I think there's something to be said that if the qualitative stuff looks bad (e.g. experts are negative, studies are bunch of non-randomized stuff susceptible to endogeneity concerns), you always have the option of just implementing aggressive discounts to the CEA, and the fact that it looks too good means the CEA is done poorly, not that the CEA-focused approach is unreliable.

CEAs are always very uncertain and fragile, and I generally wouldn't put too much stock on the precise multiple relative to GiveWell. For us, we only really care if it's >GW, and we use >=10x as a nominal threshold to account for the fact that GiveWell generally spends a lot more time and effort discounting, relative to ourselves or Foudners Pledge or other researchers.

Overally, health policy interventions are uniquely cost-effective simply because of (a) large scale of impact (policies affect people across the whole country, or at least region or city); (b) low cost per capita (due to economies of scale), and this tends to outweigh the risk of failure (typically, we're looking at 10% success rates, though better charities in promising countries are closer to 30-50% maybe). Still, it's enormously risky, and remember that the median outcome is zero impact 

I would agree for Westminster, but this is relatively tractable in LMICs where governments rely a lot on external NGOs for policy development and implementation + US legislators always have the habit of adding riders to bills. The title may be misleading unless qualified.
 

Edit: As a former civil servant in Singapore, and as someone whose friends are all still serving, our favourite joke is that we signed up to do evidence-based policy-making, but what we often do instead is policy-based evidence-making (and this is in Singapore, which is famed for fairly extreme technocracy).

Anyone who joins a high-income country civil service to make a difference should expect being stalled by bureaucracy (plus legitimate constraints on policy options, whether political or budgetary or operational, that aren't necessarily apparent to the public). Also, it really makes a difference if it's a politician pushing it vs someone in the civil service hierarchy; civil servants are deferential to democratically elected politicians in a way they aren't to their own peers or juniors in the service. Suggesting new ideas is met, at baseline, by the dead-eye stare of someone who knows this just creates more work for them.

Hi Vasco, thanks for flagging out. I've updated our Work With Us page to include the direct donation link (additional 1-3% in fees, but more convenient; if donors prefer minimizing fees, they should feel free to reach out and we can guide them through the cheaper wire transfer)

I think people who have otherwise not looked into this might reasonably update a bit on the fact that someone (us) looked into this, but fundamentally, they shouldn't update a lot or let this (or anything, really) change their minds without having looked into this themselves and satisfied their own worldview.

And to give more information on why I'd rather not publish our human-animal welfare comparisons - I try to regularly review this issue (e.g. there was a considerable revision after the RP moral weights were published, and a smaller one earlier this year), but to not touch this outside those regular revisions (I tend to let myself get sucked into spending too much time thinking about fundamental normative and epistemic issues in a way that is probably not very useful).

Publishing and inviting public suggestions/comments/criticisms would almost certainly cause me to spend too much time on this right now, in a way that would be detrimental to our other ongoing research (mainly effective giving) and our outreach/donor advisory work with non-EA donors (mainly GHD, some AW). On this issue, I'd rather just wait and see what is published this year (and your work is certainly very relevant/useful) and then re-evaluate at one go, maybe in early 2026.

Hi Vasco,

I think our estimates rely too much on subjective input that I don't think it would be useful to others, though I will say that the RP estimates helped eliminate one major source of uncertainty (even if too much remains).

Cheers,
Joel

Generally, I'll say that even when I disagree with Vasco, I admire his willingness to go where the evidence/logic points to, even if the conclusion thus arrived at is extremely unwelcome (e.g. the meat eater problem for saving human lives, or the wild animal problem for the opposite).

FWIW CEARCH has previously looked into the meat eater problem and tried to quantify the downside for animal welfare (in equivalent human DALY terms) when saving a human life, while also trying to adjust for additional considerations such as wild animal suffering and bias (since we're keenly aware that self-interest/preconceived moral values pushes us to reason in a certain direction). Our conclusion is that saving human lives is net positive, but not as high as it would be if not for the meat eater problem.

I'm not sure how much I would update on Vasco's argument (convenient as it is for our GHD work), and my main uncertainties relate both to (a) neuron count (I wouldn't rely on LLMs to spit out the correct answer here, because even beyond the usual risks of hallucinations when there is a correct answer written out there, the true value for this probably doesn't even exist in the written literature); and (b) the neuron/welfare relationship (though Vasco's regression is pretty interesting, and a reasonably good first step).

Overall, I think more research in this area (and more funding for such research) is clearly merited.

Thanks for flagging out! Have fixed it.

Hey Vasco, the adjustment is specific to GiveWell vs us (or indeed, non-GW CEAs), since GiveWell probably is the most rigorous in discounting, while other organizations are less so, for various reasons (mainly time - that's true for us, and why we just use a rough 10x GW threshold; and it's true of FP too; Matt Lerner goes into detail here on the tradeoff between drilling down vs spending researcher time finding and supporting more high EV opportunities instead).

Relative to every other organization, I don't find CEARCH to be systematically overoptimistic in the same way (at least for our deep/final round CEAs).

For our GWWC evaluation, I think the ballpark figure (robustly positive multiplier) probably still holds, but I'm uncertain about the precise figure right now, after seeing some of GWWC's latest data (they'll release their 2023-24 impact evaluation soon).

Our grantmaking always aims at maximizing DALYS averted (with income and other stuff translated to DALYs too).

In terms of cost-effectiveness, it's nominally 30-50x GW, but GiveWell is more rigorous in discounting, so our figures should be inflated relative to GW. Based on some internal analysis we did of GW's greater strictness in individual line-item estimation and in the greater number of adjustments they employ, we think a more conservative estimate is that our estimates may be up to 3x inflated (i.e. something we think is 10x GW may be closer to 3x GW, which is why we use a 10x GW threshold for recommending GHD causes in the first place - to ensure that what we recommend is genuinely >GW, and moving money to the new cause area is +EV).

So my more conservative guess for our grantmaking is that it's closer to 9-15x GW, but again I have to emphasize the high uncertainty (and riskiness, which is the inherent price we pay for these ultra high EV policy interventions).