Thus, individual small donors with time to do research should expect to be able to find highly cost-effective giving opportunities that GiveWell missed.
This does not seem at all plausible to me. If your research is good, shouldn’t you contribute it to the commons? And if you can harness the commons, you should care about the room for funding. GiveWell does what it does for a reason and if you compete with it, you’ll become similar. If there is opportunity for more research, shouldn’t the answer be for GiveWell to expand or be duplicated? And you do give examples where the answer seems to be to duplicate it, but shift its methods and goals. But none of them seem to have to do with the size of the donor, or the ratio of research to funding.
If there is a proliferation of researchers, they won’t have track-records to harness large funding. But how do the researchers themselves know that their research is better than the alternative researchers?
Later you reveal that you were talking about is local improvements. Why not say that up front if that is what you mean? And I think that the relevant difference is not in the quote, but local knowledge that you are unable to communicate to another evaluator.
If there is opportunity for more research, shouldn’t the answer be for GiveWell to expand or be duplicated?
If you believe in constant or increasing returns to scale and largely overlapping values, sure. It’s not obvious to me why we shouldn’t expect GiveWell’s highly centralized model to outperform a model more like academia, with little centralization but lots of criticism.
There are a few ways you might expect to be able to do better:
There’s effectively a size floor on things GiveWell can afford to look at because of the amount of money they want to move and limited staff time.
GiveWell recommendations are tailored for a set of preferences that may not be the same as yours, e.g. a preference for high levels of confidence and easy to explain evidence such as RCTs, even at the expense of EV.
Some pieces of information are easier for you to learn than to communicate or for others to verify. For instance, it might make a lot of sense for you to trust a friend you’ve known since childhood a lot, another friend to trust them a little based on your say-so, but strangers on the internet to trust them not at all based on your say-so.
Aceso Under Glass’s post about Tostan is a good example of novel research that has some overlap with the second and third consideration.
I agree that in general you should contribute your research to the commons.
There’s effectively a size floor on things GiveWell can afford to look at because of the amount of money they want to move and limited staff time.
My criticism is narrow and specifically about this point.
Yes, there is a size floor. New researchers will pursue smaller projects. But that is an example of diminishing returns to research. The next researcher should pursue projects only slightly smaller than the previous researcher, for the same reason that the previous research pursued large projects.
Thanks for clarifying. I should have been able to figure this out from your original comment. Oops!
If there is opportunity for more research, shouldn’t the answer be for GiveWell to expand or be duplicated?
GiveWell has limited ability to scale even with more money—my sense is that they don’t have a bunch of potential analysts lined up whom they’d be delighted to hire (or try out) if only there were the money for it. This is a feasibility objection to expansion on current margins. However, GiveWell might reallocate resources to this from other things (e.g. outreach) if they had enough money, so if you have other strong reasons to think GiveWell can do better than you can, it might be good to give them money anyway.
On duplication, doing one’s own research is a sort of duplication, at a smaller scale. I suspect that GiveWell is way past the point at which returns to scale diminish.
(Disclosure: I worked at GiveWell in the past. I have no current affiliation with GiveWell. My opinions are my own, an not a representation of their official opinion.)
GiveWell also perceives its audience as uninterested in higher expected value but higher risk options.
Doesn’t seem entirely plausible, given that GiveWell’s second-highest rated charity and several of the other next-ranked charities are acknowledged to likely have close-to-zero impact, in exchange for high expected value. They don’t seem to shy away from a cause just because they perceive it as high-risk.
This does not seem at all plausible to me. If your research is good, shouldn’t you contribute it to the commons? And if you can harness the commons, you should care about the room for funding. GiveWell does what it does for a reason and if you compete with it, you’ll become similar. If there is opportunity for more research, shouldn’t the answer be for GiveWell to expand or be duplicated? And you do give examples where the answer seems to be to duplicate it, but shift its methods and goals. But none of them seem to have to do with the size of the donor, or the ratio of research to funding.
If there is a proliferation of researchers, they won’t have track-records to harness large funding. But how do the researchers themselves know that their research is better than the alternative researchers?
Later you reveal that you were talking about is local improvements. Why not say that up front if that is what you mean? And I think that the relevant difference is not in the quote, but local knowledge that you are unable to communicate to another evaluator.
If you believe in constant or increasing returns to scale and largely overlapping values, sure. It’s not obvious to me why we shouldn’t expect GiveWell’s highly centralized model to outperform a model more like academia, with little centralization but lots of criticism.
GiveWell did expand and created GiveWell labs. That’s now OpenPhil and OpenPhil directs some money into causes who’s effect is harder to measure.
There are a few ways you might expect to be able to do better:
There’s effectively a size floor on things GiveWell can afford to look at because of the amount of money they want to move and limited staff time.
GiveWell recommendations are tailored for a set of preferences that may not be the same as yours, e.g. a preference for high levels of confidence and easy to explain evidence such as RCTs, even at the expense of EV.
Some pieces of information are easier for you to learn than to communicate or for others to verify. For instance, it might make a lot of sense for you to trust a friend you’ve known since childhood a lot, another friend to trust them a little based on your say-so, but strangers on the internet to trust them not at all based on your say-so.
Aceso Under Glass’s post about Tostan is a good example of novel research that has some overlap with the second and third consideration.
I agree that in general you should contribute your research to the commons.
My criticism is narrow and specifically about this point.
Yes, there is a size floor. New researchers will pursue smaller projects. But that is an example of diminishing returns to research. The next researcher should pursue projects only slightly smaller than the previous researcher, for the same reason that the previous research pursued large projects.
Thanks for clarifying. I should have been able to figure this out from your original comment. Oops!
GiveWell has limited ability to scale even with more money—my sense is that they don’t have a bunch of potential analysts lined up whom they’d be delighted to hire (or try out) if only there were the money for it. This is a feasibility objection to expansion on current margins. However, GiveWell might reallocate resources to this from other things (e.g. outreach) if they had enough money, so if you have other strong reasons to think GiveWell can do better than you can, it might be good to give them money anyway.
On duplication, doing one’s own research is a sort of duplication, at a smaller scale. I suspect that GiveWell is way past the point at which returns to scale diminish.
(Disclosure: I worked at GiveWell in the past. I have no current affiliation with GiveWell. My opinions are my own, an not a representation of their official opinion.)
Doesn’t seem entirely plausible, given that GiveWell’s second-highest rated charity and several of the other next-ranked charities are acknowledged to likely have close-to-zero impact, in exchange for high expected value. They don’t seem to shy away from a cause just because they perceive it as high-risk.
I think that would be better as
because it is compatible with the parse tree.