CAVEAT : This text has been written by Dirk-Jan Koch, affiliated with the International Institute of Social Studies, Erasmus University, The Hague, The Netherlands, and Martijn Klop, independent researcher and board member of EA Netherlands. For private questions about the article, reach out to : koch@iss.nl. He will read the comments though and is happy to answer!
Abstract: The rapid rise of effective altruism (EA) has reshaped contemporary global health and development by promoting the allocation of resources towards interventions that maximize measurable impact. While widely praised for its analytical rigour and commitment to evidence-based decisionmaking, EA has also attracted critique for its reliance on quantification, its technocratic orientation and its limited engagement with structural forms of inequality. This commentary builds on these debates but advances a more specific argument: that EA systematically underestimates negative unintended effects, resulting in an asymmetrical assessment of impact.
We argue that this asymmetry reflects not only methodological biases—such as the preferential treatment of measurable outcomes— but also a deeper neglect of the distributional consequences of interventions. Framed through the guiding question, ‘effective for whom?’, the commentary examines two prominent EA supported interventions—cash transfers and bed net distribution—to demonstrate how benefits for direct recipients may be accompanied by costs for non-recipients, local systems or public institutions. These costs are often diffuse, delayed or difficult to quantify, and therefore insufficiently incorporated into cost-effectiveness models. We propose three practical shifts: systematically tracking non‑recipient and system indicators, elevating qualitative and participatory evidence, and applying conservative downside adjustments when harms are plausible but hard to quantify. EA deserves a methodology that is capable of seeing the whole picture.
At the beginning of the 20th century, the French colonial administration in Hanoi faced a very real public health crisis: disease spread by rats. Seeking a cost-effective solution, officials introduced a simple idea—pay people a small premium for every rat tail delivered to the authorities. On paper, it looked efficient, measurable and cheap. What they failed to consider—and crucially underestimated—were the side effects of the incentive itself. Instead of eliminating rats, people began cutting off tails and releasing them, or even breeding rats to collect the reward. The policy did not reduce disease; it actively exacerbated the very problem it was meant to solve. This story serves as a reminder that when the focus is narrowly placed on cost-effectiveness without anticipating unintended consequences, well-intentioned solutions can not only backfire— they can become counterproductive.
Over the past decade, the effective altruism (EA) movement has experienced a rapid ascent in influence over the last decade, reshaping contemporary philanthropy by advocating for evidence-based, high-impact giving. EA emphasizes the moral imperative to maximize the positive outcomes of charitable contributions, often prioritizing causes that offer the greatest benefit per unit of resource invested—such as global health, poverty alleviation and existential risk mitigation. This philosophical shift has catalysed the growth of organizations like GiveWell, which meticulously evaluates charities based on cost-effectiveness, transparency and measurable impact, thereby directing philanthropic capital towards interventions with the highest proven efficacy. GiveWell, one of the most prominent EA-aligned organizations, has seen a dramatic increase in funds directed through its recommendations. In 2024, GiveWell directed $400 (300, excluding donations from the private foundation Good Ventures) million in donations to its top-recommended charities, a significant rise from just $25 (15) million in 2014, and the influence of its work extends to other donors who donate based on their recommendations independently. So, while official development assistance (ODA) is facing a bust period, EA-aligned organizations are going through a boom phase (even though it is not reaching levels of ODA, let alone global health spending).
The rise of EA must be understood within a broader global context marked by persistent inequality, labour exploitation and the uneven distribution of the gains and costs of global economic integration. Much of the suffering that EA seeks to alleviate is not incidental, but structurally produced through global commodity chains, financial flows and political-economic arrangements. Within this context, EA’s focus on optimizing the allocation of charitable resources risks operating downstream of these dynamics rather than engaging with their root causes. In this sense, EA’s emphasis on cost-effectiveness may not only narrow how impact is measured, but also where responsibility is located—shifting attention towards the efficient mitigation of harm rather than its structural production. While this commentary does not seek to resolve this broader critique, it highlights a related and more immediate implication: even within its own terms, EA’s assessments remain incomplete.
Another growing strand of scholarship pushes back against EA by arguing that it neglects issues systematic biases. Three types of biases have been most reported, especially in global development: measurement, quantification and instrumental biases. Measurement biases imply that fields like public health, about which there is high-quality information, will receive more funding because in these areas, it will be easier to differentiate very good from very bad interventions. This tendency to focus disproportionately on what is known, or readily verifiable, can hence systematically skew priorities and lead to certain forms of bias and error. A second bias is called quantification bias. This materializes when the metrics used to evaluate interventions mean that it is the only thing that counts in practice. The charge of quantification bias focuses primarily upon the movement’s reliance, in its early years, on the disability-adjusted life years (DALYs) metric. Analysis conducted in these terms, therefore, tends to exclude other values, such as subjective well-being. The final methodological charge, that effective altruism is unduly instrumental, holds that it fails to take seriously the effect of politics on outcomes and tends to favour technocratic rather than systemic solutions to developmental problems (Berke, 2026). These critiques also resonate with a broader line of thought that situates EA within the rise of what Dardot and Laval (2010) describe as the ‘neoliberal individual’—a subject who is not only embedded in markets, but is required to think and act as an optimizer of outcomes. From this perspective, EA’s emphasis on maximizing impact, comparing interventions and allocating resources efficiently reflects not only a methodological choice, but a wider rationality that frames moral action in terms of calculation, performance and optimization.
At the same time, proponents of EA argue that precisely these features—its emphasis on measurement, comparability and optimization— are what make it uniquely capable of improving resource allocation at scale. It is precisely this tension—between a framework that is analytically powerful and one that is potentially system-blind—that helps explain both EA’s rise and its limits. This commentary builds on that debate by asking a question that cuts across both camps but remains insufficiently examined: effective for whom? Interventions that appear highly cost-effective for direct beneficiaries may produce costs for non-beneficiaries, for local systems or for governments—costs that are insufficiently captured in current models. While large bilateral donors, such as GIZ, are increasingly focusing on unintended effects (Rana et al., 2025)—also the negative ones—such a systematic approach is largely lacking in the EA movement. EA is rightly praised for its open culture of debate and continuous improvement. Yet, despite this openness, its treatment of unintended effects in cost-effectiveness methods remains strikingly unbalanced.
From Narrow Metrics to Broader Impact: From Positive Spillovers to Undercounted Negative Side Effects
This commentary begins with the upside: spill-over effects are now part of the analytical map. EA used to focus narrowly on the direct, short-term outcomes that randomized controlled trials (RCTs) make easiest to observe: lives saved, infections averted, income increases. Over the last decade, the toolkit has broadened to capture spill-over effects—the ways an intervention changes local markets, empowers certain people who then help others or creates multiplier effects across households and communities. GiveWell’s own writing on spillovers and its sustained interest in cash transfers provide clear evidence of this shift (GiveWell, 2024b).
However, the same intellectual machinery that helped identify positive spillovers has not been applied with comparable rigour to negative spillovers and unintended harms. The resulting asymmetry is not benign; it can reverse priorities and produce programmes that are net-harmful in specific contexts.
Biases that critics of the EA movement have levelled in general appear particularly applicable with respect to negative unintended effects. Measurement bias is especially relevant here: harms are often diffuse, delayed or concentrated among non-recipients— which makes them inherently harder to capture in short RCTs or in single-indicator cost-effectiveness models. Quantification bias reinforces this problem. Donor incentives and headline metrics reward visible, immediate benefits and low administrative ratios; they do not reward the careful measurement of political, social or system-level harms.
Below are two concrete examples where this asymmetry becomes visible: cash transfers and bed-net distribution, two of the flagship interventions of the EA community. Both examples make the same underlying issue visible: cost-effectiveness depends on perspective, and the current EA framework does not consistently ask effective for whom.
Cash Transfers Reconsidered: When Positive Narratives Obscure Systemic Costs
A landmark study by Angelucci and De Giorgi (2009) was the first to look at what they call ‘indirect effects’ (i.e., unintended effects of 4 Effective Altruism’s Blind Spot: The Systematic Underestimation of Harm Progress in Development Studies (2026) pp. 1–7 cash transfers on non-beneficiaries). Their findings were positive: an increase of 100 pesos to recipient households leads to an increase in consumption of 11 pesos amongst non-recipient households. It turns out that non-participants received more gifts and loans from those receiving money from the programme. The programme reduced the ‘price’ of helping other people financially, so the entire community benefited in ways that were initially overlooked. There is also a multiplier effect: those who receive the money spend it on goods and services, and that those selling these goods and services will also buy additional goods and services, setting in motion a virtuous circle, at least in theory (although, as will become clear, this dynamic does not always play out as expected). At any rate, to obtain a complete picture, Angelucci and De Giorgi (2009) propose that cash transfer programme results should be measured at the community level rather than the individual one—a recommendation that fundamentally challenges narrow evaluation designs.
The World Bank did not initially take this advice on board when it carried out an impact assessment of a cash transfer programme in the Philippines that it had helped to develop. After a couple of years of intervention, the first positive results were starting to show up in the health of children whose parents were part of the programme. The children of participants saw an impressive 40% reduction in stunting. In a somewhat self-congratulatory randomized control evaluation, the World Bank evaluators concluded that: ‘the findings of the impact evaluation support administrative and other assessments that have found that Pantawid Pamilya is achieving most of its key objectives, with an additional four out of ten children following the standard lines’ (World Bank, 2015). Crucially, they stopped their assessment there, and the programme was therefore framed as an undisputed success.
But the reality was considerably messier than this impact evaluation suggested. Localized inflation, a delayed response of government-wide social programmes, and jealousy are potential side effects that receive only scant—and arguably insufficient— attention in cost-effectiveness calculations.
Other World Bank researchers (Filmer et al., 2018) went back and followed the advice of Angelucci and De Giorgi (2009), performing a more holistic analysis that included the effects on non-participants. The results were striking. Children in non-participant families saw a 34% increase in stunting. In other words, the positive effects for the participant group were almost completely offset by negative effects for the non-participant group. What appears effective at the level of recipients is far less so once non-recipients are included in the frame. How did that happen? Stunting increased the most in villages that were relatively remote and where many residents were receiving cash benefits. The price of food had risen significantly: eggs, for instance, had increased by 25%. Other foodstuffs, such as rice, had also become so expensive that non-participants had been forced to consume less. Apparently, there had not been a sufficient supply response (e.g., more people raising chickens), or that response took too long to materialize. This is a clear example of an unintended price effect that was not captured by the initial evaluation due to its narrow research design. While GiveWell has paid attention to localized inflation and acknowledges that in remote areas this can be a serious issue—arguing that ‘a conservative adjustment seems appropriate’ (2024a)—it still does not systematically integrate these dynamics into its cost-effectiveness models.
Crowding out governmental social safety programmes is another potential negative side effect that is often acknowledged but Koch and Klop 5 Progress in Development Studies (2026) pp. 1–7 insufficiently taken seriously. GiveWell explicitly raises this question: ‘Could GiveDirectly delay the scale-up of national social security programs?’ (GiveWell, 2025). However, the answer provided is strikingly limited: ‘We think the question of whether NGO programming crowds out state programming is an important one, but we don’t think it’s especially pertinent to GiveDirectly compared to some of the other programs we recommend’. In other words, the issue is recognized, but not analytically pursued or incorporated into decision-making.
There are also other non-financial negative side effects of cash transfers, such as jealousy and social friction. Targeted programmes that give money to some households while excluding neighbours can generate resentment, social tension and reductions in informal mutual aid. Recipients report low but significantly higher (38%) levels of hostility than controls (GiveWell, 2018a). In this case, a small correction is applied: a 2% discount is included in cost-effectiveness estimates to account for potential negative psychological spillovers (GiveWell, 2025). While it is commendable that such effects are acknowledged, the magnitude of this adjustment raises questions. Given the potential scale and persistence of these social dynamics, a 2% correction appears conservative at best. It would be more convincing if additional resources were invested in determining which types of cash transfer programmes minimize such effects and if these findings were more rigorously integrated into cost-effectiveness assessments.
Bed Nets and Hidden Trade-offs: The System-level Costs That Remain Invisible
Let us now turn to bed nets, which represent a classic vertical health intervention. Insecticidetreated bed nets have saved lives and are rightly celebrated by EA and GiveWell. However, as with other forms of vertical programming, the question is less whether bed nets are effective, and more whose outcomes are being counted—and whose are not.
One such effect is the distortion of labour markets within health systems. Large, wellfunded vertical programmes—and the NGOs that implement them—often offer better pay and working conditions than public-sector positions. This can generate a brain drain from the public sector, drawing doctors, nurses and programme managers away from routine services and into project-based work. The consequence is a reduction in the availability and quality of general care. This phenomenon is not speculative; multiple country-level studies have documented precisely these effects (GiveWell, 2023; Mussa et al., 2013). Yet, despite this evidence, such dynamics are not systematically incorporated into cost-effectiveness calculations.
A second side effect concerns resource allocation within health systems. Vertical funding streams, which are tightly earmarked for specific interventions, can limit the flexibility of health systems and reduce investment in core capacities such as administration, logistics and civil-service development. This can weaken the overall resilience of health systems. For example, statistical analysis by Chima and Franzini (2016) shows that a one-dollar increase in HIV aid per capita in Nigeria was associated with a decrease in the probability of receiving routine vaccinations by between 8% and 31%. This represents a clear trade-off, yet it is rarely accounted for in headline effectiveness metrics.
A third issue is the externalization of costs. Programmes that are celebrated for their low overhead—such as large-scale bed-net distribution campaigns costing only a few dollars per unit—often rely on local or national systems to absorb the follow-up costs. These include distribution logistics, replacement, community engagement and waste management.
When these costs are not fully accounted for, the intervention appears more efficient than it actually is. GiveWell does acknowledge these dynamics, noting that for every $1 million spent on bed nets, the Ugandan government incurs approximately $47,000 in in-kind costs (GiveWell, 2023). However, they also argue that the resources governments contribute would have been used less efficiently elsewhere, often assuming a fourfold difference in effectiveness (GiveWell, 2018b). This is a strong—and arguably contestable—assumption.
Indeed, research on fungibility suggests that such assumptions reflect a deeper issue: the belief that external actors can allocate resources more effectively than domestic institutions. This has been criticized as reflecting a neo-colonial bias in development thinking (Rana and Koch, 2020). While this critique may be uncomfortable, it highlights the importance of treating local decision-making capacity as analytically relevant rather than implicitly inferior.
Taken together, these dynamics suggest that bed-net interventions—while highly effective in the short term—may have longerterm system-level consequences that are insufficiently captured in current models. Ignoring these effects risks overstating their net impact.
2. Scaling Impact, Scaling Blind Spots: Why Effective Altruism Must Confront Its Own Methodological Limits
What this commentary has shown is a failure to consistently ask a basic question: effective for whom? EA’s analytical framework remains strongest at capturing benefits for direct recipients, but systematically weaker at accounting for costs borne by others—non-recipients, local systems and governments. The EA ecosystem has developed increasingly sophisticated methods for identifying and quantifying positive spillovers, yet the same methodological rigour is not consistently applied to negative side effects. This imbalance risks systematically skewing decision-making. It risks overstating the net value of interventions that generate visible benefits while producing less visible harms. And as EA organizations expand their influence, these asymmetries have tangible consequences: they shape funding decisions, policy dynamics and real-world outcomes.
Addressing this issue requires the EA framework extending it beyond its current methodological comfort zone: the tools that capture upside do not automatically capture downside.
First, monitoring frameworks should be expanded to include non-recipient and system-level indicators on a routine basis. Capturing such effects often requires moving beyond standard RCT-based designs towards approaches that can detect system-level and distributive dynamics. Outcomes such as social cohesion, price effects and institutional dynamics are measurable and should not be treated as peripheral.
Second, qualitative and participatory methods are not just complements but necessary tools for identifying forms of harm that resist quantification. Ethnographic research, community feedback mechanisms and grievance monitoring can capture important dynamics that standardized metrics overlook. These approaches can be integrated without sacrificing rigour.
Third, cost-effectiveness models should incorporate explicit adjustments for downside risk. If certain harms are difficult to quantify but plausibly large, they should be incorporated through conservative assumptions rather than excluded entirely. GiveWell already applies conservative adjustments in some areas; extending this principle to negative side effects would be consistent with its broader methodology.
To conclude, EA’s methodological strength lies in its attention to evidence, modelling and trade-offs. Its own history demonstrates a capacity for self-correction—from focusing narrowly on direct outcomes to incorporating spillover effects, and from high initial confidence to more nuanced reassessments. The next logical step is clear: apply the same level of analytical ambition used to identify upside, while recognizing that downside often requires different kinds of evidence, including qualitative, political and system-level analysis. Only then will EA’s framework be capable of capturing the full consequences of the interventions it promotes.
Acknowledgement
The authors would like to thank David Nash and an anonymous reviewer for useful comments on earlier drafts of this viewpoint.
Authors’ Affiliation Dirk-Jan Koch is the corre sponding author (koch@iss.nl) and is affiliated with the International Institute of Social Studies, Erasmus University, The Hague, The Netherlands.
Martijn Klop is an independent researcher. Declaration of Conflicting Interests
Martijn Klop is a board member of Effective Altruism in the Netherlands. He has written this viewpoint with total academic freedom.
Funding The authors received no financial support for the research, authorship and/or publication of this article.
References
Angelucci, M. and De Giorgi, G. 2009: Indirect effects of an aid program: How do cash transfers affect ineligibles’ consumption? American Economic Review 99(1), 486–508. https://doi.org/10.1257/ aer.99.1.486
Berke. 2026: Effective charity ≠ effective development (global health edition). EA Forum. https://forum. effectivealtruism.org/posts/p7EbDRK4FCs2795uh/ effective-charity-effective-development-globalhealth
Chima, C. and Franzini, L. 2016: Spillover effect of HIV-specific foreign aid on immunization services in Nigeria. International Health 8(2), 108–15. https://doi. org/10.1093/inthealth/ihv036
Dardot, P. and Laval, C. 2013 [2010]: The new way of the world: On neoliberal society. Verso.
Filmer, D.P., Friedman, J., Kandpal, E., et al. 2018: Cash transfers, food prices, and nutrition impacts on nonbeneficiary children [Policy Research Working Paper 8377]. World Bank. https://documents.worldbank.org/en/publication/ documents-reports/documentdetail/98903152 2077749796/
Mussa, A.H., Pfeiffer, J., Gloyd, S.S., et al. 2013: Vertical funding, non-governmental organizations, and health system strengthening: Perspectives of public-sector health workers in Mozambique. Human Resources for Health 11(1), 26. https://doi. org/10.1186/1478-4491-11-26
Rana, Z., Koch, D.J., and Diem, C. 2025: Progress and pitfalls when evaluating the unintended effects of public policy: The case of German international development cooperation. Evaluation 31(4), 559–79.
World Bank. 2015: Philippines conditional cash transfer program impact evaluation 2012. https://hdl.handle. net/10986/13244
Effective Altruism’s Blind Spot: The Systematic Underestimation of Harm, by Dirk-Jan Koch and Martijn Klop
CAVEAT : This text has been written by Dirk-Jan Koch, affiliated with the International Institute of Social Studies, Erasmus University, The Hague, The Netherlands, and Martijn Klop, independent researcher and board member of EA Netherlands. For private questions about the article, reach out to : koch@iss.nl. He will read the comments though and is happy to answer!
Abstract: The rapid rise of effective altruism (EA) has reshaped contemporary global health and development by promoting the allocation of resources towards interventions that maximize measurable impact. While widely praised for its analytical rigour and commitment to evidence-based decisionmaking, EA has also attracted critique for its reliance on quantification, its technocratic orientation and its limited engagement with structural forms of inequality. This commentary builds on these debates but advances a more specific argument: that EA systematically underestimates negative unintended effects, resulting in an asymmetrical assessment of impact.
We argue that this asymmetry reflects not only methodological biases—such as the preferential treatment of measurable outcomes— but also a deeper neglect of the distributional consequences of interventions. Framed through the guiding question, ‘effective for whom?’, the commentary examines two prominent EA supported interventions—cash transfers and bed net distribution—to demonstrate how benefits for direct recipients may be accompanied by costs for non-recipients, local systems or public institutions. These costs are often diffuse, delayed or difficult to quantify, and therefore insufficiently incorporated into cost-effectiveness models. We propose three practical shifts: systematically tracking non‑recipient and system indicators, elevating qualitative and participatory evidence, and applying conservative downside adjustments when harms are plausible but hard to quantify. EA deserves a methodology that is capable of seeing the whole picture.
Key words: Effective altruism, Cost-effectiveness analysis, Unintended consequences, Distributional inequality, Evidence-based decision-making.
At the beginning of the 20th century, the French colonial administration in Hanoi faced a very real public health crisis: disease spread by rats. Seeking a cost-effective solution, officials introduced a simple idea—pay people a small premium for every rat tail delivered to the authorities. On paper, it looked efficient, measurable and cheap. What they failed to consider—and crucially underestimated—were the side effects of the incentive itself. Instead of eliminating rats, people began cutting off tails and releasing them, or even breeding rats to collect the reward. The policy did not reduce disease; it actively exacerbated the very problem it was meant to solve. This story serves as a reminder that when the focus is narrowly placed on cost-effectiveness without anticipating unintended consequences, well-intentioned solutions can not only backfire— they can become counterproductive.
Over the past decade, the effective altruism (EA) movement has experienced a rapid ascent in influence over the last decade, reshaping contemporary philanthropy by advocating for evidence-based, high-impact giving. EA emphasizes the moral imperative to maximize the positive outcomes of charitable contributions, often prioritizing causes that offer the greatest benefit per unit of resource invested—such as global health, poverty alleviation and existential risk mitigation. This philosophical shift has catalysed the growth of organizations like GiveWell, which meticulously evaluates charities based on cost-effectiveness, transparency and measurable impact, thereby directing philanthropic capital towards interventions with the highest proven efficacy. GiveWell, one of the most prominent EA-aligned organizations, has seen a dramatic increase in funds directed through its recommendations. In 2024, GiveWell directed $400 (300, excluding donations from the private foundation Good Ventures) million in donations to its top-recommended charities, a significant rise from just $25 (15) million in 2014, and the influence of its work extends to other donors who donate based on their recommendations independently. So, while official development assistance (ODA) is facing a bust period, EA-aligned organizations are going through a boom phase (even though it is not reaching levels of ODA, let alone global health spending).
The rise of EA must be understood within a broader global context marked by persistent inequality, labour exploitation and the uneven distribution of the gains and costs of global economic integration. Much of the suffering that EA seeks to alleviate is not incidental, but structurally produced through global commodity chains, financial flows and political-economic arrangements. Within this context, EA’s focus on optimizing the allocation of charitable resources risks operating downstream of these dynamics rather than engaging with their root causes. In this sense, EA’s emphasis on cost-effectiveness may not only narrow how impact is measured, but also where responsibility is located—shifting attention towards the efficient mitigation of harm rather than its structural production. While this commentary does not seek to resolve this broader critique, it highlights a related and more immediate implication: even within its own terms, EA’s assessments remain incomplete.
Another growing strand of scholarship pushes back against EA by arguing that it neglects issues systematic biases. Three types of biases have been most reported, especially in global development: measurement, quantification and instrumental biases. Measurement biases imply that fields like public health, about which there is high-quality information, will receive more funding because in these areas, it will be easier to differentiate very good from very bad interventions. This tendency to focus disproportionately on what is known, or readily verifiable, can hence systematically skew priorities and lead to certain forms of bias and error. A second bias is called quantification bias. This materializes when the metrics used to evaluate interventions mean that it is the only thing that counts in practice. The charge of quantification bias focuses primarily upon the movement’s reliance, in its early years, on the disability-adjusted life years (DALYs) metric. Analysis conducted in these terms, therefore, tends to exclude other values, such as subjective well-being. The final methodological charge, that effective altruism is unduly instrumental, holds that it fails to take seriously the effect of politics on outcomes and tends to favour technocratic rather than systemic solutions to developmental problems (Berke, 2026). These critiques also resonate with a broader line of thought that situates EA within the rise of what Dardot and Laval (2010) describe as the ‘neoliberal individual’—a subject who is not only embedded in markets, but is required to think and act as an optimizer of outcomes. From this perspective, EA’s emphasis on maximizing impact, comparing interventions and allocating resources efficiently reflects not only a methodological choice, but a wider rationality that frames moral action in terms of calculation, performance and optimization.
At the same time, proponents of EA argue that precisely these features—its emphasis on measurement, comparability and optimization— are what make it uniquely capable of improving resource allocation at scale. It is precisely this tension—between a framework that is analytically powerful and one that is potentially system-blind—that helps explain both EA’s rise and its limits. This commentary builds on that debate by asking a question that cuts across both camps but remains insufficiently examined: effective for whom? Interventions that appear highly cost-effective for direct beneficiaries may produce costs for non-beneficiaries, for local systems or for governments—costs that are insufficiently captured in current models. While large bilateral donors, such as GIZ, are increasingly focusing on unintended effects (Rana et al., 2025)—also the negative ones—such a systematic approach is largely lacking in the EA movement. EA is rightly praised for its open culture of debate and continuous improvement. Yet, despite this openness, its treatment of unintended effects in cost-effectiveness methods remains strikingly unbalanced.
From Narrow Metrics to Broader Impact: From Positive Spillovers to Undercounted Negative Side Effects
This commentary begins with the upside: spill-over effects are now part of the analytical map. EA used to focus narrowly on the direct, short-term outcomes that randomized controlled trials (RCTs) make easiest to observe: lives saved, infections averted, income increases. Over the last decade, the toolkit has broadened to capture spill-over effects—the ways an intervention changes local markets, empowers certain people who then help others or creates multiplier effects across households and communities. GiveWell’s own writing on spillovers and its sustained interest in cash transfers provide clear evidence of this shift (GiveWell, 2024b).
However, the same intellectual machinery that helped identify positive spillovers has not been applied with comparable rigour to negative spillovers and unintended harms. The resulting asymmetry is not benign; it can reverse priorities and produce programmes that are net-harmful in specific contexts.
Biases that critics of the EA movement have levelled in general appear particularly applicable with respect to negative unintended effects. Measurement bias is especially relevant here: harms are often diffuse, delayed or concentrated among non-recipients— which makes them inherently harder to capture in short RCTs or in single-indicator cost-effectiveness models. Quantification bias reinforces this problem. Donor incentives and headline metrics reward visible, immediate benefits and low administrative ratios; they do not reward the careful measurement of political, social or system-level harms.
Below are two concrete examples where this asymmetry becomes visible: cash transfers and bed-net distribution, two of the flagship interventions of the EA community. Both examples make the same underlying issue visible: cost-effectiveness depends on perspective, and the current EA framework does not consistently ask effective for whom.
Cash Transfers Reconsidered: When Positive Narratives Obscure Systemic Costs
A landmark study by Angelucci and De Giorgi (2009) was the first to look at what they call ‘indirect effects’ (i.e., unintended effects of 4 Effective Altruism’s Blind Spot: The Systematic Underestimation of Harm Progress in Development Studies (2026) pp. 1–7 cash transfers on non-beneficiaries). Their findings were positive: an increase of 100 pesos to recipient households leads to an increase in consumption of 11 pesos amongst non-recipient households. It turns out that non-participants received more gifts and loans from those receiving money from the programme. The programme reduced the ‘price’ of helping other people financially, so the entire community benefited in ways that were initially overlooked. There is also a multiplier effect: those who receive the money spend it on goods and services, and that those selling these goods and services will also buy additional goods and services, setting in motion a virtuous circle, at least in theory (although, as will become clear, this dynamic does not always play out as expected). At any rate, to obtain a complete picture, Angelucci and De Giorgi (2009) propose that cash transfer programme results should be measured at the community level rather than the individual one—a recommendation that fundamentally challenges narrow evaluation designs.
The World Bank did not initially take this advice on board when it carried out an impact assessment of a cash transfer programme in the Philippines that it had helped to develop. After a couple of years of intervention, the first positive results were starting to show up in the health of children whose parents were part of the programme. The children of participants saw an impressive 40% reduction in stunting. In a somewhat self-congratulatory randomized control evaluation, the World Bank evaluators concluded that: ‘the findings of the impact evaluation support administrative and other assessments that have found that Pantawid Pamilya is achieving most of its key objectives, with an additional four out of ten children following the standard lines’ (World Bank, 2015). Crucially, they stopped their assessment there, and the programme was therefore framed as an undisputed success.
But the reality was considerably messier than this impact evaluation suggested. Localized inflation, a delayed response of government-wide social programmes, and jealousy are potential side effects that receive only scant—and arguably insufficient— attention in cost-effectiveness calculations.
Other World Bank researchers (Filmer et al., 2018) went back and followed the advice of Angelucci and De Giorgi (2009), performing a more holistic analysis that included the effects on non-participants. The results were striking. Children in non-participant families saw a 34% increase in stunting. In other words, the positive effects for the participant group were almost completely offset by negative effects for the non-participant group. What appears effective at the level of recipients is far less so once non-recipients are included in the frame. How did that happen? Stunting increased the most in villages that were relatively remote and where many residents were receiving cash benefits. The price of food had risen significantly: eggs, for instance, had increased by 25%. Other foodstuffs, such as rice, had also become so expensive that non-participants had been forced to consume less. Apparently, there had not been a sufficient supply response (e.g., more people raising chickens), or that response took too long to materialize. This is a clear example of an unintended price effect that was not captured by the initial evaluation due to its narrow research design. While GiveWell has paid attention to localized inflation and acknowledges that in remote areas this can be a serious issue—arguing that ‘a conservative adjustment seems appropriate’ (2024a)—it still does not systematically integrate these dynamics into its cost-effectiveness models.
Crowding out governmental social safety programmes is another potential negative side effect that is often acknowledged but Koch and Klop 5 Progress in Development Studies (2026) pp. 1–7 insufficiently taken seriously. GiveWell explicitly raises this question: ‘Could GiveDirectly delay the scale-up of national social security programs?’ (GiveWell, 2025). However, the answer provided is strikingly limited: ‘We think the question of whether NGO programming crowds out state programming is an important one, but we don’t think it’s especially pertinent to GiveDirectly compared to some of the other programs we recommend’. In other words, the issue is recognized, but not analytically pursued or incorporated into decision-making.
There are also other non-financial negative side effects of cash transfers, such as jealousy and social friction. Targeted programmes that give money to some households while excluding neighbours can generate resentment, social tension and reductions in informal mutual aid. Recipients report low but significantly higher (38%) levels of hostility than controls (GiveWell, 2018a). In this case, a small correction is applied: a 2% discount is included in cost-effectiveness estimates to account for potential negative psychological spillovers (GiveWell, 2025). While it is commendable that such effects are acknowledged, the magnitude of this adjustment raises questions. Given the potential scale and persistence of these social dynamics, a 2% correction appears conservative at best. It would be more convincing if additional resources were invested in determining which types of cash transfer programmes minimize such effects and if these findings were more rigorously integrated into cost-effectiveness assessments.
Bed Nets and Hidden Trade-offs: The System-level Costs That Remain Invisible
Let us now turn to bed nets, which represent a classic vertical health intervention. Insecticidetreated bed nets have saved lives and are rightly celebrated by EA and GiveWell. However, as with other forms of vertical programming, the question is less whether bed nets are effective, and more whose outcomes are being counted—and whose are not.
One such effect is the distortion of labour markets within health systems. Large, wellfunded vertical programmes—and the NGOs that implement them—often offer better pay and working conditions than public-sector positions. This can generate a brain drain from the public sector, drawing doctors, nurses and programme managers away from routine services and into project-based work. The consequence is a reduction in the availability and quality of general care. This phenomenon is not speculative; multiple country-level studies have documented precisely these effects (GiveWell, 2023; Mussa et al., 2013). Yet, despite this evidence, such dynamics are not systematically incorporated into cost-effectiveness calculations.
A second side effect concerns resource allocation within health systems. Vertical funding streams, which are tightly earmarked for specific interventions, can limit the flexibility of health systems and reduce investment in core capacities such as administration, logistics and civil-service development. This can weaken the overall resilience of health systems. For example, statistical analysis by Chima and Franzini (2016) shows that a one-dollar increase in HIV aid per capita in Nigeria was associated with a decrease in the probability of receiving routine vaccinations by between 8% and 31%. This represents a clear trade-off, yet it is rarely accounted for in headline effectiveness metrics.
A third issue is the externalization of costs. Programmes that are celebrated for their low overhead—such as large-scale bed-net distribution campaigns costing only a few dollars per unit—often rely on local or national systems to absorb the follow-up costs. These include distribution logistics, replacement, community engagement and waste management.
When these costs are not fully accounted for, the intervention appears more efficient than it actually is. GiveWell does acknowledge these dynamics, noting that for every $1 million spent on bed nets, the Ugandan government incurs approximately $47,000 in in-kind costs (GiveWell, 2023). However, they also argue that the resources governments contribute would have been used less efficiently elsewhere, often assuming a fourfold difference in effectiveness (GiveWell, 2018b). This is a strong—and arguably contestable—assumption.
Indeed, research on fungibility suggests that such assumptions reflect a deeper issue: the belief that external actors can allocate resources more effectively than domestic institutions. This has been criticized as reflecting a neo-colonial bias in development thinking (Rana and Koch, 2020). While this critique may be uncomfortable, it highlights the importance of treating local decision-making capacity as analytically relevant rather than implicitly inferior.
Taken together, these dynamics suggest that bed-net interventions—while highly effective in the short term—may have longerterm system-level consequences that are insufficiently captured in current models. Ignoring these effects risks overstating their net impact.
2. Scaling Impact, Scaling Blind Spots: Why Effective Altruism Must Confront Its Own Methodological Limits
What this commentary has shown is a failure to consistently ask a basic question: effective for whom? EA’s analytical framework remains strongest at capturing benefits for direct recipients, but systematically weaker at accounting for costs borne by others—non-recipients, local systems and governments. The EA ecosystem has developed increasingly sophisticated methods for identifying and quantifying positive spillovers, yet the same methodological rigour is not consistently applied to negative side effects. This imbalance risks systematically skewing decision-making. It risks overstating the net value of interventions that generate visible benefits while producing less visible harms. And as EA organizations expand their influence, these asymmetries have tangible consequences: they shape funding decisions, policy dynamics and real-world outcomes.
Addressing this issue requires the EA framework extending it beyond its current methodological comfort zone: the tools that capture upside do not automatically capture downside.
First, monitoring frameworks should be expanded to include non-recipient and system-level indicators on a routine basis. Capturing such effects often requires moving beyond standard RCT-based designs towards approaches that can detect system-level and distributive dynamics. Outcomes such as social cohesion, price effects and institutional dynamics are measurable and should not be treated as peripheral.
Second, qualitative and participatory methods are not just complements but necessary tools for identifying forms of harm that resist quantification. Ethnographic research, community feedback mechanisms and grievance monitoring can capture important dynamics that standardized metrics overlook. These approaches can be integrated without sacrificing rigour.
Third, cost-effectiveness models should incorporate explicit adjustments for downside risk. If certain harms are difficult to quantify but plausibly large, they should be incorporated through conservative assumptions rather than excluded entirely. GiveWell already applies conservative adjustments in some areas; extending this principle to negative side effects would be consistent with its broader methodology.
To conclude, EA’s methodological strength lies in its attention to evidence, modelling and trade-offs. Its own history demonstrates a capacity for self-correction—from focusing narrowly on direct outcomes to incorporating spillover effects, and from high initial confidence to more nuanced reassessments. The next logical step is clear: apply the same level of analytical ambition used to identify upside, while recognizing that downside often requires different kinds of evidence, including qualitative, political and system-level analysis. Only then will EA’s framework be capable of capturing the full consequences of the interventions it promotes.
Acknowledgement
The authors would like to thank David Nash and an anonymous reviewer for useful comments on earlier drafts of this viewpoint.
Authors’ Affiliation Dirk-Jan Koch is the corre sponding author (koch@iss.nl) and is affiliated with the International Institute of Social Studies, Erasmus University, The Hague, The Netherlands.
Martijn Klop is an independent researcher. Declaration of Conflicting Interests
Martijn Klop is a board member of Effective Altruism in the Netherlands. He has written this viewpoint with total academic freedom.
Funding The authors received no financial support for the research, authorship and/or publication of this article.
References
Angelucci, M. and De Giorgi, G. 2009: Indirect effects of an aid program: How do cash transfers affect ineligibles’ consumption? American Economic Review 99(1), 486–508. https://doi.org/10.1257/ aer.99.1.486
Berke. 2026: Effective charity ≠ effective development (global health edition). EA Forum. https://forum. effectivealtruism.org/posts/p7EbDRK4FCs2795uh/ effective-charity-effective-development-globalhealth
Chima, C. and Franzini, L. 2016: Spillover effect of HIV-specific foreign aid on immunization services in Nigeria. International Health 8(2), 108–15. https://doi. org/10.1093/inthealth/ihv036
Dardot, P. and Laval, C. 2013 [2010]: The new way of the world: On neoliberal society. Verso.
Filmer, D.P., Friedman, J., Kandpal, E., et al. 2018: Cash transfers, food prices, and nutrition impacts on nonbeneficiary children [Policy Research Working Paper 8377]. World Bank. https://documents.worldbank.org/en/publication/ documents-reports/documentdetail/98903152 2077749796/
GiveWell. 2018a: Cash transfers in a nutshell. https:// www.givewell.org/international/technical/programs/ cash-transfers
GiveWell. 2018b: Revisiting leverage. The GiveWell Blog. https://blog.givewell.org/2018/02/13/revisitingleverage/
GiveWell. 2023: Mass distribution of insecticide-treated nets. https://www.givewell.org/international/ technical/programs/insecticide-treated-nets
GiveWell. 2024a: Re-evaluating the impact of unconditional cash transfers. The GiveWell Blog. https://blog. givewell.org/2024/11/12/re-evaluating-the-impactof-unconditional-cash-transfers/
GiveWell. 2024b: Spillover effects of GiveDirectly’s cash transfers program. https://www.givewell.org/ international/technical/programs/cash-transfers/ spillovers
GiveWell. 2025: GiveDirectly’s cash for poverty relief program. https://www.givewell.org/international/ technical/programs/givedirectly-cash-for-povertyrelief-program
Mussa, A.H., Pfeiffer, J., Gloyd, S.S., et al. 2013: Vertical funding, non-governmental organizations, and health system strengthening: Perspectives of public-sector health workers in Mozambique. Human Resources for Health 11(1), 26. https://doi. org/10.1186/1478-4491-11-26
Rana, Z., and Koch, D.J. 2020: Is it time to ‘decolonise’ the fungibility debate? Third World Quarterly 41(1), 42–57. https://doi.org/10.1080/01436597.2019.16 56060
Rana, Z., Koch, D.J., and Diem, C. 2025: Progress and pitfalls when evaluating the unintended effects of public policy: The case of German international development cooperation. Evaluation 31(4), 559–79.
World Bank. 2015: Philippines conditional cash transfer program impact evaluation 2012. https://hdl.handle. net/10986/13244