Topic 36 of 64· Social sector and philanthropy metrics
Effective Altruism Cost-Effectiveness
Effective altruism (EA) cost-effectiveness reasoning ranks charitable interventions by the amount of good — most often expressed as lives saved, or health gained, per dollar spent — and directs money toward whichever intervention buys the most good at the margin. GiveWell is the field's most influential practitioner: it publishes explicit, updated cost-per-life-saved and cost-per-outcome estimates for a small list of "top charities," and recommends donors give to whichever currently has room for more funding at the best rate.
Why it matters
GiveWell states cost-effectiveness as the leading criterion in its published methodology: it looks for evidence-backed interventions, estimates their cost-effectiveness in a common unit, and ranks across completely unrelated causes — bednets against malaria, vitamin A supplementation, cash transfers, vaccine incentive payments — on that single axis. This is a direct import of QALY/DALY-style reasoning from health economics into philanthropy: just as a health system asks "how many QALYs per pound at the margin," GiveWell asks "how many lives, or life-years, per dollar at the margin," and treats causes as substitutable once converted into that common unit. See cost-effectiveness analysis in government for the public-sector cousin of this reasoning framework.
The most cited GiveWell figure concerns the Against Malaria Foundation (AMF), which distributes insecticide-treated bednets. In GiveWell's published worked example (drawn from 2020 funding data), roughly $4,500 funded enough nets to avert one death, after accounting for imperfect net usage, baseline mortality without nets, and adjustment for funging — the possibility that AMF would have received some of that funding from other donors regardless. GiveWell is explicit that this figure moves over time and across geographies as malaria prevalence, net costs, and funding gaps change, and that the cost to save a life is generally expected to rise over time as the cheapest opportunities are taken up first; it is a worked illustration of the method, not a fixed price.
The maths
Cost-effectiveness = Cost of intervention / Units of good produced
(e.g. $ per life saved, $ per DALY averted, $ per QALY)
GiveWell's chain for a bednet programme, illustratively:
$ per net purchased and delivered
÷ share of nets actually used
÷ people protected per net
× baseline annual mortality without nets
× reduction in mortality attributable to net use (from RCT evidence)
× years of protection per net
÷ adjustment for funging (money displacing other donors' funding)
= $ per life saved (net of counterfactual funding effects)
This chain matters because each step is a place cost-effectiveness estimates commonly go wrong — see the pitfalls below — and because it makes explicit that "cost per life saved" is never a raw observed price; it is a modelled estimate built from several separately uncertain inputs.
Worked example
Two hypothetical interventions, both evidence-backed, competing for the same marginal £100,000:
- Bednets (AMF-style): roughly $4,500 per life saved on GiveWell's published worked example drawn from 2020 data, i.e. very roughly 20 lives saved per £100,000 depending on the exchange rate and year used.
- Deworming programme: no plausible mortality benefit at all, but strong evidence of long-run income gains from childhood deworming; GiveWell values it in income-gain terms, not lives saved, which makes it hard to compare directly against bednets without a shared unit. GiveWell uses an explicit "moral weights" framework to convert both into one internal unit for ranking.
The EA method's discipline is forcing this comparison into the open rather than funding both because both "sound good." See social return on investment for the equivalent forcing function used by UK social enterprises and local commissioners, which asks the same question — what is the best return per pound — in a monetized-value idiom rather than a lives/DALYs idiom.
Software engineering connection
Engineers building donor platforms, grant-matching tools, or impact dashboards for EA-aligned funders (Open Philanthropy, GiveWell itself, effective giving platforms such as Giving What We Can) need to represent cost-effectiveness estimates as ranges with stated assumptions, not single numbers — the underlying model has several multiplicative uncertain inputs, and collapsing that to one figure on a dashboard misrepresents the confidence GiveWell itself states. Version each estimate by publication date; GiveWell revises its numbers, sometimes substantially, as new RCT evidence or funding-gap data arrives, and a platform that caches an old figure silently becomes wrong.
Pitfalls
- Treating a cost-effectiveness estimate as a fixed price. It is a model output with several uncertain multiplicative inputs (usage rates, baseline mortality, funging adjustment); state the date and version.
- Ignoring funging/displacement. Funding an organization that would have received the money from another donor anyway buys less counterfactual good than the headline suggests — see additionality and deadweight and displacement and attribution.
- Comparing across incompatible units without conversion. "Lives saved" and "income gained" are not directly comparable without an explicit moral-weights framework; presenting them side by side as if they were is a category error.
- Cause-area tunnel vision. Ranking only within a cause area (e.g. only global health charities) and calling the winner "the most cost-effective charity" overstates the claim; GiveWell's cross-cause ranking is deliberately narrow (global health and wellbeing), not universal.
Sources
- GiveWell, "Our criteria." https://www.givewell.org/how-we-work/our-criteria
- GiveWell, "How Much Does It Cost to Save a Life?" (February 2024 version). https://www.givewell.org/how-much-does-it-cost-to-save-a-life/february-2024-version
- GiveWell, Against Malaria Foundation review. https://www.givewell.org/charities/amf
- Giving What We Can, on cost-effectiveness across causes. https://www.givingwhatwecan.org/