Concept for Donor Coordination

Introduction

This proposal is meant to encourage comments on its content as well as comments along the lines of “I would use this,” because without many of those it will not seem like a worthwhile undertaking to implement it.

Donor Coordination

One problem that GiveWell has struggled with emerges when two donors are not fully value-aligned but can agree on wanting to fund one of GiveWell’s top charities. The result is that they wait each other out, a deadlock, both wanting the other to fund the GiveWell charity, because they value the other’s counterfactual use of the donation lower than their own. GiveWell is regularly refining its response to this problem.

For this problem to become relevant, there need to be at least two large donors or monolithic groups of donors, where large means that their planned donations are close to – for example within the same order of magnitude of – the funding gap of the charities in question. This is a good problem to have.

More commonly, however, the funding gap is large compared to the potential individual donations (where individual is meant to exclude the aforementioned monolithic groups of donors), so that the above problem becomes an edge case while centrally we face a different problem. Donors that focus their contributions on charities that have a significant evidence base and track record for impact – a large part of the “GiveWell market” – are often accused of being too focused on just these established charities thereby missing small high-impact opportunities from nonprofit startups or projects that will stay small or short-lived by design.

The distinction is similar to that between, on the one hand, passive investors that buy exchange-traded funds (ETFs) of, for example, the top 30 (DAX) or 500+ (S&P 500) companies in order to hold them, and, on the other hand, business angels or venture capitalists that invest into startups. The first group has excellent information to make relatively low risk–low return investments; the second group has to rely on rough heuristics, such as their faith into the founders, to make high risk–high return investments – of which they need to be able to make many in order to profit at least fairly reliably.

But a profit motive is an agent-relative goal. Investors (such as donors) with agent-neutral goals that are shared by at least a few others have much better opportunities for cooperation. These have largely not been tapped into. While Net Analytics is clearly focused on the low risk–low return market, this high risk–high return market also calls for a software solution to its coordination problem.

The central motivating problems are the following:

Irrational Risk-Aversion

Drops in marginal utility of a resource suggest risk aversion. In that context it is rational to prefer a low return with high probability to a high return with low probability at the same expected value. In the context of altruistic interventions,¹ the utility of marginal donations only noticeably decreases when it reaches the area of millions of dollars, some of the GiveWell top charities’ funding gaps. Since few donors have funds of that magnitude at their disposal, most risk aversion of average donors is disproportionate.

At the same time there are many donors that see a high likelihood that effective interventions are possible in a certain cause area. Unfortunately, these intervention are, by necessity, more speculative than, for example, the interventions GiveWell prioritizes. Yet there are charity startups implementing them.

The funding gaps of these charities tend to be too small for any respectable prioritization organization, like GiveWell, to warrant investing staff time into evaluating them, so donors are left to their own devices.²

When donors consider these charities, they are usually still optimistic that donating to them does yield superior impact, but they have a much harder time prioritizing between them because their central metric just remains how well they implement very similar interventions. It is well possible that the differences between these charities – charities that many impact-oriented donors are actively considering – is small enough that the value of the information would not warrant its cost.

Unfortunately, some of these donors fall into a form of analysis paralysis at this point and rather donate to the charities whose lower impact is well proven. Other donors react more rationally and donate rather arbitrarily within the group of the most highly effective charities. Others again use questionable heuristics, often aware that they are likely to be unreliable but also aware of the presumably low value of information of more thorough investigations. I aver that none of these strategies is optimal.

Competition for Exposure

The other side of the medal is that charities are aware of these dynamics. While their values may be aligned, for funding they are yet each dependent on its own pool of donors, and any cross-promotion of another charity among the first charity’s own donor base may lead to donors shifting their support to the endorsed organization. This behavior stifles cooperation.

The solution presented here will instead allow all charities in a program area to fill their funding gaps to similar degrees. If a sufficient number of donors come to accept this solution, any incentive for charities to engage in uncooperative behavior will be diminished.

Donor Coordination (working title) is a software system and strategy that fosters cooperation between value-aligned donors by allowing them do make large contributions in teams and donate to whole program areas rather than individual nonprofits. It can improve upon the current state if it is accepted and trusted by a sufficient number of donors.

High-Level Goals

The donor coordination solution can be considered successful when it achieves the following goals:

Team-level atomicity

Donors can choose portfolios with whom they are value-aligned to the point that they perceive their donations as coming from the team of donors that invests into that portfolio rather than them personally.

Program-level atomicity

1. Donors can choose charity portfolios that, as a whole, represent their moral preferences well enough that they perceive their team as donating to a program area rather than an individual charity.

2. Charities are value-aligned with the organization their donations are fungible with to the point that they make fully altruistic statements about their funding gaps.

Core Concepts

The donor coordination solution should likely take the shape of a web application to enable users of any platform to use it. The idea is roughly inspired by Wikifolio.

Visitor

An unauthenticated person viewing the website.

User

A donor, a charity, or an administrator.

Portfolio

The portfolio is an allocation rule that partitions funds among a set of charities. Every user can create portfolios, favorite or watch portfolios, and donate to portfolios.

Donor

The donor is a user other than a charity.

Charity

A charity is a user that only has the ability to enter some meta data about itself and its funding gap, and participate in discussions.

Interest Groups

The interests of beneficiaries are:

1. Beneficiaries want to maximize the available funding toward the their preferences.

2. Beneficiaries want the, at the margin, most effective interventions to receives maximal funding.

3. Beneficiaries want the funding gaps of the most effective interventions to be greater than or equal to the available funding.

In some cases the beneficiaries can give direct input, but in many cases their interests need to be represented by donors and charities because they have insufficient levels of intelligence to express them efficiently or are not yet born.

Hence the interests of donors are:

1. Donors want to maximize the available funding toward the their moral goals.

2. Donors want the, at the margin, most effective interventions realizing these moral goals to receive maximal funding.

3. Donors want the funding gaps of the most effective interventions realizing these moral goals to be greater than or equal to the available funding.

Hence the interests of the charities are:

1. Charities want to maximize the available funding toward the charity’s moral goals.

2. Charities want the, at the margin, most effective interventions realizing these moral goals to receive maximal funding.

3. Charities want the funding gaps of the most effective interventions realizing these moral goals to be greater than or equal to the available funding.

The main difference between donors and charities as two groups is the direction of the money flow. The main difference between the donors and charities internally is their different moral goal makeup.

From these primary interests follow proximate interests for value-aligned teams of donors (all donors to a program area as defined by a public portfolio):

1. Being value aligned, the members of a team are happy to make their donations fungible with the donations of all other members of the team.

2. Since their value alignment with other teams varies, there may be teams with partially opposing moral goals. Teams will want to minimize fungibility with such teams.

3. Since the funding gaps of charities are limited, teams also want to increase the funding gap of their program area by broadening its scope.

Analogously for charities:

1. The charities of popular portfolios are likely to be highly value aligned and thus happy to calculate their funding gaps cooperatively.

2. Since their value alignment with charities of other program areas varies, there may be portfolios of charities with partially opposing moral goals. Charities will want to increase their scale in order to be able to enter greater funding gaps so portfolio authors can minimize fungibility with such opposing program areas.

Clearly, the last two interests of the donor teams are in conflict. Small donation flows will favor portfolios of small, pure clusters of charities while greater donation flows will necessitate compromise in order to form greater, less pure clusters with larger funding gaps.

For simplicity I assume that all donors are perfectly informed and their only differences are differences of value alignment. This is unlikely to be the case in practice, but the only difference between a donor that is not value aligned and a donor that acts as if they were not value aligned because of lacking information is that the latter can be educated.

This educational mission is without the purview of Donor Coordination, but the software should provide the platform that donors will need to educate each other because this may be important for fostering user activity.

Functional Requirements

1. Visitors can create donor accounts.

2. Administrators can create administrator accounts.

3. Administrators can create charity accounts.

4. Visitors can view public portfolios including their descriptive statistics.

5. Donors can add public portfolios to their watch list.

6. Donors can donate to public portfolios.

7. Donors can author public portfolios.

8. Donors can draft and test portfolios in a private or draft state.

9. Donors can comment on portfolios.

10. Charities can enter new funding gaps for themselves.

11. Charities can enter new system-external donation flows.

12. Administrators have all privileges.

At some point moderator accounts will become necessary, so moderators do not need to enjoy the same level of trust as administrators to contribute to the community maintenance.

Challenges and Proposed Solutions

Descriptive Statistics

The functional requirements mention descriptive statistics. These are important for portfolio authors and other donors to decide how to structure a portfolio so not to duplicate very similar ones or which portfolio to donate to. At least two metrics are required:

1. The sum of the funding gaps of the charities in a portfolio \(P\), \(\operatorname{gap}(P) = \sum\limits_{c}^{P} \operatorname{gap}(c)\).

2. A ranked list of the portfolios with the highest fungibility but lowest similarity. One idea may be the quotient, \(\operatorname{compromise}(P, P') = \frac{\operatorname{fungibility}(P, P')}{\operatorname{similarity}(P, P')}\), of the following metrics:

   1. \(\operatorname{fungibility}(P, P') = \sum\limits_{c}^{P \cap P'} \operatorname{gap}(c)\)

   2. \(\operatorname{similarity}(P, P') = |\bigcup\limits_{c}^{P} \operatorname{donors}(c) \cap \bigcup\limits_{c}^{P'} \operatorname{donors}(c)|\).

The fungibility and similarity metrics should also be displayed in isolation, particularly as a guide for authors of portfolios of new charities when the combined compromise metric is undefined.

It may also become necessary to take weights into account, and the formulas will surely need to be tweaked further once real data become available.

Funding Gaps

There needs to be a common definition of a funding gap, so that charities have hard, unyielding guidelines as to what figure to enter for a given year.

Prioritization organizations already face a similar problem: Imagine two charities, charity A with the ability to invest $100 million with some baseline effectiveness \(e\) on average and charity B with the ability to invest $10 million with an average effectiveness of \(10e\) within a given year. Further assume that the charities are value aligned to simplify the problem to one dimension of impact.

A commonly used uncertainty discount is 3% p.a. and for simplicity we assume that suffering in the world, absent the interventions, remains constant, so that aggregate suffering increases linearly over time.

A donor that wants to invest $100 million now has the choice to donate it to charity A, knowing that it will be invested in the same year, or to charity B, knowing that $10 million of it will be invested in the same year, $90 million of it will wait on the charity’s bank account at an interest rate of maybe 1% for another year, $80 million plus interest will wait for two years, and so on.

Clearly, a definition of funding gaps that only takes into account a charity’s ability to invest some amount per year would set very different bars for the marginal impact of the last dollar of that funding gap.

Since the donor coordination solution addresses coordination problems that arise when the funding gaps of the individual charities do not warrant the attention of a prioritization organization, we assume that there are no meaningful differences of their relative effectiveness, so we face a simpler version of this problem.

One solution may be to adopt GiveWell’s excess assets policy: “We seek to be in a financial position such that our cash flow projections show us having 12 months’ worth of unrestricted assets in each of the next 12 months.”

Allocation

Another open question is the allocation of donations within the portfolio. Conceptually, donors donate to program areas, but factually they will have to transfer their donation to a specific organization. Splitting it up across several organizations would be an unnecessary hassle of the donor, so the algorithm that suggests the specific organization should know some ideal allocation and then recommend a recipient organization such that the actual allocation comes closest to the ideal allocation. It could also take tax deductibility into account as a tie breaker.

The simplest option might be an equitable allocation where the algorithm aims to assign the same level of funding to each charity after taking donations external to the system into account.

Another option may be to prioritize small funding gaps as an additional incentive for charities not to exaggerate their funding gaps in the moral lies scenario. However that would have little effect since the charities in a given program area are value aligned and can thus easily conspire with each other, and it may have the detrimental effect that charities would be incentivized to be tardy with entering new funding gaps.

Donation Swaps

Donors often agree on donation swaps where each partner donates to the charity of choice of the other partner in order to harness the tax deduction of the charity in the respective country.

In order to help portfolio authors to trade off fungibility against funding gaps, there would need to be a ranking of other portfolios that the given portfolio is most fungible with. However, portfolios whose audience is very similar are least interesting to portfolio authors, so the ranking should be sorted by something like the fungibility per cardinality of the cut set of donors, and here donation swaps would add noise to the calculation.

It needs to be either clear to the donors that they need to enter the donation of their swap partner as their donation or the software should allow them to mark donations as swaps and enter their partner. The first is probably the better solution for an MVP, but the second may be more foolproof.

Remaining Challenges

Moral Lies

When there is a pair of program areas such that the teams of each see the team of the other as an opposing team, but there is some set of charities that they can agree on, and the available funding is close to or greater than the available funding gaps of their program areas without the consensus charities, the intended result is that donors compromise and add charities to their portfolios that increase the funding gap at the cost of greater fungibility.

But charities are of course value aligned with these teams. Hence it will be ethical for them to lie about their funding gaps, inflating them, to drive the opposing donors to fund the more fungible funding gaps. Analogously, the opposing team’s charities can also inflate their funding gaps; they even have to lest their cause suffer. When one group defects in such a fashion, the cooperation breaks down. A classical example of the prisoner’s dilemma.

In practice, the donor coordination solution will be used mostly or at least at first only by donors that are all fairly value aligned at least to the extend that they value the type of moral plurality that exists among them. Hence this problem may not manifest any time soon.

Market Research

My experience that such a software would be helpful is based on reports of friends, some of whom are donors and some employees of affected charities. Unless, however, there is a sizable number of prospective users that are interested in the project, charities will not have sufficient faith into the growth of the user base to warrant their time investments.

Apart from surveys among likely prospective users, one central market research tool needs to be a minimal viable product (MVP). Other donors and nonprofit staff have considered opening a group on a social network such as Facebook to bring together all participants in whose actions need coordination. The group would provide a means for communication but would leave any functions beyond that to the participants to be implemented in a manual, ad-hoc fashion. This way it will become clear which processes are in most urgent need of automation. It will also become clear if the community is large enough to sustain a more comprehensive solution like the one proposed here.

Community Building

An important strategic and marketing problem is the following: Entering funding gaps will only warrant the effort for the charities if they can expect significant donation flows from the donor coordination solution. For donors the donor coordination solution is only interesting when the program areas they want to donate to are well represented by charities working on them.

One solution may be for administrators to regularly poll information on funding gaps from charities and invite them to claim their accounts themselves. That way, the administrators will have added effort during the startup phase, which will be increasingly outsourced to the charities as donors come to accept the system.

To achieve said donor acceptance, it would be helpful if the project were run by a reputable organization with considerable reach, and if the project collected early signups prior to its launch, both in order for it to launch with momentum. Until such an organization has been found, I cannot consider this challenge solved.