For the past several years, the EA community has run donor lotteries, in which the allocator of a pool of donations is decided quasi-randomly, with additional weight towards those who have contributed the most. This post reframes that system as ‘donation-weighted quasi-random donor pooling’ and proposes alternative systems, which might have some advantages:
- Alternative 1: Voting donor pooling
- Alternative 2: Random donor pooling
- Alternative 3: Threshold random donor pooling
- Alternative 4: Reverse-donation-weighted
- Alternative 5: Demographic-weighted
- Alternative 6: Location-weighted
- Alternative 7: Cause area donor pooling
(Note I am not familiar with the laws around donor lotteries in e.g. the US or UK, so am unsure which of these alternatives would be legal/practical – these are theoretical alternatives.)
In donor pooling, instead of each donor allocating their donation individually, donors ‘pool’ their donation, then adopt some decision procedure to determine who will ‘allocate’ (distribute, grant, disperse, etc) the pool.
The person picked to allocate the funds is commonly described as the ‘winner’. I will instead use the more neutral term ‘allocator’. There certainly are advantages to being selected the allocator, mainly the opportunity to allocate more funds to your priorities than you would otherwise have been able to do so. However, there are also disadvantages, such as taking on responsibility for your group, and the time cost those who have become allocators have typically spent.
Donor pooling has several advantages. First, it saves everyone’s time. There are also gains from specialisation – 1 allocator spending 50 hours researching the best opportunity will likely produce better results than 50 donors spending 1 hour. Second, there are opportunities that are only available to an allocator with a large pool. Charities are more willing to provide information and spend time on discussions.
One way to donor pool is in a pool with a predetermined allocator(s). In the EA community, the EA Funds (amongst others) play this role. One can choose to pool with an allocator based on their track record or philosophy of allocation. Allocators can specialise and build up experience, perhaps making them better allocators. Those well-suited to make money and donate might not be the same people as those well-suited to allocate. However, the allocator need not be predetermined.
Alternative 1: Voting donor pooling
Donors vote on who should be selected allocator. Those eligible to be selected could be restricted to pool donors. Eligibility could be broader, either from a list of people who have indicated willingness to be an allocator, or asking someone to be an allocator ‘cold’ – e.g. without indicating willingness beforehand. A variety of voting procedures are possible: consensus, majority, or plurality; single vote, single transferable vote, ranked voting, or approval voting; with or without weighting.
The ‘donor lottery’ differs from funds or voting by incorporating more randomness. One is less sure who will be selected as allocator. More randomness in allocators is likely to mean more randomness in what gets funded.
Randomness has several advantages. First, it can expand the search strategy for good donation targets. There is an ‘explore/exploit’ trade-off in donations. Exploit gives to options we know are good. Explore tries new options, which may prove to be better than existing ones but risk being worse. Randomness moves us more towards ‘explore’, expanding the search space. This is especially useful if we have not explored the space fully and believe there is high ‘value of information’.
Second, randomness is valuable to the extent that it is difficult to predict which charitable interventions will ‘pay off’. In such a situation, it could make sense to spread one’s bets as widely as possible, rather than clustering on options we (perhaps mistakenly) think have higher probability of good outcomes. This is similar to OpenPhil's Hits-based Giving.
Third, randomness can reduce groupthink. Small groups can have biases in what they fund, due to having similar information available, or a bias towards the familiar. If they make mistakes and fund something inefficient, it can take longer for them to correct.
Fourth, groupthink may also be tied to elitism, which incorporating randomness can therefore help address. Typical allocators might not just give to the same, small group of charities – they might do so because that group is better connected. Rather than pooling with a typical allocator who might allocate to groups they already know or are well-represented, randomness can get around that form of bias, and grant opportunities to newer, smaller, or less-well connected charities.
Fifth, randomness could affect the recruitment of allocators e.g. at EA funds or foundations. At the moment, these allocators are recruited through hiring rounds or informal routes (e.g. interpersonal networks), both of which might privilege some allocators, and therefore some views. Being randomly selected could give someone a chance to ‘prove themselves’, which could result in better donation decisions for the ultimate beneficiaries.
The purest form of randomness would be:
Alternative 2: Random donor pooling
Donors contribute to a donor pool (the pool could either have a cap or not). The person who will allocate that donor pool is chosen at random from every donor.
This suffers the major disadvantage of ‘bad actors’ contributing a trivial amount ($1) for the chance of being chosen as the allocator. Instead, we might consider:
Alternative 3: Threshold random donor pooling
This introduces a threshold. This could be in the donation – one has to donate e.g. a minimum of $100 to even enter the pool. It could instead be in the selection – one has to donate a minimum of $500 to be eligible to be selected as the allocator. The second is slightly preferable, as someone might want to contribute but not be eligible for selection.
One can move from a random process to a quasi-random process by giving additional weight (in the probability of selection) to various factors.
The current ‘donation-weighted’ system gives more likelihood of being selected to those who have contributed the most. The analogy is to lotteries, where if one buys more tickets one has a (slightly) larger chance of ‘winning’. This may incentivise donors to contribute more to the pool. It might also seem ‘fairer’ in some sense to the donor. Finally it has some nice implications for expected value calculations (where e.g. allocating $1,000 directly is equivalent to pooling and having a 10% likelihood of allocating $10,000).
However, it also has disadvantages. It reflects pre-existing income and wealth (and therefore power) inequalities. One might find that inherently uncomfortable, however one should also note that this system removes many of the advantages of randomness. Randomness expands the search space by increasing the chances of people not typically selected to allocate large pots of money; conversely, donation-weighting shrinks the search space by decreasing the chances of new allocators. The views of the rich on which is the most effective place to donate to are well-represented in overall donation. Donation-weighting moves us away from an ‘explore’ strategy. If people are already rich and donating considerably, they may already be susceptible to groupthink, perhaps in a way that reinforces elitism. Finally, donation-weighting reduces the recruitment potential of a randomly selected allocator ‘proving themselves’.
There are several alternatives to donation-weighting.
Alternative 4: Reverse-donation-weighted
This reverse-weights donations. This system would give more likelihood of being selected to those who have contributed the least, above some threshold to prevent free riding. A donor’s incentive to give more than the threshold would be due to the gains that donor felt would come from pooling and randomisation. The advantage of this system is to expand the search space by increasing the chances of those who do not typically allocate large sums, increasing the range of views represented in allocation decisions.
Weighting with more information
Asking for more information from donors in the pool allows one to weight in favour of characteristics that are underrepresented in donor circles, or that one wants more representation of. Again, I am not familiar with the laws around donor lotteries in e.g. the US or UK, so am unsure which of these alternatives would be legal/practical – they are theoretical alternatives.
Alternative 5: Demographic-weighted
This allocates additional weight to various demographics of the donors, such as gender, sexuality or ethnic identity. In the US or UK, some demographics are overrepresented in the allocation of the overall ‘pot’ of donations formed by all charitable giving. The demographics of allocators are skewed whiter, male and straight – reflecting overall distribution of resources and power. This is also true (to some extent) of the EA community, and possibly of the donor pools of the ‘donor lottery’.
One might be concerned that this is shrinking the search space and/or inefficiently biasing the allocation of donations, to the detriment of the ultimate beneficiaries. One could therefore weight towards underrepresented demographics, to increase the chance of donors with those characteristics being selected. Like Reverse-donation-weighted, the incentive to give for a donor whose demographic characteristics were not weighted towards would be the gains that donor felt would come from pooling, randomisation, and underrepresented voices.
Alternative 6: Location-weighted
This weights donors’ locations, with some locations increasing the chance of being selected. This could be weighted towards donors located in ‘EA hubs’ (such as the Bay Area or Oxford) as donors in those locations might be aware of non-public donation opportunities. More likely, it could be weighted away from those locations, due to concerns about group-think, elitism or expanding the search space. To extend this second alternative, it could be weighted towards donors in the Global South, as this could bring perspectives into allocation that are currently less well-represented in the EA community.
Finally, the current system does not restrict based on cause-area. This means a donor pool in which, for example, 90% of the donors (or donations) intend to support one cause-area, might face an outcome where an allocator is chosen from the other 10% with a different cause-area. In a simple risk-neutral expected-value calculation, this should not matter (10% of 10,000 = 1,000). But it might matter if the donors were cautious (risk-sensitive) or impatient (time-sensitive - e.g. if there was a particular window open now that would not be open in a year or two) – or simply not all perfect rational maximising Homo economicus.
Alternative 7: Cause area donor pooling
Some quasi-random allocator-selection, but pools have designated cause-areas. For example, a longtermist pool, an animal welfare pool, a global development pool. Donors know that the pool they are contributing to will be spent in the cause area they wish to support.
This post is intended to reframe donor lotteries and prompt discussion about alternatives to the current model. It is not intended to advocate for any particular alternative. If I had to choose, my personal preference would probably be threshold random donor pooling in a cause-area, to increase the benefits of randomness while keeping the assurance of allocation within a cause area. I am skeptical that the incentive effects of donation-weighting are worth the costs in reducing randomisation. If one forgoes many of the benefits of randomisation, why not just donate to a pool with a predetermined allocator?