For this post, GiveWell staff members wrote up the thinking behind their personal donations for the year. We made a similar post a year ago.
|Organization||GiveWell allocation||Our allocation|
I agree with the relative strengths and weaknesses of each organization as summarized in this table but I differ from the reasoning behind the GiveWell allocation in two ways:
- I put less weight on explicit cost-effectiveness analysis. Conceptually, my goal is to maximize the impact of my charitable contribution, but I put less weight on the explicit cost-effectiveness models we create than others. Those models are highly dependent on hard-to-estimate inputs and often include many assumptions that we were forced to make based on very limited information. The models are available here; we plan to more clearly and explicitly lay out the inputs about which we are least certain and the ones that are most likely to materially affect our the bottom line in a post in the near future.
- I put more weight on organizational strength. GiveDirectly stands out from our the other top charities on three factors listed in our summary table: transparency and communication, ongoing monitoring and likelihood of detecting future problems, and organizational track record. As a general observation, I have consistently found it more difficult than I expected to accomplish what initially appeared to be relatively easy goals (personal or organizational). This has reinforced my intuition that excellent organizations are significantly more likely to succeed than others, and GiveDirectly’s overall excellence increases my estimate of the impact it will achieve.
I’m not donating to GiveWell, because I don’t find it appropriate to donate to my employer, for similar reasons to what Holden wrote last year. Absent that consideration, I would consider direct support of GiveWell to be the best giving opportunity.
In last year’s post, I laid out my thoughts on giving now vs. giving later, and why I find it important to give something now (and to give it somewhere besides GiveWell itself). I don’t have much to add to that. I agree with other staff members that this year’s giving opportunities are stronger than last year’s, but in my mind the difference isn’t big enough to cause me to deviate from my regular giving, which I generally determine as a percentage of my income.
I expect each of our top charities to receive significant funding this year, so considerations of “upside” and of helping an organization grow aren’t as salient to me as they’ve often been in the past. With that in mind, and with fairly significant (though highly uncertain) estimated differences in cost-effectiveness, I’m fairly comfortable privileging health interventions over cash transfers, up to the point where room for more funding becomes a major issue.
I believe there are strong reasons to give this year, rather than waiting for future opportunities. These include:
- GiveWell’s top charities this year offer strong opportunities to do good. As a point of comparison, donating to GiveDirectly this year is probably at least as strong an opportunity as it was last year when I allocated all of my giving to GiveDirectly, and this year I plan to allocate only a small portion to GiveDirectly because of opportunities to accomplish as much or more elsewhere.
- I do not feel compelled to wait for Open Philanthropy Project recommendations. I put more weight than some of my fellow GiveWell staff members on high likelihood of impact, in part to help emotionally motivate myself to give in the future. Also, given the size and complexity of the task that the Open Philanthropy Project has taken on, it could be several years before there are opportunities available to small donors that the project believes are as strong as GiveWell’s recommendations.
Where should I give?
I plan to follow GiveWell’s recommended allocation this year by giving a gift to GiveWell for regranting to top charities at its discretion. Allocating funding between charities involves complicated trade-offs on many different dimensions: program cost-effectiveness, confidence in the organization’s ability to carry out the program, confidence in the organization’s ability to spend funds in the near term, judgements about the upside and risks of helping an organization scale up, etc. I don’t feel fully confident in my own ability to weigh these factors and fully understand what we know about each factor for each charity, but I feel confident in the process that we used to create the overall allocation: detailed analysis for each charity, write-ups that are reviewed at multiple levels (from high-level reviews by a Co-Executive Director to careful checks of each footnote citation), and several rounds of staff debate. Relying on this allocation helps counteract potential personal biases from spending much more time thinking about AMF and SCI (as part of contributing to that research) than about other top charities and standouts.
I have some concerns about our allocation, but ultimately I don’t think they are important enough to outweigh the above benefits. My main concerns are:
- Are we making too small a bet on AMF? AMF will receive most of my contribution, but in some ways, the bet GiveWell is making on AMF is conservative. Among our top charities, AMF is working on the program that I believe is the most cost-effective. It is extremely transparent and has strong processes for monitoring net distributions. It has now allocated most of the ~$10 million that it received due to GiveWell’s recommendation two years ago. I think it’s possible that we should be making a stronger bet now, perhaps advocating that AMF gets ~$20 million (twice what they got before) rather than $10-15 million, in order to see what AMF can accomplish at that level in the next two years. Because it can take AMF a year or more to finalize distributions, caution now limits our ability to learn about AMF’s capacity to move funds over the next two years. However I have not decided to allocate a larger percentage of my giving to AMF, because (a) total funding available is limited and I believe the other opportunities that are available are stronger than the ‘big bet on AMF’ option, and (b) there will be opportunities over the next 6 months to a year to learn about AMF’s progress and to reevaluate AMF’s room for more funding.
- Will SCI be able to productively spend as much as we think they will? Our estimate of SCI’s room for more funding is more speculative than that for other top charities, primarily because of our lack of understanding about how SCI has spent funds in the past and our ongoing challenges communicating with SCI about its future plans. I’m concerned that SCI will not be able to use funds to scale up deworming programs as quickly as it thinks, because of non-monetary barriers (such as lack of mapping data for new country programs, limited in-country capacity), and that as a result it may use funds for less beneficial purposes (such as increasing budgets without increasing the number of treatments, funding treatments in low prevalence areas). Having worked on GiveWell’s investigations of SCI for several years, I may have instincts on this issue that other staff don’t have; on the other hand, my close involvement may bias me toward weighing communication difficulties too strongly. In the end, I think the room for more funding we have estimated for SCI is a reasonable guess based on what SCI has told us and its strong track record of starting and scaling up deworming programs.
- Are we discounting a one-time opportunity to learn about the effects of cash transfers? GiveDirectly may use funds it receives in December 2014 to increase the sample size of its study of general equilibrium effects of cash transfers, which could increase the benefits of giving to GiveDirectly this year, since transfers would not only benefit families directly but would also improve our understanding of the impact of cash transfers. It’s not clear how likely it is that additional donations from individuals would increase the sample size – the ideal sample size according to GiveDirectly would require $15 million in cash transfers and GiveDirectly has raised a portion of this already and may be able to raise the remainder from other sources.
- Is our allocation to Deworm the World Initiative too high? The case for DtWI’s room for more funding in the next year seems weaker to me than that of other top charities. But, ultimately I’m OK with GiveWell’s recommendation, because the amount that GiveWell is recommending is modest and the potential upside of supporting a strong organization working on a cost-effective program is high enough to justify the investment.
However, I am planning three small deviations from straightforward adherence to our recommended split:
- I’m planning to give GiveWell itself 10% of my gift. I think GiveWell is a relatively mature research product at this point, one that I value greatly, and I think it makes sense for individuals who value the research to give part of their gift to pay for it.This is a bit of an evolution for me, because I’ve previously written that I disagreed with Holden and Elie about the potential risks from being too dependent on Good Ventures. I still don’t think that it would be a particular problem for large donors who rely on our research, especially Good Ventures, to cover the bulk of our operating costs (since they also account for the bulk of our money moved).The main reason I decided to give to us is non-consequentialist: I don’t like the dynamic of us asking donors (other than Good Ventures) who are considering giving a large amount of money based on our research to direct it to us instead of our top charities. We do this, rather than making the rounds of all our donors to ask them to give ~10% of their gifts to GiveWell, because it is much more time-effective. I think that’s the correct decision on our part between the two choices, but it would be better still for us not to have to because our donors had adopted a norm of giving a portion of their gift to cover GiveWell’s operating expenses.10% is a relatively arbitrary amount, and it’s significantly more than our top charities recommendations cost to deliver. Our work on top charities takes up approximately half of GiveWell’s overall budget, with the other half directed to the Open Philanthropy Project. Our budget for the next 12 months is about $2 million, so the actual cost of delivering our top charity recommendations on a forward-going basis looks like ~5% of money moved. Since some people are unlikely to pitch in, I think it makes sense for those who are willing to do so to err slightly on the higher side.Because money donated to our operating budget is largely fungible with money donated to top charities, I see this as a relatively costless decision.
- I’m not planning to give to Evidence Action’s Deworm the World Initiative. Though I think they’re an outstanding organization, I think they’re likely to raise enough from other donors to meet their funding needs for this year. Additionally, the marginal projects they’ve described strike me as quite different than simply delivering more deworming (though they still strike me as good). If donors don’t fill their two-year gap this year, I think we’ll know about and be able to address it next year.
- These previous two changes would result in my gift being 64% AMF, 13% SCI, 13% GiveDirectly, and 10% GiveWell. Although I think our overall funding target for AMF is approximately correct (I wouldn’t want to see them double it), I think it’s probably a better giving opportunity, all things considered, than our other top charities, and I don’t think our funding target is sufficiently precisely estimated that I should defer strongly to it at the level of a few percent. Accordingly, I’m planning to round my donation to: 70% AMF and 10% each to SCI, GiveDirectly, and GiveWell.
The decision to deviate slightly from our recommendations (which I had not been initially planning to do except for the 10% to GiveWell) was partly informed by reading other staff member’s contributions to this post and realizing that many of us were inclined to defer to GiveWell’s recommended split. I’m deviating partly to try to communicate that I don’t think total deference is the appropriate stance: these are hard decisions and disagreements should be expected. I don’t think donors who use our research should feel that there is a very high bar for deviation from our recommended allocation.
- Donating to GiveWell for re-granting to top charities seems strictly better than donating according to the recommended allocation, because it leaves GiveWell greater discretion in allocating funds in the event that we move more money to top charities than predicted. (Recommended allocations were determined based on an estimate of expected money moved, and I think this estimate was conservative).
- I expect there would be significantly diminishing marginal returns for funds used this year beyond the maximum for each charity. Underlying this intuition are concerns about the challenges of quickly expanding staff capacity, finding additional locations with comparably high disease prevalence and implementation capacity, etc. I therefore want GiveWell to have discretion in how to allocate funds beyond the maximums.
I am giving 30% of my donation to GiveDirectly, specifically.
- In addition to the positive factors that others have stated, I see GiveDirectly’s commitment to research as one of its most notable strengths relative to the other top charities. Its work in this area seems potentially very high leverage, given that it is aimed at influencing the policies and portfolios of major funders. For example, GiveDirectly’s study on macroeconomic effects was designed to address a gap in the evidence base for cash transfers that policy makers cited as limiting their ability to implement cash programs. GiveDirectly is also pursuing projects in partnership with aid agencies to “benchmark” traditional aid programs against cash transfers.
- This year, GiveWell’s recommendations did not put much weight on “upside” factors like an organization’s potential to influence other funders and programs. In GiveDirectly’s case, this factor is highly speculative, because the work is relatively new, so there is a very limited track record. However, I feel we have seen some positive signals already and I expect we will report more on this work in the coming year.
- I want to support GiveDirectly so that it has greater flexibility to pursue research and partnerships, both from the additional funding and the time not spent on fundraising.
I am giving 10% of my donation to GiveWell’s Standout charities.
- As a GiveWell donor and employee, I want to reward our Standout charities for their transparency and commitment to generating and/or acting on rigorous evidence. I also want to increase the incentives among Standouts (and all charities) for continued openness and participation in GiveWell’s process. I think there are relatively high learning benefits for GiveWell in continuing to assess these charities, which seem unique to me in ways that depart from a strict adherence to our criteria.
- GiveWell’s Standout charities seem to be within what I consider to be a close range of the estimated cost-effectiveness of our top charities. Though I have lower confidence in the impacts of Standouts, there are various “upside” factors which I can imagine contributing to their true long-run impacts. These factors include institutional capacity building, increasing competition in health products and services, and causing attitudinal and behavior changes in health that may be self-perpetuating over time. (Some of these factors may apply to top charities as well.) While “upside” is extremely difficult to rigorously assess, especially on short time horizons, it still matters in modeling a charity’s true impacts. I have tried to keep these kinds of considerations in mind without putting too much weight on them, because they are highly speculative.
Finally, I am saving 10% of my giving until I have more time to look into opportunities that are not GiveWell-recommended. If I do not allocate these funds elsewhere, I will give them mid-year after GiveWell has refreshed our room for more funding estimates for recommended charities.
As with last year, I plan to give money from outside my charitable budget for membership in civic organizations, helping out my neighbors, supporting family members’ and friends’ projects, and contributing to my religious community.
I see GiveWell’s allocation as our organization’s best attempt at approximating some combination of “where GiveWell would give if we controlled all donors’ contributions”, “where GiveWell thinks our audience would want to give if they had all the information we have”, and “where GiveWell thinks people should give if they want to follow GiveWell’s systematic (and notably skeptical) approach to evaluating charities”, all of which are generally very similar and sometimes hard to distinguish.
As such, and because I believe that any attempt at ideal allocation will be imprecise, I would expect personal giving allocation of a GiveWell staff member to vary somewhat from the official GiveWell allocation. Specifically, a staffer might have different intuitions or values than the ones that are best represented in the GiveWell allocation, or might even believe that the GiveWell allocation should have been something different. Additionally, a staffer (or any donor to GiveWell’s recommendations) could also try to anticipate other GiveWell donor’s gifts and attempt to give in reaction to those contributions, rather than attempting to give to each organization in proportion to their total ideal breakdown. (See our recent blog post for an elaboration of this dilemma.) I don’t believe GiveWell staff or donors should do this; instead I think it is better for us to give as though other GiveWell donors with broadly similar goals are giving the same way. (See this discussion of Newcomb’s problem for an illustration of one of the primary reasons I believe this.)
My planned allocation
I support GiveWell’s allocation and plan to allocate the bulk of my giving accordingly. My personal allocation for altruism-motivated giving has just two differences:
- For similar reasons to Alexander (above), I think it is appropriate for all donors that use GiveWell’s work to allocate 5-10% of their GiveWell-directed giving to GiveWell (unrestricted, for GiveWell operations). If nothing else, this serves as a recognition that the funds required to run GiveWell become somewhat fungible with the funds directed to the charities we recommend, since we turn to our biggest supporters for unrestricted support when we need it (who often reduce the amount they give to our recommended charities by the amount they give to us).
- I believe there is a significant chance that our cost-effectiveness analysis (CEA) of DMI significantly underestimates their cost-effectiveness because it does not account for one possibly important factor: the fact that DMI and its funders decided to invest several million dollars in an RCT of DMI’s impact, hoping and expecting to find a statistically significant effect. I believe it was appropriate to leave this consideration out of GiveWell’s CEA, but I am also personally willing to put weight on this factor. (For more details, see a forthcoming blog post on this consideration.) As a result, DMI looks more competitive with our top charities and I would like to see them get ~5% of GiveWell directed giving.
I feel confident giving to GiveWell’s recommended allocation for reasons similar to those laid out in Natalie’s post. The charities that I was most uncertain about donating to were DtWI (because of uncertainty about room for more funding) and SCI (because of communication difficulties and uncertainty about how past funds have been used). However, my concerns about donating to both of these organizations were ultimately outweighed primarily by the very high estimated cost-effectiveness of deworming programs.
Last year, I saved about half of my charitable giving budget because I thought it might have been a year in which the top charities had unusually low estimated cost-effectiveness. (It now seems that this was likely true for the reasons that Natalie alluded to in her “Should I wait to give?” section.) This year, I plan to give the funds that I saved from last year and my full charitable giving budget for this year.
Giving to GiveWell’s discretionary fund seems to be the easiest way to ensure that my donation will go to the charities where GiveWell thinks additional funds will have the most impact, regardless of how much total money is donated to top charities this year. We plan to post more details about how this fund will allocate resources if one of our recommended charities surpasses its “maximum target for individuals” (as laid out here) in an upcoming blog post. As far as I can tell, there is no downside to giving to GiveWell’s discretionary fund relative to following the allocation through other methods, so I would recommend that donors with similar goals to mine also give to the discretionary fund.
I’m giving the rest to AMF because I think it is significantly more cost-effective than GiveDirectly. I believe both AMF and SCI have room for more funding (RFMF) beyond the targets we set for them, whereas I don’t believe DtWI has RFMF to scale up deworming programs. (We think AMF could likely use substantially more than our target amount, while DtWI told us it had limited RFMF for deworming scale-up, and I’m also concerned about fungibility with other Evidence Action programs. I’m drawing this analysis mostly from information presented in our top charities blog post.)
I believe that SCI is less cost-effective than most other staff who published cost-effectiveness analyses (CEAs). I think SCI is about 3 times as cost-effective as GiveDirectly, most other staff think roughly 5-10 times. Also, compared to what most other staff wrote in their CEAs, I believe that the income gains from deworming are worth more relative to saving lives via bednets (I would trade a 25% income gain for 8 people for saving one life, whereas most staff would trade a 25% gain for about 40 people for saving one life). The former disagreement favors AMF, while the latter favors SCI. I also consider that AMF seems more competent and transparent than SCI, which is not captured in the CEA. When I net all of this out (with the help of some informal calculations on top of my published CEA), AMF seems to come out slightly ahead.
I would like to see SCI get a decent share of donations to top charities this year, and I might also like to see DtWI get a share as well (though I haven’t thought about it carefully). However, unlike many other staff, I don’t feel beholden to give to these charities in the proportions that I’d like to see individual donors collectively give. Instead, of these three charities, I’m giving just to AMF, where I believe my marginal dollar will do the most good. I believe that AMF is a better giving opportunity this year than GiveDirectly was last year, so I feel vindicated in my decision last year to hold some of my giving budget for this year.
This year I made my annual donation as well as donating a portion of last year’s gift that I saved for this year because I expected an improvement in the available giving options. I split my giving between an unrestricted donation to GiveWell and a donation to the Against Malaria Foundation. I chose the Against Malaria Foundation largely for reasons described by Ben above. I value cost effectiveness calculations relatively highly and therefore prefer bed nets to cash. I am more sure of the quality of AMF’s implementation than SCI’s. I donated unrestricted to GiveWell for reasons described in last year’s staff giving post.
In making this decision, I read our post on the top charities, walked myself through the logic of the recommended allocation, and had a few conversations with my coworkers about their giving plans. I don’t have any strong disagreements.
Ideally, the effective altruist community would be large enough to close the funding gaps of all our top charities. Then we wouldn’t have to make all these tough trade-offs. Based on GiveWell’s projections, it looks like GiveDirectly will still have a substantial amount of room for more funding at the end of this giving season. Though I don’t see this gap as extraordinary, I’m hoping that it still motivates casual donors to give more than we project them to.
Last year, I saved about half of my charitable giving budget, because I thought my first year working at GiveWell might influence my choice of where to give more than other years and I was optimistic that GiveWell would find better giving opportunities in the near future. This year, I plan to give the portion of my charitable giving budget from last year that I saved and my full charitable giving budget for this year.
Recently, many staff discussions on personal giving have focused on whether or not a donor who agrees with GiveWell’s targets should give according to GiveWell’s recommended allocation. We have discussed how our positions as GiveWell employees may impact the answer to this question, because (1) our posts here may end up influencing other donors, and (2) to the extent that we have personal disagreements with GiveWell’s allocation, we should be asking whether we believe that represents deeper flaws in the process used to determine the allocation.
Currently, I think I should give in the way that I would like other GiveWell donors to give, even if I am not sure that fellow donors will behave the same way. I am not confident in this opinion, and hope that further discussions will clarify my views.
Because I am a relatively new employee at GiveWell, I have limited experience thinking about the trade-offs between our charities. I largely agree with what Natalie wrote above about trusting in the process that is used to arrive at this allocation, and how this process can prevent us from personal biases (or ignorance). I also do not think that my personal values differ greatly from those of GiveWell–values which are reflected in its recommended allocation.
Natalie’s questions above about our allocation cover the main concerns I have: Should we be directing more funds to AMF? Are DtWI’s proposed uses of GiveWell donations cost-effective enough to be worth funding? I also wonder whether we are under-weighing the benefits of GiveDirectly’s possible influence on other charities. Two of our standout charities this year ran RCTs on their own programs. If GiveDirectly’s success even partially causes other organizations to run their own trials (or be more transparent, or improve monitoring), it may be adding more value than we currently account for. Because I am highly uncertain that adjusting for any of these concerns (mine or Natalie’s) beyond the degree to which they are already considered would improve GiveWell’s targets, I am sticking to the current allocation.
I have divided my giving this year into three categories:
- Donations as gifts
For the holidays this year, I am using $50 and $100 donations to GiveDirectly as gifts to my family and friends. In some cases, I am giving material gifts in addition to donation-gifts. In aggregate, these donation-gifts make up a substantial portion of my overall giving for the year. GiveDirectly is my charity of choice for donation-gifts because its model is easy to explain and sidesteps many common concerns about charitable giving. I hope that these donation-gifts encourage recipients to think more about effective giving.
- Donations to services
I donate some money to services that I use regularly, such as the Wikimedia Foundation, which runs Wikipedia. I conceive of this giving as payment for services rendered, rather than as an altruistic donation.
- Core giving
As a member of Giving What We Can, I have pledged 10% of my earned income to effective charities. This money is the core of my charitable giving (the donation-gifts mentioned above are not included in the 10%). I am following GiveWell’s suggested allocation with my donation this year, with the exception of donating to ICCIDD and reducing my donation to SCI. I have reservations about SCI’s organizational structure. These reservations, coupled with questions about the evidence base for deworming, weaken the case for SCI in my view. ICCIDD is compelling to me because of the dedication of its staff, the potential impact of salt iodization, and its pressing need for further funding. My cost-effectiveness outputs consistently place ICCIDD’s program within comparable range of SCI’s. In general, I feel confident in all of the organizations we are recommending this year, and would happily give to any of them over a charity we have not investigated.
The giver’s dilemma as a GiveWell employee
As recently noted on this blog, there are multiple issues surrounding donor coordination, mostly regarding the fungibility of donor money and the preservation of donor autonomy. I see two general models for donor behavior:
- Giving as if you controlled the entire pot – In this model, donors that subscribe to GiveWell’s value system follow the GiveWell-suggested allocation, without using timing or information gains to their advantage (e.g. donors do not delay their donations in hope of receiving more up-to-date information about the funding situations of charities in question).
- Giving as a market – In this model, donors that subscribe to GiveWell’s value system attempt to make the best buy they can make on the margin, taking into account timing and information differences between donors (e.g. donors delay donating in order to receive updates on the funding situations of charities then give accordingly, making the donations of previous donors fungible).
I have not come to a resolution on the giver’s dilemma, and I see merit in both of the above models. This issue is especially salient to me as a GiveWell employee, as I likely have information advantages relative to a generic donor, which would make the second model more attractive.
However, under either of the above models, it makes sense for me to follow GiveWell’s recommended allocation, except where I have substantive disagreement with it (e.g. my decision to give to ICCIDD and reduce my donation to SCI). Under the first model, following the allocation is the default position. Under the second, I occupy a unique market position as a member of GiveWell. Arguably, this unique position enables me to more effectively allocate my personal giving (by understanding organizational room for more funding and donor intent with greater granularity than the average donor). However, the largest impact of my giving may not be the immediate effect of my dollars, but the signal that my allocation sends. Giving in a radically different allocation to GiveWell’s recommendation signals either:
- a lack of confidence in our process
- a difference in values or intuition
- a desire to use a unique position to optimize personal giving, rather than using this position to signal a vote of confidence to other donors
I believe that GiveWell’s process is sound and has produced recommendations worth following. Further, my values and intuition largely align with GiveWell’s recommended allocation, with the exception of SCI. Finally, I believe that the long-term impact of sending a relatively consistent signal as an organization is greater than that of optimizing my personal giving for the year. For these reasons, I am following GiveWell’s recommended allocation closely.
A final caveat is that I am new to working at GiveWell, and new to thinking seriously about charitable giving. I have a low confidence in my understanding of the organizations we consider relative to my co-workers, and I expect my understanding to develop substantially over time. In the meantime, I am reluctant to anchor myself to strongly mainstream or contrarian positions, and feel that I have struck a reasonable balance in this regard with my giving decisions this year.
By and large, I agree with the rationale for this allocation explained in the blog post. In some cases, I have different intuitions on pieces of our analysis, but I believe that the process we followed is significantly more likely to lead to a good recommendation than my personal intuitions (I agree with Natalie, who elaborates on this; her examples of personal intuitions are reasonably representative of my own). I truly believe that donors who are “looking to give as we would” should follow this allocation; I think it’s an important signal to GiveWell followers to reinforce this confidence via my personal giving decision.
To some extent, other staff members disagree, and I hope to resolve disagreements like these in the future. However, putting this into context, I think most staff disagreement is on the optimal allocation of donations across our top charities rather than the list of top charities. I believe that donating to our top charities at all (rather than other charities) is a significantly more impactful decision. I personally (and I believe GiveWell as an organization) care more about getting the list of top charities right than the optimal allocation. While I’d ideally prefer all of the staff reconcile their independent views until we come to an agreement, practically, I do not think it is very important.
Process for putting together this post:
- Elie emailed all staff telling them we were planning to run a staff members donations post this year, as we did last year.
- Staff held 3 meetings to discuss where they were planning to give and why. Meetings were voluntary and attended by some staff but not all. Elie and Holden attended the 3rd meeting but not the first two.
- Staff who chose to participate sent their drafts to Elie and Holden and copied an internal email list that allowed other staff members to see their submissions.
- Entries were arranged in order of length of time employed full-time at GiveWell.
- The draft post was sent out to all staff. At that point staff were able to make final edits.
It is understandable that each individual employee of an orginization might disagree with that orginization’s decisions and reasoning to some extent, but don’t the orginization’s final decisions represent a compromise between all these disagreements? If so, why isn’t the compromise reached by several well informered and qaulified people better than the decision each individual comes to on his own?
This exercise is useful – thank you.
Uri: Individuals might feel that they have an intuitive grasp on the ‘correct’ tradeoff that they can’t (yet) fully articulate and communicate to other staffers. A compromise will favor positions that can be argued and defended well, which isn’t automatically superior to intuitive judgment calls — there are some biases that specifically affect explicit reasoning and group rationality.
Another possible reason for disagreement is that GiveWell staff simply have different values. Discussing things enough should give everyone the same beliefs (ideally, if not in practice), but there’s much less reason to expect discussion to result in everyone having the exact same preferences (even in principle). For example, some staff express a preference for being sure they’re having at least some impact, whereas others are happier to risk having no impact for the sake of having a chance at a much larger impact. This difference might go away with enough discussion, but it might not.
A third possible reason for disagreeing with the consensus position is to signal to donors that there’s a lot of uncertainty surrounding that position, and to help empower donors to think for themselves about the relevant tradeoffs rather than automatically accepting the consensus. Even if you value the consensus position, you may want to encourage outsiders to criticize and nitpick that consensus in order to produce an even more robust ‘super-consensus’ between GiveWell’s recommended allocation and the recommendations of other informed parties.
Thanks for the feedback, Maxwell.
Uri & Rob: I think Rob’s answer is right. His #3 covers my own reason for deviating (slightly) from the GiveWell recommendation, and I think his #1 and #2 also play a role in other cases.
Hi Rob, I largely disagree with your first 2 points, but find your 3rd point compelling enough to win the argument. In any case I think (or at least hope) that making my disagreement explicit might be useful for someone:
“Individuals might feel that they have an intuitive grasp on the ‘correct’ tradeoff that they can’t (yet) fully articulate and communicate to other staffers.”
Sure, but they should also realize that the same is true for other people, hence the attempt at a compromise. The compromise should take in all the factors that go into each individual’s decision making process.
“A compromise will favor positions that can be argued and defended well” Not true – if absolutely necessary, the compromise can come at the very end of the decision making process, in which case all indefensible factors can be taken into account. Something like an average (but not an average since that is overly simplistic in this situation). Also I am highly dubious of positions that cannot be argued or defended (we should try to minimize their effect on our decisions), but that is a topic for another time.
“GiveWell staff simply have different values” – Again, why should anyone’s private values trump the correct aggregation of everyone’s values. Values are susceptible to altercation through rational
discourse, a person can be wrong about what he or she values.
I just thought of another argument in favor of individual decision making over compromise. A distribution of giving preferences might lead to better results than a compromise. After all, it comes to something like a compromise when all the donations are tallied up in the end.
Actually this is more of the beginning of an idea than a fully mature argument…
“all indefensible factors can be taken into account”
Because GiveWell deliberately errs on the side of transparency (and also, to some extent, on the side of conservatism and skepticism), it might be in GiveWell’s interest to promote a ‘consensus’ position that assigns too little weight to these unargued intuitions. GiveWell might (rightly) worry that it would harm their organization to give too many arguments of the form ‘we can’t put into words why we think this, but we think you should just trust us’; they’re forced to do this sometimes, but doing it too much can make them look sloppy (especially if they’re sometimes wrong on these topics). The individuals might then have more freedom than the organization to promote the highest-expected-value allocation of money. I wouldn’t be surprised if this were one reason a lot of staffers are giving less to deworming causes than GiveWell itself recommends.
An alternative hypothesis is that GiveWell staff just find it too difficult and time-consuming to try to quantify everyone’s implicit intuitions, so they’ve adopted the heuristic of excessively privileging explicit argument in their consensus positions just to encourage more productive discussion and less bald assertion.
“they should also realize that the same is true for other people, hence the attempt at a compromise.”
That’s true, but different people may have different amounts of hard-to-articulate intuition; or their intuitions may be unusually high-quality or low-quality. Assuming this is an issue of belief calibration and not of action coordination, how much I defer to others’ not-explicitly-justified conclusions should depend on how much well-calibrated content my implicit world-model has. (Hopefully it’s possible to do this without being completely ruined by bias.)
Holden’s giving matches GiveWell’s recommendations more than Elie’s does, but that could be because Holden has an easier time putting his expectations into words (thus allowing them to be incorporated into the consensus), or because Holden trusts his (Holden’s) hunches less than Elie trusts his (Elie’s) hunches, or because Holden’s unargued hunches just happen to coincide more with the hunches and/or verbal arguments of the rest of the group; and none of those positions is automatically irrational. (I don’t know the actual reason; this is just an example.)
“Again, why should anyone’s private values trump the correct aggregation of everyone’s values.”
Values don’t aggregate in the way that beliefs do. That is, there are theorems to the effect that two perfectly rational agents with the same priors and common knowledge of each other’s beliefs should end up with identical beliefs; but there are no corresponding theorems saying that two perfectly rational agents will need to end up with the same preferences. Presumably all GiveWell staff care in some way about others’ values, but that doesn’t mean they care in exactly the same way about others’ values; even if every GiveWell staffer were a perfect utilitarian, they might be different varieties of utilitarian, who use the word ‘utility’ to refer to irreconcilably different mental and physical states. E.g., there’s no reason to expect a preference utilitarian and a hedonic utilitarian to end up with the same preferences, even if they have identical beliefs (and common knowledge of each other’s preferences and beliefs).
Rob, I think we have very different theories about the role intuitions should play in reasoning. I wouldn’t accept a single statement from Givewell to the effect that: “we can’t put into words why we think this, but we think you should just trust us”, let alone several. In my opinion (and very briefly), relying on intuition is more a matter of reaching bedrock in one’s rational analysis, then weighing competing intuitions, and finally banking on one, because we have to act. Givewell staff usually explain their thinking to an extent that satisfies these requirements. When they don’t, I think they hope their readers will call them out on it.
In addition, anyone how has an above average number of intuitions involved in their beliefs should aim to improve them, probably through articulation, and should not act until he or she has done so. The average already includes plenty of room for intuition. It goes without saying that no one with low-quality intuitions should be employed by Givewell unless they never rely on said intuitions.
In short, I see intuition as a necessary evil. I think Givewell’s research bears this out, many of its conclusion are counter-intuitive, but supported by hard data (e.g. GiveDirectly vs. Heifer).
Your point about values is true but misses my point. From the fact that perfectly rational agents do not lead to hold identical values, it does not follow that all those preferences and values are equally correct. Maybe all of them are completely wrong; maybe all of them are partially, yet equally, wrong; maybe a few are less wrong than others. In all but the first case there is a high probability that a consensus will, at the very least, nudge us in the right direction. I think it is preferable to assume other people’s values have some merit in them, and to take value uncertainty into account when forming beliefs, then to choose a single value system with which to interpret every fact. Compromising is not a matter of changing each person’s values, so the divergence that remains in the end is not a problem.
Comments are closed.