Summary

• I tried to improve on Giving What We Can’s (GWWC’s) estimates for the future recorded donations of 10 % pledgers by stipulating the logarithm of the donations per pledger-year for pledges started in the same year evolves linearly (exponential growth; model 1), or quadratically (model 2) with the year of the pledge. I also combine the estimates from models 1 and 2 with a geometric mean weighted by their coefficients of determination (model 3).
• My models lead to a value of a 10 % Pledge and cost-effectiveness of GWWC in 2023 and 2024 very similar to those from their Lifetime Giving Method. I got:
  • A value of a 10 % Pledge of 24.0 k, 21.4 k, and 21.7 k$ of additional donations to highly effective charities, 1.56, 1.39, and 1.41 times GWWC’s estimate from the Lifetime Giving Method.
  • A giving multiplier of 9.22, 8.34, and 8.42, 1.47, 1.33, and 1.34 times GWWC’s estimate from the Lifetime Giving Method.
• I still recommend funding the Centre for Exploratory Altruism Research’s (CEARCH’s) High Impact Philanthropy Fund (HIPF) due its effects on soil animals, which I think are practically proportional to the increase in agricultural-land-years per $. I estimate it increases agricultural land 9.42 times as cost-effectively as GiveWell’s top charities, and 2.65 times as cost-effectively as GWWC in 2023 and 2024 based on my preferred model 3.

Context about GWWC’s estimates

GWWC estimated their impact in 2023 and 2024 in terms of additional donations to highly effective charities, like GiveWell’s top charities, based on 2 methods.

As in our 2020–2022 impact evaluation, we used two complementary methods to estimate the counterfactual impact GWWC generated through pledges in 2023–2024:

• Lifetime Giving Method: Estimate the expected lifetime counterfactual value of donations attributable to each new pledge (by pledge type), then multiply by the number of new pledges acquired during 2023–2024.
• Realised Giving Method: Estimate the counterfactual value of donations made by pledgers in 2023–2024.

[...]

It is important to note that when measured over a long period of time for a single group of pledgers, these two estimates should theoretically be approximately the same. The choice of how to weight the two methods simply reflects when we attribute GWWC credit for the pledge donations that we cause:

• The Realised Giving Method gives GWWC credit at the time the donations occur
• The Lifetime Giving Method gives GWWC credit at the time the pledgers makes the commitment to making the donations (at the time of pledging)

In the Lifetime Giving Method, GWWC calculates the future recorded donations of 10 % pledgers per pledger-year by year of the pledge from the product between:

• The fraction of pledgers recording at least one donation, which decreases exponentially from 59.4 % in the 1st year of the pledge to 27.8 % in the 40th year.
• 8.49 k$ of recorded donations per pledger-year, which is their estimate for pledges started from 2020 to 2023, and pledgers recording at least one donation during this period.
• A discount factor equal to (1 - “annual discount rate”)^(“year of the pledge” - 1)*“pledge is on”, where:
  • “annual discount rate” is 3.50 %.
  • “year of the pledge” starts at 1.
  • “pledge is on” is 1 for “year of the pledge” no larger than 35.1, GWWC’s estimate for the lifespan of a 10 % pledge in years, and 0 otherwise.

GWWC also applied adjustments to account for non-recorded donations, and counterfactuals.

My estimates of future recorded donations

I tried to improve on GWWC’s Lifetime Giving Method as follows:

• For each year s pledges started from 2009 to 2022, I run regressions of the logarithm of the recorded donations D of 10 % pledgers per pledger-year on:
  • The year of the pledge y (model 1). This respects the equation ln(D(s, y)) = a(s) + b(s)*y <=> D(s, y) = e^a(s)*e^(b(s)*y), which stipulates donations per pledger-year evolve exponentially with the year of the pledge.
  • The year of the pledge, and this year squared (model 2). This respects the equation ln(D(s, y)) = A(s) + B(s)*y + C(s)*y^2 <=> D(s, y) = e^A(s)*e^(B(s)*y + C(s)*y^2), which stipulates donations per pledger-year evolve superexponentially with the year of the pledge.
• I obtained the future non-discounted recorded donations per pledger-year from the mean regression parameters weighted by the product between recorded donations and years of data.
• I calculated the future recorded donations per pledger-year multiplying the future non-discounted recorded donations per pledger-year by the discount factor.

Here are my calculations.

I present below the relative annual variation of the recorded donations per pledger-year for model 1, which I determined from D(s, y + 1)/D(s, y) - 1 = e^b(s) - 1. For the mean slope weighted by the importance of the year the pledge started, I got a relative annual variation of the recorded donations per pledger-year of -3.87 %.

The table below has the mean regression parameters, and coefficient of determination (R^2) weighted by the product between recorded donations and years of data. I also combine the estimates from models 1 and 2 with a geometric mean weighted by their coefficients of determination (model 3).

I show below a graph with my and GWWC’s estimates for the future recorded donations per pledger-year.

Cost-effectiveness of GWWC

I made 3 copies of the sheet with the calculations of GWWC’s last impact evaluation, replaced their future non-discounted recorded donations per pledger-year with my estimates for models 1, 2, and 3, and set the weight of the Lifetime Giving Method to 1. I also made a copy of GWWC’s sheet where I simply set the weight of the Lifetime Giving Method to 1.

My models lead to a value of a 10 % Pledge and cost-effectiveness of GWWC in 2023 and 2024 very similar to those from their Lifetime Giving Method. I got:

• A value of a 10 % Pledge of 24.0 k, 21.4 k, and 21.7 k$ of additional donations to highly effective charities, 1.56 (= 24.0*10^3/(15.4*10^3)), 1.39 (= 21.4*10^3/(15.4*10^3)), and 1.41 (= 21.7*10^3/(15.4*10^3)) times GWWC’s estimate from the Lifetime Giving Method.
• A giving multiplier of 9.22, 8.34, and 8.42, 1.47 (= 9.22/6.29), 1.33 (= 8.34/6.29), and 1.34 (= 8.42/6.29) times GWWC’s estimate from the Lifetime Giving Method.

GWWC estimated 45 % of the pledge donations from 2023 to 2024 went to global health and wellbeing, of which 94 % were high-impact donations. So 42.3 % (= 0.45*0.94) of the pledge donations from 2023 to 2024 went to high-impact global health and wellbeing interventions. I believe these are the overwhelming driver of GWWC's cost-effectiveness accounting for effects on soil animals, which I think is practically proportional to the increase in agricultural-land-years per $. In addition, I guess high-impact global health and wellbeing increase agricultural-land-years as cost-effectively as GiveWell’s top charities. As a result, for my preferred model 3, I estimate GWWC in 2023 and 2024 increased agricultural land 3.56 (= 8.42*0.423) times as cost-effectively as GiveWell's top charities.

I still recommend funding HIPF. I estimate this increases agricultural land 9.42 (= 1.29*10^3/137) times as cost-effectively as GiveWell’s top charities, 2.65 (= 9.42/3.56) times as cost-effectively as GWWC in 2023 and 2024 based on my preferred model 3.

Acknowledgements

Thanks to Aidan Whitfield for feedback on the draft. The views expressed in the post are my own.