Thanks for sharing! That’s really interesting. Couple of thoughts:
(1) For us, CEARCH uses n=1 when modelling the value of income doublings, because we’ve tended to prioritize health interventions where the health benefits tend to swamp the economic benefits anyway (and we’ve tended to priortize health interventions because of the heuristic that the NCDs are a big and growing problem which policy can cheaply combat at scale, vs poverty which by the nature of economic growth is declining over time).
There, the modelling is more precise, and we use n=1.26 as a baseline estimate, per Layard, Mayraz and Nickell’s review of a couple of SWB surveys (https://www.sciencedirect.com/science/article/abs/pii/S0047272708000248). Would be interested in hearing how your team arrived at n=1.87 - I presume this is a transformation of an initial n=1 based on your temporal discounts?
Nicolaj correct me if I’m wrong – I think it’s derived here in the OP:
(Quantitatively it would be captured by η=1.87 when combined with the improving circumstances component. That comes from solving the last equation in Rethink Priorities’ 2023 report for η given r=0.026 and g=0.03—i.e., assuming that the compounding non-monetary benefits factor also reflects diminishing marginal utility from income doublings. As a result I’m assuming the discount rate reflects η=1.87 for the remainder of the post.)
That last equation on pg 48 is 𝑟_𝐺𝑖𝑣𝑒𝑊𝑒𝑙𝑙 = (1 + δ)(1 + 𝑔)^(η−1) − 1. δ is the pure time preference rate, for which GiveWell’s choice is δ = 0%; pg 30 in the RP report above summarizes the reasoning behind this choice.
Thank you for outlining what you’re doing at CEARCH—I appreciate it. I’ve put the Layard, Mayraz, and Nickell review on our list of sources to look at as we investigate the right choice of η more. As for where η=1.87 comes from, I saw that Mo already answered that question (thank you!). Let me know if something is still unclear.
Hi Nicolaj,
Thanks for sharing! That’s really interesting. Couple of thoughts:
(1) For us, CEARCH uses n=1 when modelling the value of income doublings, because we’ve tended to prioritize health interventions where the health benefits tend to swamp the economic benefits anyway (and we’ve tended to priortize health interventions because of the heuristic that the NCDs are a big and growing problem which policy can cheaply combat at scale, vs poverty which by the nature of economic growth is declining over time).
(2) The exception is when modelling the counterfactual value of government spending, which a successful policy advocacy intervention redirects, and has to be factored in, albeit at a discount to EA spending, and while taking into account country wealth (https://docs.google.com/spreadsheets/d/1io-4XboFR4BkrKXgfmZHQrlg8MA4Yo_WLZ7Hp6I9Av4/edit?gid=0#gid=0).
There, the modelling is more precise, and we use n=1.26 as a baseline estimate, per Layard, Mayraz and Nickell’s review of a couple of SWB surveys (https://www.sciencedirect.com/science/article/abs/pii/S0047272708000248). Would be interested in hearing how your team arrived at n=1.87 - I presume this is a transformation of an initial n=1 based on your temporal discounts?
Cheers,
Joel
Nicolaj correct me if I’m wrong – I think it’s derived here in the OP:
That last equation on pg 48 is 𝑟_𝐺𝑖𝑣𝑒𝑊𝑒𝑙𝑙 = (1 + δ)(1 + 𝑔)^(η−1) − 1. δ is the pure time preference rate, for which GiveWell’s choice is δ = 0%; pg 30 in the RP report above summarizes the reasoning behind this choice.
Hi Joel,
Thank you for outlining what you’re doing at CEARCH—I appreciate it. I’ve put the Layard, Mayraz, and Nickell review on our list of sources to look at as we investigate the right choice of η more. As for where η=1.87 comes from, I saw that Mo already answered that question (thank you!). Let me know if something is still unclear.
-Nico