Following up on our open thread in December, we wanted to have another one.
Our goal is to give blog readers an opportunity to publicly raise comments or questions about GiveWell or related topics (in the comments section below). As always, you’re also welcome to email us at email@example.com or to request a call with GiveWell staff if there’s feedback or questions you’d prefer to discuss privately. We’ll try to respond promptly to questions or comments.
If you have questions related to the Open Philanthropy Project, you can post those in the Open Philanthropy Project’s open thread.
I’ve been curious about life-extension research as a possible target for funding. The hard cap on human lifespan seems like a really obvious target for increasing QALYs. If the research is too speculative at the moment to provide sufficient confidence for a GiveWell recommendation, that itself would be fodder for an interesting and important discussion.
There’s a weird middle ground between giving people cash (a la GiveDirectly) and giving people interventions that they need, regardless of their own awareness of the need (ex, AMF). However, the charity Watsi seems to be trying to fill the intermediate space (paying for the medical interventions people need). Correct me if I’m wrong, but I can’t find any indication that GiveWell has evaluated Watsi. Have/Are/Will you?
Tony: life extension is an interesting topic. As groups like SENS have pointed out, even total elimination of major diseases like cancer adds surprisingly little to longevity, while conversely, even a small reduction in the acceleration of mortality yields large gains.
The problem, of course, is that anti-aging treatments have a long and illustrious history of failure, and you have to assume that many, if not all, of the things SENS and others will try are going to fail too, which makes it difficult to evaluate such longshot projects.
If you set your sights lower, there are a lot of worthy interventions into all-cause mortality reduction. Reductions in all-cause mortality are very difficult to detect in typical clinical trials of less than a thousand people, less than 10 years followup, and mid to old age like 30-60s, so getting trials big enough costs a lot, but are still worthwhile to fund. I’ve done some noodling around with modeling a few interventions & aging in http://gwern.net/Longevity if you’re interested in playing with numbers.
It would be cool if research writeups had the author name(s) listed. Sometimes I have questions about the writeups and my current strategy is to pick a GW employee at random and ask them.
What ever happen to Givewell’s consideration of climate mitigation/climate engineering pilot studies? What about the one run by U Washington?
Almost all animal suffering isn’t caused by humans, but occurs in the wild. Right now we don’t have any idea what to do about it, but this seems like an area where Open Phil is uniquely positioned to have an impact. You could request grant proposals on reducing wild-animal suffering and probably do a good job of motivating more much-needed work in this area. Have you thought about this?
Obviously I’m not claiming that we will immediately know how to solve wild animal suffering after making a few grants. But it doesn’t seem particularly harder to solve than, say, AI safety, which is an area that’s getting some attention from Open Phil. And no one else right now has the funding necessary to motivate more research, so you’re uniquely positioned to make something happen.
Gwern, I am a pretty big fan of the idea of improved brain transplant technology. We can do lots of other sorts of organ transplants, and so it seems like that is good evidence it should be possible. If you can transplant people’s brains into other bodies, then you can get rid of 90% of the dangers of aging. Only brain related dangers would remain. Seems like potential huge progress.
Of course, this idea is also extremely icky to me, but I think the cost of breaching the taboo is worth it, and if we don’t breach any taboos we will be much less likely to succeed.
Michael – re: questions on research writeups: Please feel free to email your questions to firstname.lastname@example.org. We monitor that inbox and get the most knowledgeable person to answer the question, if necessary (it also helps us to have questions come through a central location). You can also post your questions on whichever open thread was most recent. We could create a more permanent solution for public questions than periodic open threads if there’s demand for that.
Tony, Durbrow, and Michael – We’ve just published a separate open thread for questions related to the Open Philanthropy Project. We will copy over and address your questions related to life-extension research, geoengineering, and wild animal suffering in the comments section on that post: https://www.openphilanthropy.org/blog/march-2016-open-thread.
Tony, re: Watsi – it does not qualify as eligible for a GiveWell recommendation (though it is possible that this will change if we conduct more investigation into surgery as an intervention; here’s our latest content on that). For what it’s worth, I don’t see much of a fundamental difference between Watsi and AMF on the axis you describe: AMF is generally distributing nets to people who want them, and both Watsi and AMF are quite far from giving people full leeway over how they use funds. I think the case for either (relative to GiveDirectly) would have to be made on a cost-effectiveness basis, incorporating quality of evidence.
(Your links are malformed, Holden.)
How important is preventing deforestation? What would be the consequences if humanity kept destroying all remaining rainforests?
Anne, We have not focused on deforestation in GiveWell’s research, although it has come up — albeit in a minimal way — in our investigation into human-caused climate change as a cause area for the Open Philanthropy Project, a separate research process open to higher-risk, higher-reward opportunities and aimed at larger, institutional philanthropists. Following our initial climate change investigation, we are not currently pursuing deforestation as a potential Open Philanthropy Project focus area.
Thank you for your reply!
Anne, if you’re interested in deforestation as an issue, you may want to look at Cool Earth and the analysis of them at Giving What You Can. Cool Earth works on protecting tropical rainforest in partnership with indigenous people.
According to Giving What You Can, Cool Earth is the most efficient climate change charity they looked at, with a likely efficiency in the range $0.3/t to $1.32/t in terms of charitable dollars per ton of atmospheric CO2 or equivalent averted. In terms of effects on human health — by $/life saved or $/DALY — they estimate that it is much less efficient than the most efficient charities (AMF, etc.). So I’d say how attractive Cool Earth is as a charity (or more generally, preventing rainforest deforestation is as an intervention) rests in large part on how you weigh benefits to people versus benefits to ecosystems (preventing extinctions, etc.).
Have you thought about collating and linking to analyses on other websites that appear to be of good quality and of relevance to GiveWell’s mission?
This might be a way to add an “extra-shallow” level of coverage for causes that GiveWell has not (yet) analyzed. An example (motivated by my previous comment!) might be climate change (GiveWell/OP does cover the sub-cause of solar radiation management, but not really I think aside from that). If you agree that Giving What You Can’s analysis on climate change as a cause seems good, then you could link to it (or whatever source(s) you think make most sense) on your Climate Change page and perhaps give a brief summary. If later you came back and filled in that page with GiveWell researched content, you might drop the outside links or include it as a little See Also section. A thought.
When you launched GiveWell, there was very little if any such content accessible elsewhere. Even today, many websites that are recommending effective charities are doing so one the basis of GiveWell’s excellent analysis. Still, there is more and more analysis generated by other websites and I wonder if it makes sense for GiveWell to use or link to some of that in some way.
Do you have a timeline of when you’ll be publishing your annual review? I’m mostly interested in your annual review of money moved along with the distribution by charity and size buckets, as you have done in previous years.
Colin — Thanks for this suggestion. We do periodically recommend external sources on our website, but only on a case by case basis, depending on our familiarity with the source and how helpful we believe it would be to readers of our website. We expect it would be challenging to do this more comprehensively.
> If you can transplant people’s brains into other bodies, then you can get rid of 90% of the dangers of aging. Only brain related dangers would remain. Seems like potential huge progress.
As the saying goes, the first 90% takes 90% of the time, then the last 10% takes the other 90% of the time. You haven’t made that huge progress because you still have to contend with the Gompertz curve.
Think about how many people get senile dementia, Alzheimers etc long before their body gives out, without any of the massive trauma that comes with transplants of pretty much anything (kidney, liver, heart transplants: all of these come with substantial health penalties from the transplant and iatrogenics; the point is that they are sometimes better than the alternative). This is the fundamental problem of aging: if you solve A completely, that just means B will kill you a year later.
You can see this playing with the simulation. Try simulation a scenario in which instantaneous mortality is reduced by 33% because of, say, a silver bullet for cancer being discovered, and see what sort of life expectancy gain that translates into. It’s nice, but it’s not remotely what you need to pay for the R&D to do whole-body transplants, much less cover the $100k+ a hypothetical operation would entail.
Vipul, We’re close to publishing our annual review. In the meantime, I can share some preliminary figures, with the caveat that the final numbers could look significantly different.
Thank you, Catherine! The spreadsheet is good enough for me for now.
I was wondering whether your critical opinion on DCP-2 (https://www.givewell.org/international/technical/criteria/cost-effectiveness#Thecosteffectivenessfiguresweuse) had changed with the publication of DCP-3 last year? Are you more/equally/less confident in their numbers, and do you rely on them (again)? If so, why?
(I am particularly interested in their chapter on Mental, Neurological and Substance Use Disorders but that’s a long shot; I expect that you would not have looked into this specifically)
Thanks so much in advance for any info!
Konstantin — We’re aware of and following the DCP-3, but have not yet formed a view on its content.
Thank you, Catherine. Would be interested in hearing with what you eventually conclude.
Today (April 25th) is World Malaria Day. I wrote a short article encouraging people to give $2.50 to The Against Malaria Foundation, GiveWell’s current top charity:
Linchuan — Thanks for spreading the word about AMF!
Does GiveWell have an official stance on population ethics as it relates to saving lives in the short term?
A GiveWell-commissioned report (https://blog.givewell.org/2014/04/17/david-roodmans-draft-writeup-on-the-mortality-fertility-connection/) suggests that population will hardly change as a result of AMF saving lives. GiveWell’s cost-effectiveness model for AMF (https://www.givewell.org/international/technical/criteria/cost-effectiveness/cost-effectiveness-models) assumes that saving one life creates ~35 DALYs, and uses this as the basis for how good it is to save a life. But if AMF causes populations to slightly decline, that means it’s actually removing (human) DALYs from the world, so this assumption is massively off–you can’t justify AMF by saying it creates more happy human life, because it doesn’t.
You could instead justify AMF’s life-saving effects by saying it’s inherently good to save a life, in which case GiveWell’s cost-effectiveness model shouldn’t interpret the value of lives saved in terms of DALYs created/destroyed. (I don’t know how much this CE model mattered for your decision to rank AMF as your #1 charity.) That means GiveWell is probably way overestimating the value of AMF, and the value of AMF depends heavily on how you convert lives saved to DALY-equivalents when saving a life doesn’t actually create more life-years.
One component of GiveWell’s recommendation is that AMF is ~10 times more cost-effective than GiveDirectly, which depends on the assumption that saving a life creates 35 DALYs. How do you compare lives saved to life-years created? Can you justify ranking AMF as your top charity on this basis?
Michael — thanks for the question. This is a good example of the type of subjective, moral judgments that our cost-effectiveness analyses incorporate. Assuming the values laid out in your comment, you’re right that AMF is less cost-effective. However, many people (myself included, and I think many GiveWell staff) consider preventing deaths to be valuable but don’t consider causing additional births to be valuable in the same way (ie, they wouldn’t consider preventing additional births in cases where the parents choose not to have a child to be a negative thing). That is, they would consider saving the life of someone who already exists (and would have died without a bednet) to be a good thing but would not consider the decline in the fertility rate due to a decline in the mortality rate to be a bad thing, much less one that offsets the gains from saving the life of the child that did already exist.
If you disagree with the assumptions GiveWell staff make in the cost-effectiveness analysis spreadsheet and would rather input your own values than follow the median of staff estimates, you can do so here: https://www.givewell.org/international/technical/criteria/cost-effectiveness/cost-effectiveness-models. You could also support a charity whose main benefit isn’t saving lives. We believe the primary benefit of GiveWell top charities GiveDirectly, Deworm the World Initiative, and the Schistosomiasis Control Initiative is to improve lives, rather than save them.
Sean, I think that’s a great response to Michael.
A different issue on the cost-effectiveness analysis: GiveWell uses a value of ~35 DALY’s per life saved by a bed net, which is less than the life expectancy at age 5, by more than a factor of 2. The difference comes about from applying an annual discount rate of 3%. (There’s also the relatively minor countervailing issue of age weights, but let’s ignore that for present purposes. This is all drawn from this paper cited by your spreadsheet. See Table 5.1, the YLL(3,1) column for Age 5.)
I’m curious why do folks at GiveWell think, philosophically, that it’s appropriate to apply a discount factor here? I mean, that may be standard practice in this context, but why do you think that practice is correct?
I think of discount factors in this kind of context as principally about reflecting increasing uncertainty as you go farther out into the future.
So if you imagine a bed net distribution that will happen 3 years from now, then it makes sense to me to apply a discount factor to reflect those 3 years. Maybe bed nets will become less much effective over that period for some reason (e.g. because of evolution in the vector) or malaria treatment will dramatically improve so that far fewer die.
But on the other hand if we’re talking about a 5 year old dying today, I don’t see why you discount the future years of her life. It doesn’t matter for her what happens with mosquitoes or malaria treatment or anything else in the future years.
Thanks for your response. I started writing a reply, but it ended up getting a bit long for a comment, so I wrote something longer about it: http://effective-altruism.com/ea/xo/givewells_charity_recommendations_require_taking/
I’d appreciate if you could read it and tell me what you think.
Hey Colin, thanks for the question. This is a good question, and staff don’t all agree on the answer. Briefly, the idea of discounting in this way comes from an impression that saving a DALY now is more valuable than saving one in the future, even if the certainty is the same. That leads to discounting future DALYs even if they’re part of the same life saved. One way to think about this would be to consider, in isolation, whether you’d rather save a DALY now for $10 or save one in 10 years for the same $10. The former seems better, if only because the latter seems inferior to investing your money in the meantime and giving more to the same place later. It’s not always clear whether this logic fully justifies discounting benefits in a given case, but we generally use a discount rate to avoid background implications such as the idea that the two hypothetical options here are equivalent. Of course, this is debatable, and so we’re planning to write more about this in more depth in the next few months.
Sean, thanks for the reply.
I’m confused by the investment argument you’ve stated for discounting future DALY’s (even in the absence of uncertainty). The argument seems to be: say the price per DALY today is $x ($10 in your example). You can spend that $x now to save a DALY or you can invest it for 10 years and then (in expected value) you’ll have more than $x. So far so good. But then you jump to the conclusion that that means you can save more than one DALY in the future with the money. But there’s a huge assumption here! You seem to be assuming that the price per DALY is constant (or maybe that it is constant in EV). I can see no reason to expect that (and substituting a high inflation currency for the dollar it seems obviously false).
A weaker form of this argument would be to assert that the expected growth in the cost per DALY is less than the expected investment return. Maybe there are good reasons to expect this, I don’t know. But my first instinct would be to think if anything the inequality, at least today, goes in the opposite direction. Currently, interest rates are low. On the other hand, it’s quite plausible to imagine e.g. the funding gap for bed nets and deworming getting closed in the new few years (which would be wonderful!), with the remaining best interventions with capacity materially more expensive in terms of $ per marginal DALY. After all, the dollars directed by GiveWell are growing exponentially (which is exciting! congrats!), and my feeling is also outside of GiveWell more and more giving (while still a small fraction of total giving) is being directed with similar concerns in mind.
Michael – Thanks for the thoughts. This is one of a number of elements of our cost-effectiveness analysis that we haven’t written about as extensively as we’d like to, so we put this on our list of things to try to write more about in the future. One thing I’ll mention is that I think it’s not clear from David Roodman’s report that lives saved are offset 1:1 by decreased mortality; I’d say that it’s unclear what the net effect on the population is.
Colin – Over the long run, I agree that we should expect DALYs to be more expensive in the future. But over the next few years, we don’t see a particular reason to expect this. And one of the purposes of the discount rate is to enable comparisons to things like GiveDirectly, where we expect very little of this kind of decay. Ideally we would have a complete and detailed model that incorporates flow-through effects of giving now instead of later, increasing “price of doing good over time,” etc. but in the absence of that model, using a uniform discount rate is simple, and seems better than using no discount rate (which seems to imply that e.g. it’s never a good idea to give now rather than saving). But if you don’t agree with the discount rate we’ve used, we encourage you to put in your own and see whether it changes the bottom line.
Discounting and GiveDirectly
Sean, even for GiveDirectly, I’m not convinced it’s reasonable to expect that DALY/$ should be roughly constant over time. True, a dollar given 10 years from now should end up with about the same nominal value in dollars in the recipient’s hands as today (and likely more in local currency, say Kenyan shillings). But to talk about DALY’s, I’d argue the more relevant variable is the transfer amount as a fraction of income (so a log utility framework). Given that, we expect the $/DALY to decay with the growth rate in income of the typical recipient. (Unlike bed nets and deworming, I think we fundamentally don’t expect capacity to much of an issue for cash transfers for the foreseeable future, except for transitory expansion issues.)
Discounting to handle unmodeled uncertainty
What about uncertainty as a basis for discount rates? The idea as you know is that a discount rate is used as a method to capture the effect of uncertainties that are otherwise unmodeled including “unknown unknowns” (this only makes sense for uncertainties that grow with time, not e.g. uncertainty whether an intervention works).
Decomposing the discount rate
Do people at GiveWell tend to agree with this concept of uncertainty-driven discounting? If so, I’d suggest, if you also decide to keep the investment return concept, that you decompose
your discount rate into its two components. So your default discount rate might be 3%=1%+2%, with 1% the investment component and 2% the (default) uncertainty component. Then, for specific cases you can think about how it might make sense to deviate from this default. For bed nets, you might model the uncertainty component as 2% for the first 2 years and 0% thereafter (using 2 years as an estimate of the average delay to the death averted). So the total discount rate would have the term structure of 3% for the first 2 years and 1% thereafter. For another intervention, e.g. in an area with rapid innovation, you might apply an uncertainty discount rate above 2% (and total discount rate above 3%).
Of course, none of this really matters for AMF specifically. It’s already a GiveWell top charity; so if you were to double the DALY/$ efficiency for AMF (as I’d advocate), needless to say AMF would still be a top charity. But it does illustrate how big an effect discount rates can have. And therefore why they are worth some serious thought (which you are giving it, hence the upcoming blog post!).
Colin – I agree that some decay in the DALY/$ for GiveDirectly could happen as the incomes of the very poorest people in the world rise, but it’s still possible that in a decade a substantial number of people will be living at a similar income level as current GiveDirectly recipients do today, so it’s hard to say exactly. Either way, we’d guess the decay will be substantially smaller than for the most cost-effective giving opportunities, where e.g. bednets may not be necessary at all at some point in the future.
Staff have different opinions on using uncertainty as a basis for discount rates, and I think that some agree with that type of uncertainty-driven discounting, so the current discount rate input is used as a catch-all for different intuitions about discounting.
You’re right in your general point that the discount rates can be a big driver of our CEA models, which due to the uncertainty surrounding them contributes to our overall impression that they’re largely subjective and leads to not to want to put too much weight on them. In general, skepticism about discounting (or a preference for a lower discount rate) should lead people to move more towards favoring AMF and deworming over GiveDirectly, relative to our position (which as you point out already recommends giving to AMF currently, so doesn’t affect the current recommendation but could have a practical effect in the future).
Unrelated question: Has GiveWell studied subsidized or free agricultural microinsurance, such as providing rainfall insurance to poor farmers, as an intervention? The idea as you may know is that when protected by insurance, farmers can invest more in fertilizer and the like for their crops. See e.g. IPA’s work on that. It would be interesting to try to understand how the effect on income or wealth per charitable dollar compares to unconditional cash transfers (like Give Directly).
Hey Colin – we haven’t studied that; it’s a potentially interesting area to compare to unconditional cash transfers but we’re not aware of a strong evidence base for that type of program.
Thanks Sean. I don’t know much about this, but FYI you can see a summary of 10 randomized studies on rainfall-based index insurance here. (There has also been some work on yield-based crop insurance, but that is more prone to moral hazard issues.)
Thanks, Colin! We’ve put this on our list of things to look into further.
Comments are closed.