Iāll start us off with the standard argument against donation splitting. Weāll start with the (important!) assumption that you are trying to maximize[1] the amount of good you can do with your money. Weāll also take for the moment that you are a small donor giving <$100k/āyear.
There is some charity that can use your first dollar to do the most good. The basic question that this line of argument takes is: is there some amount of money within your donation budget that will cause the marginal effectiveness of a dollar to that charity to fall below that of the second best charity.
For example, you could imagine that Acme Charity has a program that has only a $50k funding gap. After that, donations to Acme Charity would go towards another program.
The standard argument against donation splitting, which seems right to me, is that the answer to that question is āprobably not.ā
Hereās the donation splitting policy that I might argue for: instead of ādonate to the charity that looks best to youā, Iād argue for ādonate to charities in the proportion that, if all like-minded EAs donated their money in that proportion, the outcome would be bestā.
Hereās the basic shape of my argument: suppose there are 1000 EAs, each of which will donate $1000. Suppose further there are two charities, A and B, and that the EAs are in agreement that (1) both A and B are high-quality charities; (2) A is better than B on the current margin; but (3) A will hit diminishing returns after a few hundred thousand dollars, such that the optimal allocation of the total $1M is $700k to A and $300k to B. What policy should each EA use to decide how to allocation their donation? It seems like the two sensible policies are:
Donate $700 to A and $300 to B (donation splitting); or
Donāt donate all at the same time. Instead, over the course of giving season, keep careful track of how much A and B have received, and donate to whichever one is best on the margin. (In practice this will mean that the first few hundred thousand donations go to A, and then A and B will each be receiving donations in some ratio such that they remain equally good on the margin.)
But if you donāt have running counters of how much has been donated to A and B, the first policy is easier to implement. And both policies are better than the outcome where every EA reasons that A is better on the margin and all $1M goes to A.
Now, of course EAs are not a monolith and they have different views about which charities are good. But I observe that in practice, EAsā judgments are really correlated. Like I think itās pretty realistic to have a situation in which a large fraction of EAs agree that some charity A is the best in a cause area, with B a close second. (Is this true for AMF and Malaria Consortium, in some order?) And in such a situation, Iād rather that EAs have a policy that causes some fraction to be allocated to B, than a policy that causes all the money to be allocated to A.
Note that how this policy plays out in practice really does depend on how correlated your judgments are to those of other EAs. If Iām wrong and EAsā judgments are not very correlated, then donating all your budget to the charity that looks best to you seems like a good policy.
JP Addison
I like this position ā Iām already not sure how much I disagree. Some objections that might be more devilās advocate-y or might be real objections:
I agree correlation is important. Iām not sure how to define it and, once defined, whether it will be correlated enough in practice.
Roughly speaking, what decision theory /ā unit of analysis are we using here? It seems like your opening statement assumes we can set the norm for all EA. Whereas Iām thinking more about what Iād recommend to an individual who asked for my opinion. I want to avoid unilaterally doing something that only pays off when everyone does it, unless I really think that everyone will do it.
Eric Neyman
Cool, yeah, I agree that āhow much correlationā and āwhich decision theoryā are important uncertainties/ācruxes. Maybe to spell this out more, I think my argument requires several assumptions:
EAsā donations are quite correlated: the ways that different individual EA donors make decisions is correlated enough that, without communication/ācoordination between different donors, youād end up with a large chunk of EAs donating to a small set of orgs, maybe so much so that an individual EA would look at those donations and be like āit would be better if the donations were more diffuseā
Large unit of analysis: we are discussing a policy recommendation for EA donors writ large, rather than how an individual donor should behave. Or, we are making discussing what an individual donor should do, but in an evidential decision theory mindset where if the donor follows a given policy, thatās evidence that other donors will as well. Or something like that. I generally find this confusing to think about.
Lack of successful communication/ācoordination: even supposing that EAsā donations are very correlated, if there were enough communication bandwidthāe.g. if there were running donation counters on every orgās webpage and all the EA donors paid attention to those countersāthen this correlation wouldnāt pose a big problem. But in practice we donāt have such good coordination.
Small EA donors collectively hit diminishing returns: e.g. if all EAs agreed that the best org were Charity A, then if all the EAsā donations went to Charity A then Charity A would be oversaturated (i.e. it would be better if the EAs were to donate more to the next best charity on the margin).
JP Addison
Iām happy to concede assumption 3. It seems possible that assumption 4 isnāt true in some cases, but itās gotta be true for at least some cause areas, and Iām happy to drop it for now. Maybe a commenter could provide a useful analysis.
Points 1&2 both seem like important cruxes.
Correlation
JP Addison
I think we should first deal with a conceptual question about what type of correlation weāre interested in. Are we interested in a question of cross-cause splitting? Are we asking about correlation among all EAs? Or among donors who donate to a particular cause.
Ways JP would investigate the correlation question:
First approach: Do a research project where one asks GiveWell, Giving What We Can, etc about the correlation among donations they have visibility into.
Second approach: I think of like, 5 friends who donate to the same cause. And we can ask them where theyāre donating this year and where they donated last year. Maybe we use the CEA where are you donating post. Or the where are you donating thread. Which we could actually do within the course of this dialogue.
Eric Neyman
I think the relevant question of correlation is, like, is it enough to collectively hit diminishing returns? So for example, if there were zero of any kind of correlation, then the answer is clearly no, and if all EA donors were perfectly correlated, then it would depend on Assumption 4 above.
The truth is somewhere in the middle. To simplify, we can talk about correlation between cause areas (this means something like ālots of EAs decide to donate to AI safety, enough to collectively hit diminishing returns within the cause area, and they wish more of them had donated to farm animal welfare insteadā) and correlation within a cause area (ālots of EAs decide to donate to the Humane League, enough to collectively hit diminishing returns, and they wish more of them had donated to the Good Food Institute insteadā).
My guess is that correlation within a cause area is a bigger deal. Would you agree?
JP Addison
I agree the within cause area seems like the natural unit of analysis. Cross causes Iām much more suspicious of donation splitting, basically for the reason that I think assumption 4 fails.
Eric Neyman
Another thing we can look at is the EA Forum donation election, where we already have āpre-votesā. One basic question is, if small EA donors donated in proportion to these pre-votes, would that be good, or would some charities be oversaturated?
(This isnāt exactly the right question because the set of EA donors voting in the election will be more correlated than the set of all EA donors, but itās a start.)
JP Addison
Hereās a very hacky BOTEC approach we worked out via live collaboration:
We can use an old estimate of EA Fundās amount of donations processed, and this summary of how much the LTFF got this year to arrive at very rough estimate that charities will receive $50,000 per pre-vote.
We then can guesstimate that this giving season CE will get ~$3M.
From this we agree on a wild guess that this would leave CE in a ādiminishing marginal returns but not severelyā situation.
Eric Neyman
Over the course of talking to you/āthinking about this, I think Iāve gone from ācorrelation between EA donors is a medium-to-large problemā to ācorrelation between EA donors is a small-to-medium problemā. (I do still feel kinda conceptually confused about how to think about correlation and have a ton of uncertainty.) But probably(?) it would be better if donations to EA orgs were spread more uniformly than the above pre-votes.
JP Addison
A shift in my thinking from this correlation conversation is Iāve gone from thinking about correlation as a number where it needs to be pretty high to make Ericās argument go through, to:
Is the amount of current donation splitting plus correlation enough that in practice āāEA shouldāā³ donation split more
The answer to which seems quite plausibly to be yes.
Unit of analysis
JP Addison
Letās grant for a moment that we believe that currently:
CEās Incubated Charities Fund is the best target for donations
After Giving Season this year, CE will probably have done well enough to hit diminishing marginal returns enough to lower it beneath GiveWellās All Grants Fund
I want to defend for the moment the position that you should donate 100% to the All Grants Fund.
Reasoning:
Itās the thing which will causally have the most impact.
With the situation as you predict it to be, the world where you donate 100% to the AGF will be a better world than the world where you donate 70:30 to CE and the AGF.
Thereās not enough (correlation!) evidence to make an evidential or other non-causal decision theorist deviate from the causal strategy.
Like, in practice there are bunch of messy humans with pretty different views in pretty different situations weighing pretty different concerns.
Eric Neyman
Yeah, I think that probably makes sense in a context where youāre an individual donor who knows what strategies the other donors are using (whether theyāre splitting or not, and whether theyāre thinking about oversaturation or not). But I also think that:
Insofar as weāre in the business of making a recommendation/ābest guess about how EA donors as a whole ought to behave, and people are listening to us, then coordinating on some sort of mild donation splitting strategy would be reasonable.
Letās say youāre worried that your favorite charity will get oversaturated and so you decide to donate to your second-favorite charity. Are you right, or will other donors reason similarly to you and so your favorite charity will be underfunded? I guess this is just a matter of how much correlation you expect, but this time the thing that matters is correlation in meta-level strategy rather than object level charity opinions.
(But this argument feels too meta in a way that I feel like is detached from reality, so I probably donāt endorse it.)
JP Addison
Insofar as weāre in the business of making a recommendation/ābest guess about how EA donors as a whole ought to behave, and people are listening to us, then coordinating on some sort of mild donation splitting strategy would be reasonable.
I am kinda interested in a proposal here. I think if you could coordinate with a large enough group of donors, then I do think the donation splitting procedure in Ericās opening dialogue comment would do better than independent actions. Given that this dialogue is a public discussion, this post does seem like a good opportunity to do so.
I havenāt (within the time bounds of this dialogue) come up with a specific proposal, but Iād encourage commenters to make them. Maybe an assurance contract?
Dialogue on Donation Splitting
Correlation
Unit of analysis