I chose 0.8 by writing cost-effectiveness estimates of corporate campaigns using 80% credence intervals for the inputs and then calculating the standard deviation of the result. I didn’t quite do it that way for GD and AMF; I tried to estimate the standard deviation from looking at GiveWell’s historical estimates and its the variation in its employees’ current estimates. This was a somewhat rough process but I believe 0.1 and 0.3 are approximately correct.
Also, in your model, QALY improvement is a particularly important cell, but I don’t see much quantitative discussion (though there is some qualitative discussion) in the OpenPhil posts. How did you arrive at your number of 0.5 to 1.5, normally distributed? Do you give any credence to Hsiung’s view that cage-free is net negative toward hens (though disputed heavily by Bollard)? Do you give any credence to the anti-welfarist argument that cage-free has bad long-term effects on creating a society complacent to some level of harm toward animals?
I give some credence to both those things, yes. The anti-welfarist argument doesn’t affect this calculation because this calculation only looks at the direct effects of cage-free campaigns, but it does affect my estimate of the long-term value of the campaigns.
The number for QALY improvement is mostly based on my best guess and other people’s best guesses; it’s hard to say with high accuracy what number we should use.
It means that the chicken’s life is 1 chicken-QALY better per year. There’s a separate figure to adjust chicken sentience to human sentience, where I assume that 1 chicken QALY is worth about 0.3 human QALYs.
Yes, so that means if the difference between cage-free and caged is 1 QALY then it’s as big a difference as between non-existence and healthy life. So like if I were living on a factory farm for a year, and you gave me the option to reduce my lifespan by 1 year but I get to spend my year on a factory farm without a battery cage, that seems like a reasonable deal to me.
No. 1 QALY/year is how good a normal life is. But a life could be better than that, and a life could be more bad than a good life is good. If I’d be willing to give up 10 years of normal life to avert 1 year on a factory farm, then that means a year on a factory farm is worth −10 QALYs.
The Guesstimate model isn’t great, you should look at the one on my spreadsheet instead. My most up-to-date estimate actually has a sigma of 0.56. It’s not actually a standard deviation, it’s the standard deviation of the log-base-10 of the distribution, which means the difference between the mean and one standard deviation above the mean is 0.56 orders of magnitude.
It depends on what you mean by “robust”.
I chose 0.8 by writing cost-effectiveness estimates of corporate campaigns using 80% credence intervals for the inputs and then calculating the standard deviation of the result. I didn’t quite do it that way for GD and AMF; I tried to estimate the standard deviation from looking at GiveWell’s historical estimates and its the variation in its employees’ current estimates. This was a somewhat rough process but I believe 0.1 and 0.3 are approximately correct.
Also, in your model, QALY improvement is a particularly important cell, but I don’t see much quantitative discussion (though there is some qualitative discussion) in the OpenPhil posts. How did you arrive at your number of 0.5 to 1.5, normally distributed? Do you give any credence to Hsiung’s view that cage-free is net negative toward hens (though disputed heavily by Bollard)? Do you give any credence to the anti-welfarist argument that cage-free has bad long-term effects on creating a society complacent to some level of harm toward animals?
I give some credence to both those things, yes. The anti-welfarist argument doesn’t affect this calculation because this calculation only looks at the direct effects of cage-free campaigns, but it does affect my estimate of the long-term value of the campaigns.
The number for QALY improvement is mostly based on my best guess and other people’s best guesses; it’s hard to say with high accuracy what number we should use.
Thanks. Can you elaborate on what a 1 QALY improvement means in this context? Each chicken’s overall life is improved by 1 QALY?
It means that the chicken’s life is 1 chicken-QALY better per year. There’s a separate figure to adjust chicken sentience to human sentience, where I assume that 1 chicken QALY is worth about 0.3 human QALYs.
What does 1 QALY per year mean? Isn’t 1 QALY per year already the difference between non-existence and an ideal, healthy life?
Yes, so that means if the difference between cage-free and caged is 1 QALY then it’s as big a difference as between non-existence and healthy life. So like if I were living on a factory farm for a year, and you gave me the option to reduce my lifespan by 1 year but I get to spend my year on a factory farm without a battery cage, that seems like a reasonable deal to me.
How does the estimate go above 1 QALY/year? Isn’t that the maximum possible?
No. 1 QALY/year is how good a normal life is. But a life could be better than that, and a life could be more bad than a good life is good. If I’d be willing to give up 10 years of normal life to avert 1 year on a factory farm, then that means a year on a factory farm is worth −10 QALYs.
I can see your model of cage-free campaigns here, how do you translate that standard deviation (37K on one run, 80K on another) into 0.8?
The Guesstimate model isn’t great, you should look at the one on my spreadsheet instead. My most up-to-date estimate actually has a sigma of 0.56. It’s not actually a standard deviation, it’s the standard deviation of the log-base-10 of the distribution, which means the difference between the mean and one standard deviation above the mean is 0.56 orders of magnitude.