Thanks both! We aim for a 5x multiplier on the marginal growth activities we undertake (e.g., pledge booths at EAGs) as well as our average multiplier. On your question @AgentMa🔸, there are a few reasons we aim for a multiplier of 5x as opposed to just >1x, most of which relate to how our multiplier is estimated. As one salient example, a quirk of how we estimate our multiplier is that we categorise the places donors and pledgers give to as either effective or not/unknown in a binary fashion (see our effectiveness coefficient). This is largely for pragmatic reasons (e.g., we aim to be somewhat worldview/cause neutral), but a consequence is that for any given 'worldview' an individual might hold our multiplier is lower than it appears on this 'aggregate' view that treats all 'effective' charities as equal.
Hi Vasco, thanks for the question! On average, our expenditure per 10% Pledge has definitely increased since 2020-2022, a period when our costs were unusually low. It's fair to say that the decline in the giving multiplier since then is largely driven by our costs increasing faster than pledge growth. While we aim to keep our multiplier above 5x, we don't necessarily aspire to have a multiplier of 30x, as we did in the past. A key reason for this is that we think there is probably a tradeoff between cost-effectiveness and scale (i.e., if our multiplier is 30x, there are probably highly cost-effective activities we are leaving on the table).
Hi Vasco, thanks for the question! I think our ideal approach would look like what you suggest: i.e., us doing all our evaluations within each cause area in parallel. This would make it easier for us to be as consistent as possible in the standards we are applying to the evaluators and be more comparative. Unfortunately, in practice this hasn't worked out in the past based on our own capacity constraints and difficulty coordinating evaluations to all occur at the same time. That said, where possible we still try to do at least two parallel evaluations in each cause area (although even this isn't always possible).
Thanks Vasco! I think you are right that we should get a better sense of the marginal multiplier by seeing how the multiplier responds to investments we are making over the next few years. That said, I agree with @GV 🔸 that we might expect some lag between the investments and the returns - particularly for more 'top-of-funnel' work. As a result, our 2026 multiplier might not fully capture the impact of our investments.
Thank you for the very thoughtful comment! As mentioned in the report, we’ve really appreciated the open and thoughtful way you have engaged with us throughout this evaluation. We look forward to seeing you implement your planned improvements and to re-evaluating HLI’s work in the future!
Hi Vasco, thanks for your response! Sorry for my delay in getting back to you, I have just got back from leave. I have tried to leave responses to your main points below, but if I have missed anything please let me know.
Re: our marginal multiplier: This is not something we have explicitly tried to model. The most relevant information I can provide here is that our bar for undertaking new pledge acquisition activities is that they must at least exceed 5x in expectation, but this is still not the same as our marginal multiplier for a number of reasons. While we hope to publish an estimate of our average multiplier for 2025 in early 2026, I currently don’t expect to try to explicitly estimate our marginal multiplier. One key reason for this is that, unless we have identified a very scalable method for growing pledges, our marginal multiplier estimate would change quite quickly as we receive more funding and so may only be relevant for a brief period.
Re: how we report the number of pledgers on the website: Thanks for sharing your thoughts here! I continue to think that the statement on our website is accurate and that it isn’t misleading to use the terms ‘community’ and ‘pledging’ here. Simply, these are the numbers of people in our community who have taken pledges with GWWC. I don’t believe the text makes a claim about the number of pledgers who are reporting their donations (which is not a requirement of the pledge) or the number who are fulfilling their pledge (which we don’t have a reliable estimate of). It isn’t clear to me why we should think that the median person who reads the statement would assume that 90% of pledgers are donating.
Re: modelling pledge value for different cohorts: This kind of regression modelling will be something we continue to consider implementing in future evaluations, but currently it isn’t clear enough to me that these models will be better predictors of future cohort pledge donations than the 'average of recent years' method we currently use. The trends to date have simply been too noisy for me to feel confident in any given mathematical model. I also think these models involve some tradeoffs in terms of time investment and legibility and that we also need to factor in these considerations when selecting our approach.
Re: Trial Pledges: We have not estimated the fraction of impact we attribute to the 🔸10% Pledge that was caused by the 🔹Trial Pledge, but I would roughly guess for recent cohorts it is somewhere in the vicinity of 5–20%. It is difficult to come up with a precise estimate because we don’t know how causally responsible the 🔹Trial Pledge is for the 🔸10% Pledge in these cases (as you note).
Hi Ramiro, thanks for sharing your concerns. In my response to Vasco’s comment, I explain why I don’t think our communications around the number of people who have signed the pledge is misleading. As for whether we take more credit than is due for pledge donations, I want to flag two important ways we try to ensure we aren’t overestimating our impact (among others):
We do not take credit for donations not recorded on the platform unless we have evidence from our surveys that additional pledge donations occurred (see our recording coefficient).
We only take credit for pledge donations that our survey of pledgers indicates we are causally responsible for (see our counterfactuality coefficient). In this impact evaluation, we estimated that we caused 33% of pledge donations.
If this is still your preference, you can resign from your pledge using this form. I hope I have understood your concerns, but please let me know if you have any further questions or concerns.
Hi Vasco, thanks for your engagement! I have put together some responses to your questions/comments below. Please let me know if I missed anything or you have further questions.
> The Centre for Exploratory Altruism Research (CEARCH) estimated GWWC's marginal multiplier to be 17.6 % (= 2.18*10^6/(12.4*10^6)) of GWWC's multiplier. This suggests GWWC's marginal multiplier from 2023 to 2024 was 1.06 (= 0.176*6), such that donating to GWWC over that period was roughly as cost-effective as to GiveWell's top charities. A marginal multiplier of 1 may look bad, but is actually optimal in the sense GWWC should spend more (less) for a marginal multiplier above (below) 1.
I would actually expect our marginal multiplier to be much closer to our average multiplier than the CEARCH method implies. Most importantly, I expect most of our marginal resources are dedicated to identifying and executing on scalable pledge growth strategies. I think this work, in expectation, provides a pretty strong multiplier. By comparison our average multiplier includes some major fixed costs (e.g., related to running our donation platform).
It's also worth noting that pledge growth accelerated between 2023 and 2024, such that our average multiplier for 2024 was roughly 50% higher than that for 2023. In 2025, pledge growth is currently exceeding 2024 growth (by this time in 2024 we had ~280 new 🔸10% Pledges, so far in 2025 we have ~370 new 🔸10% Pledges), although our costs are also higher.
> So I wonder whether the information below on GWWC's website is somewhat misleading.
I don't think I agree that the information on the website is misleading seeing as it just states the number of people who have taken the pledge. I think it’s important to bear in mind that the pledge has never required that pledgers record their donations with GWWC and we know that many of our most engaged pledgers do not record their donations.
> I guess pledges starting in later years are less valuable, such that you are overestimating your impact by not controlling for the year the pledge started.
The regression you suggest is something we have considered, but don’t think it is an obvious improvement over our approach of taking the mean over the most recent pledge years. While there might be an effect of the year the pledge started on average first-year pledge donations, we do not think this trend is linear. For instance, the 2021 cohort had the second highest average first-year donations across all cohorts and the five cohorts with the lowest average first-year donations were 2010, 2017, 2018, 2016 and 2012. Ultimately, this is an empirical question and my prediction is that our average method will be more predictive of the first year of pledge donations for the 2024 cohort than the regression. If you are interested in performing this analysis yourself, you can find total inflation-adjusted pledge donations by pledge cohort and year of pledge in a table in this document.
> Have you considered retiring The Trial Pledge? You estimated 96 % of your impact came from The 10 % Pledge.
We currently aren’t considering retiring the 🔹Trial Pledge. While in terms of direct donation value the 🔹Trial Pledge contributes a relatively small fraction of our pledge impact, we believe the main value add of the 🔹Trial Pledge comes from 🔹Trial pledgers ‘upgrading’ to 🔸10% Pledges. For example, roughly 10% of those who have taken a 🔹Trial Pledge are now 🔸10% Pledges and we are currently exploring ways to improve conversion rates even more. We have also seen some evidence that retention may be stronger for 🔸10% Pledges that follow 🔹Trial Pledges than for other 🔸10% Pledges.
Thanks both! We aim for a 5x multiplier on the marginal growth activities we undertake (e.g., pledge booths at EAGs) as well as our average multiplier. On your question @AgentMa🔸, there are a few reasons we aim for a multiplier of 5x as opposed to just >1x, most of which relate to how our multiplier is estimated. As one salient example, a quirk of how we estimate our multiplier is that we categorise the places donors and pledgers give to as either effective or not/unknown in a binary fashion (see our effectiveness coefficient). This is largely for pragmatic reasons (e.g., we aim to be somewhat worldview/cause neutral), but a consequence is that for any given 'worldview' an individual might hold our multiplier is lower than it appears on this 'aggregate' view that treats all 'effective' charities as equal.
Hi Vasco, thanks for the question! On average, our expenditure per 10% Pledge has definitely increased since 2020-2022, a period when our costs were unusually low. It's fair to say that the decline in the giving multiplier since then is largely driven by our costs increasing faster than pledge growth. While we aim to keep our multiplier above 5x, we don't necessarily aspire to have a multiplier of 30x, as we did in the past. A key reason for this is that we think there is probably a tradeoff between cost-effectiveness and scale (i.e., if our multiplier is 30x, there are probably highly cost-effective activities we are leaving on the table).
Hi Vasco, thanks for the question! I think our ideal approach would look like what you suggest: i.e., us doing all our evaluations within each cause area in parallel. This would make it easier for us to be as consistent as possible in the standards we are applying to the evaluators and be more comparative. Unfortunately, in practice this hasn't worked out in the past based on our own capacity constraints and difficulty coordinating evaluations to all occur at the same time. That said, where possible we still try to do at least two parallel evaluations in each cause area (although even this isn't always possible).
Thanks Vasco! I think you are right that we should get a better sense of the marginal multiplier by seeing how the multiplier responds to investments we are making over the next few years. That said, I agree with @GV 🔸 that we might expect some lag between the investments and the returns - particularly for more 'top-of-funnel' work. As a result, our 2026 multiplier might not fully capture the impact of our investments.
Thanks, Mo! The dashboard is really cool - maybe we should even build something like this for next year's evaluation.
Thanks, Vasco!
Thank you for the very thoughtful comment! As mentioned in the report, we’ve really appreciated the open and thoughtful way you have engaged with us throughout this evaluation. We look forward to seeing you implement your planned improvements and to re-evaluating HLI’s work in the future!
Hi Vasco, thanks for your response! Sorry for my delay in getting back to you, I have just got back from leave. I have tried to leave responses to your main points below, but if I have missed anything please let me know.
Re: our marginal multiplier: This is not something we have explicitly tried to model. The most relevant information I can provide here is that our bar for undertaking new pledge acquisition activities is that they must at least exceed 5x in expectation, but this is still not the same as our marginal multiplier for a number of reasons. While we hope to publish an estimate of our average multiplier for 2025 in early 2026, I currently don’t expect to try to explicitly estimate our marginal multiplier. One key reason for this is that, unless we have identified a very scalable method for growing pledges, our marginal multiplier estimate would change quite quickly as we receive more funding and so may only be relevant for a brief period.
Re: how we report the number of pledgers on the website: Thanks for sharing your thoughts here! I continue to think that the statement on our website is accurate and that it isn’t misleading to use the terms ‘community’ and ‘pledging’ here. Simply, these are the numbers of people in our community who have taken pledges with GWWC. I don’t believe the text makes a claim about the number of pledgers who are reporting their donations (which is not a requirement of the pledge) or the number who are fulfilling their pledge (which we don’t have a reliable estimate of). It isn’t clear to me why we should think that the median person who reads the statement would assume that 90% of pledgers are donating.
Re: modelling pledge value for different cohorts: This kind of regression modelling will be something we continue to consider implementing in future evaluations, but currently it isn’t clear enough to me that these models will be better predictors of future cohort pledge donations than the 'average of recent years' method we currently use. The trends to date have simply been too noisy for me to feel confident in any given mathematical model. I also think these models involve some tradeoffs in terms of time investment and legibility and that we also need to factor in these considerations when selecting our approach.
Re: Trial Pledges: We have not estimated the fraction of impact we attribute to the 🔸10% Pledge that was caused by the 🔹Trial Pledge, but I would roughly guess for recent cohorts it is somewhere in the vicinity of 5–20%. It is difficult to come up with a precise estimate because we don’t know how causally responsible the 🔹Trial Pledge is for the 🔸10% Pledge in these cases (as you note).
Hi Ramiro, thanks for sharing your concerns. In my response to Vasco’s comment, I explain why I don’t think our communications around the number of people who have signed the pledge is misleading. As for whether we take more credit than is due for pledge donations, I want to flag two important ways we try to ensure we aren’t overestimating our impact (among others):
If this is still your preference, you can resign from your pledge using this form. I hope I have understood your concerns, but please let me know if you have any further questions or concerns.
Hi Vasco, thanks for your engagement! I have put together some responses to your questions/comments below. Please let me know if I missed anything or you have further questions.
> The Centre for Exploratory Altruism Research (CEARCH) estimated GWWC's marginal multiplier to be 17.6 % (= 2.18*10^6/(12.4*10^6)) of GWWC's multiplier. This suggests GWWC's marginal multiplier from 2023 to 2024 was 1.06 (= 0.176*6), such that donating to GWWC over that period was roughly as cost-effective as to GiveWell's top charities. A marginal multiplier of 1 may look bad, but is actually optimal in the sense GWWC should spend more (less) for a marginal multiplier above (below) 1.
I would actually expect our marginal multiplier to be much closer to our average multiplier than the CEARCH method implies. Most importantly, I expect most of our marginal resources are dedicated to identifying and executing on scalable pledge growth strategies. I think this work, in expectation, provides a pretty strong multiplier. By comparison our average multiplier includes some major fixed costs (e.g., related to running our donation platform).
It's also worth noting that pledge growth accelerated between 2023 and 2024, such that our average multiplier for 2024 was roughly 50% higher than that for 2023. In 2025, pledge growth is currently exceeding 2024 growth (by this time in 2024 we had ~280 new 🔸10% Pledges, so far in 2025 we have ~370 new 🔸10% Pledges), although our costs are also higher.
> So I wonder whether the information below on GWWC's website is somewhat misleading.
I don't think I agree that the information on the website is misleading seeing as it just states the number of people who have taken the pledge. I think it’s important to bear in mind that the pledge has never required that pledgers record their donations with GWWC and we know that many of our most engaged pledgers do not record their donations.
> I guess pledges starting in later years are less valuable, such that you are overestimating your impact by not controlling for the year the pledge started.
The regression you suggest is something we have considered, but don’t think it is an obvious improvement over our approach of taking the mean over the most recent pledge years. While there might be an effect of the year the pledge started on average first-year pledge donations, we do not think this trend is linear. For instance, the 2021 cohort had the second highest average first-year donations across all cohorts and the five cohorts with the lowest average first-year donations were 2010, 2017, 2018, 2016 and 2012. Ultimately, this is an empirical question and my prediction is that our average method will be more predictive of the first year of pledge donations for the 2024 cohort than the regression. If you are interested in performing this analysis yourself, you can find total inflation-adjusted pledge donations by pledge cohort and year of pledge in a table in this document.
> Have you considered retiring The Trial Pledge? You estimated 96 % of your impact came from The 10 % Pledge.
We currently aren’t considering retiring the 🔹Trial Pledge. While in terms of direct donation value the 🔹Trial Pledge contributes a relatively small fraction of our pledge impact, we believe the main value add of the 🔹Trial Pledge comes from 🔹Trial pledgers ‘upgrading’ to 🔸10% Pledges. For example, roughly 10% of those who have taken a 🔹Trial Pledge are now 🔸10% Pledges and we are currently exploring ways to improve conversion rates even more. We have also seen some evidence that retention may be stronger for 🔸10% Pledges that follow 🔹Trial Pledges than for other 🔸10% Pledges.