I think one could probably push back on whether 7.5 minutes of [extreme] pain is a reasonable estimate for a person who dies from malaria, but I think the bigger potential issue is still that the result of the BOTEC seems highly sensitive to the “excruciating pain is 100,000 times worse than fully healthy life is good” assumption—for both air asphyxiation and ice slurry, the time spent under excruciating pain make up more than 99.96% of the total equivalent loss of healthy life.[1]
I alluded to this on your post, but I think your results imply you would prefer to avert 10 shrimp days of excruciating pain (e.g. air asphyxiation / ice slurry) over saving 1 human life (51 DALYs).[2]
If I use your assumption and also value human excruciating pain as 100,000 times as bad as healthy life is good,[3] then this means you would prefer to save 10 shrimp days of excruciating pain (using your air asphyxiation figures) over 4.5 human hours of excruciating pain,[4] and your shrimp to human ratio is less than 50:1 - that is, you would rather avert 50 shrimp minutes of excruciating pain than 1 human minute of excruciating pain.
To be clear, this isn’t a claim that one shouldn’t donate to SWP, but just that if you do bite the bullet on those numbers above then I’d be keen to see some stronger justification beyond “my guess” for a BOTEC that leads to results that are so counterintuitive (like I’m kind of assuming that I’ve missed a step or OOMs in the maths here!), and is so highly sensitive to this assumption.[5]
I think one could probably push back on whether 7.5 minutes of pain is a reasonable estimate for a person who dies from malaria
My calculation above assumed 7.56 min (= 0.126*60) of excruciating pain, not just “pain”. Examples of excruciating pain include “scalding and severe burning events [in large parts of the body]”, or “dismemberment, or extreme torture”.
To be clear, this isn’t a claim that one shouldn’t donate to SWP, but just that if you do bite the bullet on those numbers above then I’d be keen to see some stronger justification beyond “my guess” for the BOTEC.
I had to guess the intensity of excruciating pain because there is basically no empirical data informing potential estimates.
We find it unlikely that the most intense pain experienced is of an Excruciating nature as defined in the Welfare Footprint framework, since this category is by definition associated with extreme and unbearable pain, not tolerated even if for a few seconds (a definition which does not coincide with the description of the patients in the studies above).
I agree my guess is speculative. However, if one wants to argue that I overestimated the cost-effectiveness of SWP, one has to provide reasons for my guess overestimating the intensity of excruciating pain. I do not think claiming the results are unintuitive is a good way of doing this. In the context of global health and development, it would not make much sense to dismiss GiveWell’s conclusion that one can save a life for 5 k$ just because it is unintuitive that one can save 10 lives for the cost of a BMW. Instead, it would be better to critique the inputs that went into the cost-effectiveness analysis. I know you are doing better than this, because you are critiquing intermediate results instead of the final cost-effectiveness. However, I would say it would be better if you focussed on criticising my assumption that excruciating pain is 100 k times as intense as fully healthy life, or, equivalently, the direct implication that 1 day of fully healthy life is neutralised by 0.864 s (= 24*60^2/(100*10^3)) of excruciating pain.
Otherwise I could just copy your entire BOTEC, and change the bottom figure to 1000 instead of 100k, and change your topline results by 2 OOMs.
If you did that, SWP would still be 434 (= 43.4*10^3*10^3/(100*10^3)) times as cost-effective as GiveWell’s top charities. I think it is also worth wondering about whether you trully believe that updated intensity. Do you think 1 day of fully healthy life plus 86.4 s (= 0.864*100*10^3/100) of scalding or severe burning events in large parts of the body, dismemberment, or extreme torture would be neutral?
Ah my bad, I meant extreme pain above there as well, edited to clarify! I agree it’s not a super important assumption for the BOTEC in the grand scheme of things though.
However, if one wants to argue that I overestimated the cost-effectiveness of SWP, one has to provide reasons for my guess overestimating the intensity of excruciating pain.
I don’t actually argue for this in either of my comments.[1] I’m just saying that it sounds like if I duplicated your BOTEC, and changed this one speculative parameter to 2 OOMs lower, an observer would have no strong reason to choose one BOTEC over another just by looking at the BOTEC alone. Expressing skepticism of an unproven claim doesn’t produce a symmetrical burden of proof on my end!
Mainly just from a reasoning transparency point of view I think it’s worth fleshing out what these assumptions imply and what is grounding these best guesses[2] - in part because I personally want to know how much I should update based on your BOTEC, in part because knowing your reasoning might help me better argue why you might (or might not) have overestimated the intensity of excruciating pain if I knew where your ratio came from (and this is why I was checking the maths and seeing if these were correct, and asking if there’s stronger evidence if so, before critiquing the 100k figure), and because I think other EAF readers, as well as broader, lower-context audience of EA bloggers would benefit from this too.
If you did that, SWP would still be 434 (= 43.4*10^3*10^3/(100*10^3)) times as cost-effective as GiveWell’s top charities.
Yeah, I wasn’t making any inter-charity comparisons or claiming that SWP is less cost-effective than GW top charities![3] But since you mention it, it wouldn’t be surprising to me if losing 2 OOMs might make some donors favour other animal welfare charities over SWP for example—but again, the primary purpose of these comments is not to litigate which charity is the best, or whether this is better or worse than GW top charities, but mainly just to explore a bit more around what is grounding the BOTEC, so observers have a good sense on how much they should update based on how compelling they find the assumptions / reasoning etc.
I think it is also worth wondering about whether you trully believe that updated intensity. Do you think 1 day of fully healthy life plus 86.4 s (= 0.864*100*10^3/100) of scalding or severe burning events in large parts of the body, dismemberment, or extreme torture would be neutral?
Nope! I would rather give up 1 day of healthy life than 86 seconds of this description. But this varies depending on the timeframe in question.
For example, I’d probably be willing to endure 0.86 seconds of this for 14 minutes of healthy life, and I would definitely endure 0.086 seconds of this than give up 86 seconds of healthy life.
And using your assumptions (ratio of 100k), I would easily rather have 0.8 seconds of this than give up 1 day of healthy life, but if I had to endure many hours of this I could imagine my tradeoffs approaching, or even exceeding 100k.
I do want to mention that I think it’s useful that someone is trying to quantify these comparisons, I’m grateful for this work, and I want to emphasise that these are about making the underlying reasoning more transparent / understanding the methodology that leads to the assumptions in the BOTEC, rather than any kind of personal criticism!
Thanks, Bruce! Makes sense. I have now clarified in the post that the guesses for the pain intensities come from my personal time trade-offs:
My assumptions for the pain intensities are guesses for my personal time trade-offs, and imply each of the following individually neutralise 1 day of fully healthy life:
Nope! I would rather give up 1 day of healthy life than 86 seconds of this description. But this varies depending on the timeframe in question.
For example, I’d probably be willing to endure 0.86 seconds of this for 14 minutes of healthy life, and I would definitely endure 0.086 seconds of this than give up 86 seconds of healthy life.
0.1 s of excruciating pain intuitively passes very quickly, so I can easily imagine it being less than 100 times as bad as 10 s (= 0.1*100) of excruciating pain. However, I think intuitions like this are misguided holding the intensity of pain constant (i.e. only varying duration). For example, if 1 min in strong pain is 100 times as bad as 1 min in weak pain, and strong and weak pain have each a constant intensity (instead of referring to ranges of intensities), N min in strong pain should be 100 times as bad as N min in weak pain. Saying otherwise implies that the badness of 1 min in strong pain depends on how many minutes of strong pain preceded that 1 min, which contradicts the assumption that strong pain has a constant intensity.
I think 0.1 s of excruciating pain does not feel bad because one wrongly imagines the memory of the pain afterwards, which may not be so bad because 0.1 s is too little time to form memories. In addition, one may imagine that some time is needed to reach the intensity of excruciating pain, such that only a small fraction of the 0.1 s are actually excruciating pain, but the thought experiment requires 0.1 s of actual excruciating pain.
I think one could probably push back on whether 7.5 minutes of [extreme] pain is a reasonable estimate for a person who dies from malaria, but I think the bigger potential issue is still that the result of the BOTEC seems highly sensitive to the “excruciating pain is 100,000 times worse than fully healthy life is good” assumption—for both air asphyxiation and ice slurry, the time spent under excruciating pain make up more than 99.96% of the total equivalent loss of healthy life.[1]
I alluded to this on your post, but I think your results imply you would prefer to avert 10 shrimp days of excruciating pain (e.g. air asphyxiation / ice slurry) over saving 1 human life (51 DALYs).[2]
If I use your assumption and also value human excruciating pain as 100,000 times as bad as healthy life is good,[3] then this means you would prefer to save 10 shrimp days of excruciating pain (using your air asphyxiation figures) over 4.5 human hours of excruciating pain,[4] and your shrimp to human ratio is less than 50:1 - that is, you would rather avert 50 shrimp minutes of excruciating pain than 1 human minute of excruciating pain.
To be clear, this isn’t a claim that one shouldn’t donate to SWP, but just that if you do bite the bullet on those numbers above then I’d be keen to see some stronger justification beyond “my guess” for a BOTEC that leads to results that are so counterintuitive (like I’m kind of assuming that I’ve missed a step or OOMs in the maths here!), and is so highly sensitive to this assumption.[5]
Air asphyxiation: 1- (5.01 / 12,605.01) = 0.9996
Ice slurry: 1 - (0.24 / 604.57) = 0.9996
1770 * 7.5 = 13275 shrimp minutes
13275 / 60 / 24 = 9.21875 shrimp days
There are arguments in either direction, but that’s probably not a super productive line of discussion.
51 * 365.25 * 24 * 60 = 26,823,960 human minutes
26,823,960 / 100,000 = 268.2396 human minutes of excruciating pain
268.2396 / 60 = 4.47 human hours of excruciating pain
13275 / 268.2396 = 49.49 (shrimp : human ratio)
Otherwise I could just copy your entire BOTEC, and change the bottom figure to 1000 instead of 100k, and change your topline results by 2 OOMs.
Thanks, Bruce.
My calculation above assumed 7.56 min (= 0.126*60) of excruciating pain, not just “pain”. Examples of excruciating pain include “scalding and severe burning events [in large parts of the body]”, or “dismemberment, or extreme torture”.
I had to guess the intensity of excruciating pain because there is basically no empirical data informing potential estimates.
I agree my guess is speculative. However, if one wants to argue that I overestimated the cost-effectiveness of SWP, one has to provide reasons for my guess overestimating the intensity of excruciating pain. I do not think claiming the results are unintuitive is a good way of doing this. In the context of global health and development, it would not make much sense to dismiss GiveWell’s conclusion that one can save a life for 5 k$ just because it is unintuitive that one can save 10 lives for the cost of a BMW. Instead, it would be better to critique the inputs that went into the cost-effectiveness analysis. I know you are doing better than this, because you are critiquing intermediate results instead of the final cost-effectiveness. However, I would say it would be better if you focussed on criticising my assumption that excruciating pain is 100 k times as intense as fully healthy life, or, equivalently, the direct implication that 1 day of fully healthy life is neutralised by 0.864 s (= 24*60^2/(100*10^3)) of excruciating pain.
If you did that, SWP would still be 434 (= 43.4*10^3*10^3/(100*10^3)) times as cost-effective as GiveWell’s top charities. I think it is also worth wondering about whether you trully believe that updated intensity. Do you think 1 day of fully healthy life plus 86.4 s (= 0.864*100*10^3/100) of scalding or severe burning events in large parts of the body, dismemberment, or extreme torture would be neutral?
Ah my bad, I meant extreme pain above there as well, edited to clarify! I agree it’s not a super important assumption for the BOTEC in the grand scheme of things though.
I don’t actually argue for this in either of my comments.[1] I’m just saying that it sounds like if I duplicated your BOTEC, and changed this one speculative parameter to 2 OOMs lower, an observer would have no strong reason to choose one BOTEC over another just by looking at the BOTEC alone. Expressing skepticism of an unproven claim doesn’t produce a symmetrical burden of proof on my end!
Mainly just from a reasoning transparency point of view I think it’s worth fleshing out what these assumptions imply and what is grounding these best guesses[2] - in part because I personally want to know how much I should update based on your BOTEC, in part because knowing your reasoning might help me better argue why you might (or might not) have overestimated the intensity of excruciating pain if I knew where your ratio came from (and this is why I was checking the maths and seeing if these were correct, and asking if there’s stronger evidence if so, before critiquing the 100k figure), and because I think other EAF readers, as well as broader, lower-context audience of EA bloggers would benefit from this too.
Yeah, I wasn’t making any inter-charity comparisons or claiming that SWP is less cost-effective than GW top charities![3] But since you mention it, it wouldn’t be surprising to me if losing 2 OOMs might make some donors favour other animal welfare charities over SWP for example—but again, the primary purpose of these comments is not to litigate which charity is the best, or whether this is better or worse than GW top charities, but mainly just to explore a bit more around what is grounding the BOTEC, so observers have a good sense on how much they should update based on how compelling they find the assumptions / reasoning etc.
Nope! I would rather give up 1 day of healthy life than 86 seconds of this description. But this varies depending on the timeframe in question.
For example, I’d probably be willing to endure 0.86 seconds of this for 14 minutes of healthy life, and I would definitely endure 0.086 seconds of this than give up 86 seconds of healthy life.
And using your assumptions (ratio of 100k), I would easily rather have 0.8 seconds of this than give up 1 day of healthy life, but if I had to endure many hours of this I could imagine my tradeoffs approaching, or even exceeding 100k.
I do want to mention that I think it’s useful that someone is trying to quantify these comparisons, I’m grateful for this work, and I want to emphasise that these are about making the underlying reasoning more transparent / understanding the methodology that leads to the assumptions in the BOTEC, rather than any kind of personal criticism!
Though I am personally skeptical of a 50:1 shrimp:human tradeoff
E.g. is this the result of a personal time trade-off exercise?
I explicitly say “To be clear, this isn’t a claim that one shouldn’t donate to SWP”. I’m a big fan of SWP!
Thanks, Bruce! Makes sense. I have now clarified in the post that the guesses for the pain intensities come from my personal time trade-offs:
0.1 s of excruciating pain intuitively passes very quickly, so I can easily imagine it being less than 100 times as bad as 10 s (= 0.1*100) of excruciating pain. However, I think intuitions like this are misguided holding the intensity of pain constant (i.e. only varying duration). For example, if 1 min in strong pain is 100 times as bad as 1 min in weak pain, and strong and weak pain have each a constant intensity (instead of referring to ranges of intensities), N min in strong pain should be 100 times as bad as N min in weak pain. Saying otherwise implies that the badness of 1 min in strong pain depends on how many minutes of strong pain preceded that 1 min, which contradicts the assumption that strong pain has a constant intensity.
I think 0.1 s of excruciating pain does not feel bad because one wrongly imagines the memory of the pain afterwards, which may not be so bad because 0.1 s is too little time to form memories. In addition, one may imagine that some time is needed to reach the intensity of excruciating pain, such that only a small fraction of the 0.1 s are actually excruciating pain, but the thought experiment requires 0.1 s of actual excruciating pain.