GiveBetter x SF

Against Malaria Foundation

Ordinary public giving for malaria nets

Research time: ~16 min on GPT-5.6 Sol Medium
  • Research — organization and evidence review.
  • Modeling — cost-effectiveness analysis.
  • Historical estimate for research done before time tracking.

Published: 8 September 2026.

Funding limitations

Summary

What do they do? AMF finances insecticide-treated nets and works with distribution and monitoring partners. Its ordinary public-gift policy directs donations to net purchases; specified other donors and partners cover non-net costs. More

Why this approach interests us

  • A focused organization supplies an intervention with randomized child-survival evidence and an identified funding pipeline. Ordinary public gifts buy nets across campaigns chosen by AMF.

Our main reservations

  • GiveWell's August 2026 summary uses 2022–24 grant-weighted funding opportunities. Applying that historical yield to an ordinary gift is our assumption, not a guaranteed offer or a verified price for today's next dollar. The native estimate already accounts for other funders' responses. Our QALY figures are not GiveWell's moral weights or its own QALY estimates.

What do you get for your dollar?

Our conditional best estimate is $3,778 per 10 global QALYs, or $378 per QALY. A representative ordinary gift is modeled at about $3,800 per 10 global QALYs. We apply GiveWell's historical $5,500 per modeled life saved to the whole gift, then use our own finite, age-aware health conversion. This is a conditional estimate, not a promised result. Inspect the current model. Original v1 model and scenarios.

global health
$3,778 — per 10 additional QALYs; whole-gift historical-benchmark scenario
bay area health
0% credited — No quantified direct local benefit; unknown spillovers are not relabeled global impact.

1. What do they do?

AMF finances insecticide-treated nets and works with distribution and monitoring partners. Its ordinary public-gift policy directs donations to net purchases; specified other donors and partners cover non-net costs.

Campaign selection and distribution

AMF selects eligible campaigns and partners. Household registration, logistics and net distribution are necessary complements, not free resources.

Protection and follow-up

Nets reduce exposure to malaria vectors. Presence, use and condition are monitored, but those observations alone do not establish the marginal mortality effect of a new donation.

Scope of this review. Ordinary public gift to AMF's core net-purchase route, allocated among AMF-selected campaigns. Representative core-giving estimate, not unrestricted all-expense funding or a GiveWell pooled-fund gift. This is not an all-expense institutional grant or a GiveWell pooled-fund gift. No bespoke country restriction is assumed.

2. Monitoring and information sharing

Children aged 6–59 months in northern Ghana

Cluster-randomized trial, 1993–95. Binka et al. found all-cause mortality RR .83 (95% CI .69–1.00). Follow-up through 2000 found no indication of increased later mortality.

Our assessment. Historical intervention evidence and context, not another numerical multiplier on the already adjusted mortality benchmark. Modern campaign mix and adult/child effects remain uncertain.

Tanzanian children followed for 22 years

Prospective observational cohort. Early net use was associated with survival into adulthood; observational design and incomplete follow-up limit causal interpretation.

Our assessment. Historical intervention evidence and context, not another numerical multiplier on the already adjusted mortality benchmark. Modern campaign mix and adult/child effects remain uncertain.

Modern Tanzania malaria-control setting

Randomized comparison of net types. Chlorfenapyr nets improved malaria outcomes relative to standard pyrethroid nets; other comparisons were less favorable.

Our assessment. Historical intervention evidence and context, not another numerical multiplier on the already adjusted mortality benchmark. Modern campaign mix and adult/child effects remain uncertain.

3. Qualitative assessment

A focused organization supplies an intervention with randomized child-survival evidence and an identified funding pipeline. Ordinary public gifts buy nets across campaigns chosen by AMF.

Key reservations

  • GiveWell's August 2026 summary uses 2022–24 grant-weighted funding opportunities. Applying that historical yield to an ordinary gift is our assumption, not a guaranteed offer or a verified price for today's next dollar. The native estimate already accounts for other funders' responses. Our QALY figures are not GiveWell's moral weights or its own QALY estimates.
  • We assume 80% of modeled deaths avoided occur in early childhood and 20% at older ages. These are analyst priors, not measured portfolio shares. Early deaths receive a finite 40-year survival trajectory; older deaths receive 20 years. We adjust for background mortality and health quality, and discount future health at 3% annually. We do not add a further generic persistence discount centrally. None of these trajectories is a fitted national life table.
  • Every estimate charges the entire donation. A matched full-resource or net societal cost remains unknown: we cannot simply reuse the old per-net or per-child allowances after changing the model's denominator. Direct modeled SF and Bay Area health is zero; unknown indirect effects receive no credit. The global figure is not a local cost-effectiveness estimate.
  • Nets protect children and adults who are alive during their use. We assume a mean 2.5-year delay from donation to the avoided death for both age groups. The 80/20 age split is a portfolio prior informed by, but not measured from, GiveWell's older-child/adult modeling. Delay and remaining survival are separate assumptions.

Benefits not included in our estimate

  • No additional morbidity, income or education multiplier is added to this mortality calibration.
  • No direct SF/Bay health is credited; unknown indirect effects are not assumed impossible.

4. What do you get for your dollar?

$3,778 per 10 global QALYs

The original net-level model estimated about $9,700 per 10 QALYs. Replacing its compounded mortality assumptions with an external benchmark improves calibration. We also removed an unsupported extra 20% persistence reduction and stopped assigning every modeled death a young child's survival trajectory. Those last two changes work in opposite directions. We now anchor mortality to GiveWell's historical funding-opportunity estimates, then make our own explicit conversion from avoided deaths to quality-adjusted life years. The earlier models remain available as constructed stress tests. Their arithmetic was correct, but combining several uncalibrated discounts made their central estimates more pessimistic than the external mortality benchmark. We have not multiplied the new benchmark by those old discounts again.

A better life is our comparison unit of 10 additional quality-adjusted life years (QALYs), potentially spread across people. These are uncertain estimates, not measured returns or verified donation offers.

How we calculate the estimate

Gift-time incremental health: Gift × core share ÷ historical dollars per modeled death × relative mortality yield × age-weighted discounted future QALYs − new harms
$10,000 produces 26.466 modeled global QALYs; full gift ÷ QALYs × 10 = $3,778

Model inputs and assumptions
Historical cost per modeled life saved
$5,500 (range: Historical benchmark; relative-yield scenarios 0–2×). GiveWell's August 2026 summary uses 2022–24 grant-weighted funding opportunities. Applying that historical yield to an ordinary gift is our assumption, not a guaranteed offer or a verified price for today's next dollar. The native estimate already accounts for other funders' responses. Our QALY figures are not GiveWell's moral weights or its own QALY estimates. GiveWell 2022–24 opportunities, not next-gift quote.
Share of whole gift assigned to modeled core
1 (range: Central allocation; 100% sensitivity). We treat the ordinary gift as supporting AMF's representative core activity. This is not a claim that a specific new campaign is unfunded or that all overhead disappears. Already-funded grants are evidence about effectiveness, not available donation offers. Analyst prior, not overhead rate.
Early versus older death share
80% / 20% (range: 0–100% early). 80%early/20%older is an analyst prior for both. AMF Uganda example is supporting context, not a grant-weighted empirical share. NI share is not inferred from moral-weighted benefit percentages. Analyst prior.
Early-death survival trajectory
40 years; annual survival .99; utility .85 (range: Three-year tail sensitivity). All horizons, annual survival, utility and midpoint integration are analyst trajectories, not national lifetables. Finite truncation does not imply death at horizon. Not a fitted national life table.
Older-death survival trajectory
20 years; annual survival .98; utility .80 (range: Three-year tail sensitivity). All horizons, annual survival, utility and midpoint integration are analyst trajectories, not national lifetables. Finite truncation does not imply death at horizon. Not a fitted national life table.
Years from gift to prevented death
Early: 2.5; older: 2.5 (range: Explicit conditional event schedule). Nets protect children and adults who are alive during their use. We assume a mean 2.5-year delay from donation to the avoided death for both age groups. The 80/20 age split is a portfolio prior informed by, but not measured from, GiveWell's older-child/adult modeling. Delay and remaining survival are separate assumptions. Analyst timing judgment.
Annual health discount rate
3% (range: Fixed in these scenarios). Discount gift-to-event and event-to-health intervals once each; midpoint annual integration. Model convention.
New shared and independent harms
0 QALYs (range: 0–0.1 present-value QALYs per specified gift). Only new harms not already in the source benchmark. Harm inputs are whole-gift present-value QALYs for the specified gift, not per dollar or per death; shared harm scales with relative yield, independent harm does not. Re-specify harms if changing gift scale. Sensitivity, not measured harm.

Representative ordinary gift

  • central: $3,778 per 10 global QALYs. 26.466 net QALYs for the whole $10,000 gift.
  • half-yield: $7,557 per 10 global QALYs. 13.233 net QALYs for the whole $10,000 gift.
  • double-yield: $1,889 per 10 global QALYs. 52.932 net QALYs for the whole $10,000 gift.
  • zero-yield: No positive-health ratio per 10 global QALYs. 0.000 net QALYs for the whole $10,000 gift.
  • all-early: $3,472 per 10 global QALYs. 28.805 net QALYs for the whole $10,000 gift.
  • all-older: $5,844 per 10 global QALYs. 17.111 net QALYs for the whole $10,000 gift.
  • short-tail: $24,995 per 10 global QALYs. 4.001 net QALYs for the whole $10,000 gift.
  • persistence-80: $4,723 per 10 global QALYs. 21.173 net QALYs for the whole $10,000 gift.
  • shared-harm: $3,793 per 10 global QALYs. 26.366 net QALYs for the whole $10,000 gift.
  • zero-yield-shared-harm: No positive-health ratio per 10 global QALYs. 0.000 net QALYs for the whole $10,000 gift.
  • zero-yield-independent-harm: No positive-health ratio per 10 global QALYs. -0.100 net QALYs for the whole $10,000 gift.
  • core-share-one: $3,778 per 10 global QALYs. 26.466 net QALYs for the whole $10,000 gift.

Uncertainty. The scenarios change assumptions; they are not confidence intervals. A donation can have no additional health effect, in which case there is no finite positive cost per QALY. Independently caused harm can make the health result negative even when benefits are zero. Harms already included in the source benchmark are not subtracted a second time. The roughly $100 difference between the two central figures is much smaller than the uncertainty in age mix, timing and marginal funding. It does not establish that one organization is better.

5. Funding and previous grants

Confirm current campaign allocation, costs, co-funding and additionality. A provider-reported funding gap does not guarantee that the next gift causes the modeled benefit.

AMF displayed an immediate $70.67M gap on 8 September 2026: $128.38M available against $199.06M approved potential use. Rounded figures do not reconcile exactly. The $276M DRC grant announced in August is already-funded baseline, not output of a fresh public gift. This is dated provider pipeline evidence, not a verified individual donation offer. Every estimate charges the entire donation. A matched full-resource or net societal cost remains unknown: we cannot simply reuse the old per-net or per-child allowances after changing the model's denominator. Direct modeled SF and Bay Area health is zero; unknown indirect effects receive no credit. The global figure is not a local cost-effectiveness estimate.

This review does not establish a verified marginal funding offer or a complete history of grants.

We have not verified a suitable donation route for this reviewed activity. Confirm the legal recipient and intended allocation before donating.

6. Sources

  1. Ghana cluster-randomized mortality trial. onlinelibrary.wiley.com. Primary source; access and interpretation limits documented in model. Published: April 1996; retrieved: 8 September 2026.
  2. Public gifts and non-net costs. www.againstmalaria.com. Primary source; access and interpretation limits documented in model. Published: Undated, live provider page; retrieved: 8 September 2026.
  3. Global accounts, year ended June 2025. www.againstmalaria.com. Primary source; access and interpretation limits documented in model. Published: Year ended June 2025; retrieved: 8 September 2026.
  4. Immediate funding gap. www.againstmalaria.com. Primary source; access and interpretation limits documented in model. Published: Undated, live provider page; retrieved: 8 September 2026.
  5. Charity registrations. www.againstmalaria.com. Primary source; access and interpretation limits documented in model. Published: Undated, live provider page; retrieved: 8 September 2026.
  6. August 2026 DRC grant announcement. blog.givewell.org. Primary source; access and interpretation limits documented in model. Published: 24 August 2026; retrieved: 8 September 2026.
  7. Research behind the DRC grant. blog.givewell.org. Primary source; access and interpretation limits documented in model. Published: August 2026; retrieved: 8 September 2026.
  8. Modern Tanzania net-type trial. pmc.ncbi.nlm.nih.gov. Primary source; access and interpretation limits documented in model. Published: 2022; retrieved: 8 September 2026.
  9. Twenty-two-year survival cohort. www.nejm.org. Primary source; access and interpretation limits documented in model. Published: 2022; retrieved: 8 September 2026.
  10. Ghana follow-up: primary indexed abstract. www.ebi.ac.uk. Primary source; access and interpretation limits documented in model. Published: 2002; retrieved: 8 September 2026.
  11. Institutional survival-cohort record. edoc.unibas.ch. Primary source; access and interpretation limits documented in model. Published: 2022; retrieved: 8 September 2026.
  12. GiveWell: How We Produce Impact Estimates. GiveWell. Primary evaluator analysis; independent QALY conversion. Published: Mixed-vintage technical analysis; see linked source and model; retrieved: 8 September 2026.
  13. GiveWell: Mass Distribution of Insecticide-Treated Nets, especially section4.3. GiveWell. Primary evaluator analysis; independent QALY conversion. Published: Mixed-vintage technical analysis; see linked source and model; retrieved: 8 September 2026.
  14. GiveWell: New Incentives technical analysis. GiveWell. Primary evaluator analysis; independent QALY conversion. Published: Mixed-vintage technical analysis; see linked source and model; retrieved: 8 September 2026.
  15. GiveWell: NI extension of support in Northern Nigeria. GiveWell. Primary evaluator analysis; independent QALY conversion. Published: Mixed-vintage technical analysis; see linked source and model; retrieved: 8 September 2026.

Annual expenses: years and sources

Average annual expenses (three consecutive fiscal years): Not available. Organization size is separate from the modeled cost-effectiveness of a donation.

Against Malaria Foundation

Three years of organization-level expenses have not yet been verified for this report. No other organization or fiscal sponsor budget has been substituted.

Why does this differ from GiveWell?

GiveWell’s August 2026 summary estimates $5,500 per life saved for this program, based on its 2022–2024 funding opportunities. This is a modeled mortality outcome—not a QALY estimate or a quote for the next unrestricted gift.

Our original v1 scenario implied $12,449 per early death prevented, before converting survival into QALYs. It credited about 13.65 discounted QALYs per prevented death, using a finite 40-year survival curve, background mortality, .85 utility, .8 persistence and 3% discounting. Those survival inputs are judgments, not measured life tables.

Using GiveWell’s mortality estimate with that same event-time survival assumption gives $4,030 per 10 QALYs, before a new gift’s implementation delay. Using an illustrative 50 QALYs per life instead gives $110 per QALY, or $1,100 per 10 QALYs. Fifty undiscounted years are not automatically fifty discounted QALYs.

GiveWell’s aggregate lives-saved estimate may include different ages and timing from our early-child-death model. Applying the same QALYs per life here is a diagnostic assumption, not an age-matched survival calculation.

What we learned

There is no factor-of-ten arithmetic error. Our original model compounds conservative transfer and funding assumptions that were not calibrated to GiveWell’s analysis. Its result should be read as a constructed scenario, not proof that GiveWell overstates impact. We should not apply our full adjustment ladder again to GiveWell’s already-adjusted mortality estimate.

For malaria nets, trial transfer, effective protection over time and funding replacement can overlap if each is used as a generic uncertainty penalty. They must represent different causal losses. Campaign geography and partner delivery costs also differ from a simple net-purchase budget.

GiveWell evaluates specific funding opportunities. Displacing its funding may move impact to another program rather than eliminate it. Conversely, a historical average does not establish the effect of every direct gift. The current v2 report replaces that mortality ladder with the historical benchmark and age-aware timing assumptions. We retain the original model for auditability. None of these global calculations establishes SF benefit.

Inspect the comparison inputs and formulas.

How much might this help San Francisco?

Direct benefits accrue overseas. Preventing infections could indirectly protect SF residents through travel and reduced importation, but no evidence currently calibrates this charity’s local effect. We leave the SF comparison price unquantified rather than show its global price as a local bargain.

Tiny local-spillover math

If additional SF health were one millionth of the modeled direct overseas health, the same gift would cost $3.8B per 10 SF QALYs. This is an illustrative coefficient, not an estimated share or an evidence-based bound.

These scenarios carry forward the direct model’s finite schedule only as a simplifying diagnostic. They do not model an infinite transmission or innovation tail. Overseas health is not reassigned to SF; SF is nested within Bay, so the two cannot be added.

Relevant mechanisms: measles importation and rare local malaria transmission. Neither source estimates AMF or New Incentives’ effect on SF.