I tend to update articles only when I remember their content and realize that I want to change something about it. But I rarely remember it well enough once about two years have passed. Such articles are therefore likely to contain some statements that I no longer espouse or would today frame differently.
At least since GiveWell’s 2014 series of articles on donor coordination it has been clear that donor coordination is an important project to realize so large donors can split their funding rather than wait each other out, and if there are thresholds to realizing a project that neither can meet alone, they can pool resources.
But another advantage of donor coordination concerns all donors, and in the short term especially donors who would like to donate to EA metacharities. To convey the idea, I need to critique my previous proposal of expected utility auctions.
Prioritization research as GiveWell and Animal Charity Evaluators conduct it combines a number of advantages:
- EAs know where they can donate,
- it coordinates donors insofar as the estimates for room for more funding are updated in yearly cycles or more often,
- but most importantly, it generates a cash flow to charities that is a function of their efficiency at producing positive impact:
- thus linking a metric that they cannot ignore for their own sake (money) to a metric that they morally mustn’t ignore (impact),
- thus incentivizing less effective charities to become more effective (and conduct and publish research to get there) to partake of that cash flow or even to retain donors if the movement builders are very successful,
- thus creating a more efficient market for altruistic investments, which raises the effectiveness waterline of the whole charity sector.
Depending on how much money we EAs have at our disposal in the movement, there is more or less value in the the first two points – but the overwhelming value for the vast majority of society, the long-run future, and moral process is in the third point. And GiveWell has reaped all the low-hanging prioritization fruit here, so it now needs to climb higher in the metaphorical tree (with Open Phil). (ACE may well have some low-hanging fruit left to pluck.)
But prioritization organizations need to also apply prioritization to their own research. When the funding gap of a charity is so limited that the expected value of information gained by reviewing it is lower than the price in staff time paid for the review, conducting the review would come at a positive opportunity cost, which translates into suffering that could’ve been averted. Hence, charities need to be somewhat scalable to qualify for review.
In my post on expected utility auctions I explained that this leads to a recommendation gap for smaller and more speculative opportunities, especially now that it has been announced that Open Phil is not planning to put out public recommendations. Since donors don’t (and would be ill-advised to) split up their donations arbitrarily between countless charities, they cluster around the few safe bets that GiveWell and ACE have found, and insofar as they don’t, their donations are split up across scores of smaller or more speculative endeavors, so that this support becomes less visible. This looks like a streetlight effect but is really quite rational.
As a remedy I proposed to lower the bar for prioritization research, and crowdsource it to highly transparent hobbyists who can share the responsibility for the results in a StackExchange-like system.
The value of that system needs to be divided by the months of work that would have to go into community building to get it going and multiplied by the probabilities that this community building effort will be successful and that the resulting recommendations will be valuable. MVP versions of the system were proposed in the comments, but it would still require a significant leap of faith for someone to drive the project.
Instead, how about donor coordination?
The majority of the small charities that we wish to donate to that are not in the focus of GiveWell and ACE will be charities in prioritization research itself (let’s include 80,000 Hours here for the moment) and in EA movement building. These may be further subdivided, but I’ll use these two as examples for “program areas” for now. (Or “intervention areas,” but “program areas” sounds smoother to me.)
Within these two program areas there are only charities that get EA, eliminating the need for the third and major advantage of prioritization research. Hence it will be much less cost-effective to engage in. That leave the first two advantages.
Further, the donor coordination problem GiveWell faced may, to a large part, not apply. The problem only occurs when donors have substantially more faith in their own judgements than in that of the other donors, either because they’ve put a lot of effort into forming them or because they have outré moral goals. In my experience, neither is common, at least among the donors of less than €10,000 per year that I’ve talked to, who are probably not the category GiveWell worries about.
In my experience donors often have the opposite problem. They wish to drive, e.g., EA movement building with their donations but don’t know which organization to donate to, even after doing substantial research, because they all seem to do great and value-aligned work. Hence the more risk-averse donors then fall back on a GiveWell or ACE charity instead while the more risk-neutral ones pick out one at random or precisely because of its room for more funding, thinking of it as the best proxy for cost-effectiveness that they can find.
Room for more funding may be a decent proxy in some cases, but greater scale often correlates with lower cost-effectiveness, so that it also seems plausible that this metric might lead us astray – so long as we think of it as a proxy of an individual charity’s cost-effectiveness.
But here’s the rub: Unlike in the program area of net distributions, the total room for more funding in EA movement building probably doesn’t go far beyond the money that is available in the movement (if at all), so instead of thinking of our donation as one to an individual charity, we should think of it as a donation to a program area.
What we need for that is only:
a common definition of room for more funding,
a set of charities in the program area that are value-aligned, and thus can be trusted to report their room for more funding honestly,
a system for donor coordination, which can be as simple as a Facebook group as MVP.
Since every donor fills some room for more funding and thus displaces an equivalent donation, all donations are fungible and have equivalent counterfactual cost-effectiveness: the average marginal cost-effectiveness of the total filled room for more funding of the whole program area.
This happens anyway, and donor coordination can only make it better since on average highly cost-effective EA charities may currently receive more donations than they can invest at the margin, and since a lot of EA donors may currently avoid EA metacharities in favor of individual safe bets. We may not be able to detect the very best opportunities, but those wouldn’t be worth too much if they’d be filled by a fraction of the available donor money and the effectiveness waterline is generally high.
The question is no longer whether a marginal donation to GiveWell or 80,000 Hours is more cost-effective but whether prioritization research is sufficiently cost-effective to deserve our donations.
Prioritization research and EA movement building are some obvious first targets for the donor coordination project since there is a seizable consensus that these are still equally or more cost-effective than object-level donations, but what if we fill both these funding gaps?
GiveWell deprioritizes research in program areas when it finds that there are no promising charities implementing the program. If there are really no charities or no value-aligned charities, then this can’t be helped, but if there are many value-aligned charities, each of insufficient scale to warrant GiveWell’s attention, then an evaluation of the program area and a very cursory review of the charities (to ensure that they’re value-aligned) can be enough to provide the foundation for donor coordination in the area.
The meaning of room for more funding is highly unclear to me at the moment, which would need to change in order to get commensurable estimates from different charities.
Does it mean the amount of funding an organization can invest before it would choose to leave further funds on its bank account beyond the end of the fiscal year? Or before a hypothetical investment of further funds would incur a significant drop in marginal cost-effectiveness? If so, how significant?
As an example, I imagine that AMF has to be fairly liquid, so it probably has a lot of money on bank accounts with low interest. So in the hypothetical case where it gets more money than it would’ve normally invested in the year, it has to weigh whether it wants to leave it there until the next year, when it’ll get better distributions finalized, or whether it is better to invest it less well than AMF would typically invest it. Maybe there are regions where it could easily conduct distributions, e.g., because Concern Universal works there, but that it has so far avoided because the malaria incidence is slightly lower than in others or the IR rate slightly higher. In general, suffering accumulates over time, IR rates increase, incidence decreases, so the second option might sometimes be better.
So if room for more funding just means that the extraneous money will be left on the account until after the end of the year, then it’s equivalent to some degree of a drop in marginal cost-effectiveness that AMF was not ready to accept. So what drop is that? A significant drop from AMF’s accustomed level of cost-effectiveness, or a drop below GiveDirectly level, or a drop not to GiveDirectly level but significant enough that given the other factors in the evaluation, it would temporarily lose its top charity status?
For EA movement building and possibly prioritization research, the leverage multiplier for long-run estimated cost-effectiveness beyond object-level donations might serve as such a baseline below which the room for more funding of a charity is closed. It might be set higher than one to account for the increased uncertainty of meta work. “Long-run,” because in the short term, heavy reinvestment into one’s own program will lead to a quicker increase in scale.
For other program areas, different metrics would have to be developed.
And as an ancillary question, in view of home bias and the heavy commitment that founders make to their charities, how likely is it that they might rationalize in favor of their own charity being the most effective, which would increase the room it has until it reaches some lower marginal effectiveness threshold and thus make it moral for them to direct somewhat more funds toward it? By extension, staffers might assume that this partiality affects all the other charities, so that their own would end up relatively neglected, which would again be objectively bad in their view.
I’d be happy for my donations to ACE to be fungible with donations to the Global Priorities Project, GiveWell, etc. assuming that they have funding gaps, but Descartes would disagree. Hence what constitutes a program area has to be the result of a clustering of donor interests, the data for which is probably available to Giving What We Can and EA Hub.
Peter Hurford has started a Slack channel for donor coordination, which you can join if you have a Slack account. Evan Gaensbauer considers working on something like it. If you wish to help, please get in touch with them.