.
However, I think “$2 in, $2 out” obscures that mutually beneficial trades of control/influence can occur. The existence of mutually beneficial trades is, after all, fundamental to economics. When we are discussing trades in an economic marketplace, we assign value based on how each trade participant values what they gave up and what they got. If I buy a widget at $1 but would have been willing to pay up to $1.25 for it, we would say that the trade has generated 25 cents of consumer surplus. We wouldn’t say that the widget must be assigned the same value on both ends of the transaction, and thus all trades are zero-sum and surplus cannot exist. Applying this model back to matching, it would seem consistent of us to calculate aggregate donor value based on the honestly held preferences of each donor and the true economic substance of the transaction.
So it’s possible for a trade to result in more utils under the preferences of both donors and those of the facilitator. Maybe the most extreme example would be a case in which I was planning to donate $1 to Politician Red, and you were planning to donate $1 to Politician Blue, but we both surrendered full control over our donation in exchange for influencing the other to donate the $1 to AMF instead. Or less dramatically, suppose I slightly favor effective animal-welfare charities over effective health ones, while my brother is relatively indifferent between effective and his favored non-effective health charities (but strongly prefers either to any animal-welfare charity). We can compromise on the effective health charity, rather than splitting 50⁄50 between our first choices.
On those trades, it seems really incongruous to say that there was $2 of impact before, and yet only $2 after. In the me/brother example, you could say that there was only $1.10 of impact pre-trade ($1 for me, 10 cents for my brother) and $1.90 post-trade (95 cents for each dollar spent, since we stipulated that the animal-welfare charity was slightly more impactful). I would roughly allocate that impact as $1.40 for me / 50 cents for my brother. Why? It seems illogical to assign less impact for the donor’s own dollar than they would have attained prior to a mutually beneficial trade, and “split the surplus” seems like a reasonable default rule. But the end result is that the trade (which is equivalent in effect to a “match”) has indeed increased the impact attributable to both my brother and to me! So, the claim that one can often increase impact by trading control/influence with a donor who would otherwise make suboptimal choices would evaluate as true. And I think that’s the core claim that donation-matching orgs are often making, at least to EAs?