Battle of Ideas

Foreign aid does more harm than good

A steelman map of the public fight over foreign aid — arguments for and against, with sources and assumptions.

AI-generated · paired steelman agents · independently red-teamed · Pass-1 source spot-checks only · framing-fidelity not independently verified · single model family

Official development assistance and large transfers to poorer countries. Emergency humanitarian relief (earthquake, famine) is a named third camp, not stuffed into FOR. Remittances are not aid. Not a US-budget-only page — the fight is whether the transfers help.

AGAINST 5

no further strong arguments at this depth

FOR 5

no further strong arguments at this depth

People also ask

Questions people actually type. The two columns above are the cases — open a card for sources, assumptions, and counters.

Does foreign aid work?

Does foreign aid do more harm than good?

Does international aid to Africa do more harm than good?

What are the disadvantages of foreign aid?

Is foreign aid effective?

Does foreign aid really work?

Foreign aid by country

Ordering within each column: strongest first — validation tier, then source quality, then representativeness.

AGAINST · Foreign aid does more harm than good
Empirical — moderateP1

Child deaths halved on a clock that aid actually funds

UN IGME / UNICEF: the global under-five mortality rate fell from 94 deaths per 1,000 live births in 1990 to 37 in 2023 — a 61 percent decline. That is not a mood. Infectious disease, malaria, and HIV were the killers, and the tools that moved them — expanded programme on immunization, insecticide-treated nets, antiretroviral therapy — were paid for in large part by official development assistance and the large transfers this page is about. Cochrane (Pryce, Richardson, Lengeler, 2018): insecticide-treated nets versus no nets reduce all-cause child mortality by 17 percent (rate ratio 0.83; five trials, ~200,000 children; high-certainty). WHO's World Malaria Report has put cases and deaths averted since 2000 in the hundreds of millions and millions respectively, with nets the largest single tool. PEPFAR and the Global Fund are not a US-budget exhibit; they are the reason AIDS deaths fell from a 2004 peak on the order of two million a year to well under a million. hiv.gov, citing UNAIDS, reports larger infection and death declines in PEPFAR-supported countries than in the global average. AGAINST does not need every clinic photograph. It needs the inequality: a Dutch-disease paper about manufacturing shares does not outweigh children who did not die. Emergency famine airlifts are a third camp, named so they are not stuffed in here. Vaccines and nets are development assistance. They are the good.

Key assumptions

  • A large share of the child-mortality decline is caused by aid-funded tools rather than by income growth that would have happened anyway partial
  • Lives saved are the right numerator for 'good,' even if GDP per capita does not move untestable

Red team — the strongest counters

China's children did not die less because of DAC grants

The global U5MR collapse is dominated by places whose growth was not aid. Vaccines have a market price. Attributing 61 percent to ODA is a poster, not a decomposition.

Nets and ART are the boutique, not the portfolio

Cochrane on ITNs is a real good. Budget support, technical assistance, and tied consultants are the rest of the USD 200 billion. Stack-ranking the pile by its best trial is how every industry defends itself.

Sources

  • UN IGME / UNICEF child mortality U5MR 94 per 1,000 in 1990 to 37 in 2023 (−61%). Pass-1: data.unicef.org child-survival page exists. P1 checked
  • Insecticide-treated nets for preventing malaria Pryce, Richardson & Lengeler, Cochrane 2018, CD000363. ITNs vs no nets: all-cause child mortality RR 0.83 (0.77–0.89); ~5.6 lives per 1,000 children per year. Pass-1: Cochrane/PMC 6418392 exist. P1 checked
  • PEPFAR latest global results hiv.gov / State: larger HIV incidence and AIDS-death declines in PEPFAR-supported countries than globally (UNAIDS estimates). Pass-1: hiv.gov PEPFAR page exists. Treat '26 million lives' as a programme talking point, not a trial. P1 checked

Confidence, decomposed

Logical validity●●●●●
Premise support●●●●○
Representativeness●●●●●
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · Foreign aid does more harm than good
Empirical — moderateP1

When you just give the money, consumption and the local economy move

Haushofer and Shapiro (Quarterly Journal of Economics, 2016) randomised GiveDirectly's unconditional cash in rural Kenya: consumption, assets, and psychological well-being rose; 'temptation goods' did not. Egger, Haushofer, Miguel, Niehaus and Walker (Econometrica, 2022) then dumped about USD 1,000 each on more than 10,500 households across 653 villages — a fiscal shock over 15 percent of local GDP — and found large gains for recipients, positive spillovers on non-recipients and firms, minimal price inflation, and a local transfer multiplier around 2.5. This is not a ministry of planning. It is the large-transfer half of the claim, run as an experiment. AGAINST's point is not that every World Bank project is GiveDirectly. It is that the harm claim has to survive the case where the transfer actually arrives. If cash at scale raises consumption without a price spike, 'aid does more harm than good' cannot be a claim about transfers as such. It has to be a claim about the rest of the portfolio. The rest of the portfolio includes the nets and the ARVs. Remittances are not aid; they are a cousin that shows households can absorb money. GiveDirectly is aid. The RCT is the measurement.

Key assumptions

  • NGO cash in Kenya is representative enough of 'large transfers' that it speaks to the claim, not only to one charity partial
  • A local multiplier of 2.5 is a good, not a harm, even if it is not GDP-per-capita convergence with the donor untestable

Red team — the strongest counters

One NGO, one region, a one-shot thousand dollars

GiveDirectly in rural Kenya is not the Ministry of Finance in Kinshasa. A local multiplier of 2.5 over a short horizon is not a country catching up. The claim is ODA, not one charity's RCT.

Spillovers on non-recipients can be a transfer, not a free lunch

If non-recipients gain because recipients buy their maize, that is redistribution inside the village. It is not evidence the donor's dollar created development rather than a demand pulse.

Sources

Confidence, decomposed

Logical validity●●●●●
Premise support●●●●○
Representativeness●●●●○
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · Foreign aid does more harm than good
Logically validP1

The Nobel method is how you keep the good and kill the rest

The 2019 economics Nobel went to Banerjee, Duflo and Kremer 'for their experimental approach to alleviating global poverty.' Poor Economics (2011) is the book of the method: deworming, remedial tutoring, immunisation-plus-lentils, bednets — some of it works, some of the romantic portfolio does not (microcredit's income miracle did not survive the RCTs). J-PAL's point, and AGAINST's, is that 'aid' is not one object. A growth-regression null on a bag of budget support, technical assistance, and tied consultants is not a finding that treated nets fail. Easterly is right that planners lack feedback. The randomistas are the feedback. Sachs's End of Poverty and the Millennium Villages are the other pole — a big simultaneous push — and independent evaluations of the Villages are mixed enough that AGAINST should not hide behind them. The steelman does not need a big push. It needs the existence of interventions, funded as ODA, that pass a trial, plus a political process that can shift money toward those and away from the palace. 'More harm than good' is a stack ranking of the whole pile. The pile contains things that save children at a known rate. Ranking the pile as a harm is a refusal to sort.

Key assumptions

  • The aid portfolio can actually be shifted toward what RCTs pass, rather than the RCTs remaining a boutique beside the old machine partial
  • Success at the margin (a net, a tutoring programme) counts as 'good' for the claim even if country GDP does not converge untestable

Red team — the strongest counters

RCTs exist because the rest failed the exam

Deworming and lentils-for-shots are the admission that transformational aid did not know what it was doing. They are not a defence of the country strategy. Easterly can grant every J-PAL paper and keep the claim.

What scales in a trial often dies in a ministry

External validity is the live critique of the randomistas. A tutoring programme that works in a district need not survive procurement, politics, and the next government's whim. Nobel method ≠ portable good.

Sources

  • The Prize in Economic Sciences 2019 Nobel citation: Banerjee, Duflo, Kremer, experimental approach to global poverty; remedial tutoring and preventive health as examples. Pass-1: nobelprize.org press release exists. P1 checked
  • Poor Economics Banerjee & Duflo, PublicAffairs, 2011. RCT-grounded account of which antipoverty interventions work. Pass-1: publisher page exists. Not a finding that all ODA works. (pooreconomics.com did not respond — replaced.) P1 checked

Confidence, decomposed

Logical validity●●●●●
Premise support●●●●○
Representativeness●●●●●
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · Foreign aid does more harm than good
Logically validP1

A growth null is not a harm — GDP was the wrong exam

Take the FOR growth papers at full strength. Rajan–Subramanian (2008) do not find a reliable aid-to-GDP coefficient. Easterly–Levine–Roodman (2004) sink Burnside–Dollar's 'aid works with good policy' as a robust result. AGAINST can grant both and still lose the claim. 'More harm than good' is a comparison. A coefficient near zero is not a harm; it is a miss on one outcome. Life expectancy, under-five survival, and years of school are other outcomes, and they moved in the decades of high ODA. Some of that is growth in China and India that was not aid. Some of it is Ghana and Bangladesh and Ethiopia, where health programmes and cash and debt relief sat next to domestic policy. The honest residual is not zero, and it is not a factory. Dutch disease is a real channel for some real exchange rates. It does not show that the nets were a net harm. Conditionality and palace windfalls are real. They are an argument for how to give — cash to households, vaccines, published RCTs — not for the claim that the transfers, on net, made people worse. The third camp (earthquake, famine) is not this argument's alibi. This argument is the development column, scored on lives and consumption, not on a growth t-stat.

Key assumptions

  • Health and consumption gains can be attributed to aid often enough to beat the governance and Dutch-disease losses partial
  • GDP per capita is not the master outcome for 'good' untestable

Red team — the strongest counters

Moving the numerator after GDP failed is a tell

The industry advertised development. When growth coefficients vanished, the brochure became lives. Lives matter. They are also how you never lose an argument that was about whether the transfers develop.

Governance and Dutch-disease costs still sit on the books

A logical gap between 'null' and 'harm' is real. So are palace windfalls and overvalued exchange rates. AGAINST has to net those against the nets, not declare GDP the wrong exam and leave the stand.

Sources

Confidence, decomposed

Logical validity●●●●●
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
AGAINST · Foreign aid does more harm than good
Empirical — weakP1

Cutting aid is also a policy — and we have a first look at the cut

OECD preliminary figures: DAC ODA on a grant-equivalent basis was USD 214.6 billion in 2024 (−6% real from 2023) and USD 174.3 billion in 2025 (−23.1% from 2024), the largest annual contraction on the record they publish. Humanitarian ODA fell with it. That is not a thought experiment. UNAIDS and modelling groups spent 2025 writing down what a PEPFAR freeze does to deaths and infections; the direction is not subtle. AGAINST does not need a precise extra-death forecast — those models are contested — to make the logical point. If the claim is that the transfers do more harm than good, a large, sudden removal should, on net, help. Health ministries in recipient countries, and the people on ART, are not describing a help. Sachs's big-push rhetoric over-claimed what a village package could do. Easterly's planner critique is fair about the packages that fail. Neither sentence licenses a cut that takes the nets and the ARVs with it because a growth regression was null. The harm claim is a stack ranking of a mixed portfolio. The cut does not sort the pile. It shrinks the pile, good and bad together. That is the test the slogan did not want.

Key assumptions

  • 2024–25 ODA cuts fall on effective health programmes as well as on waste, not only on the waste partial
  • Near-term HIV and malaria resurgence would count as 'harm from less aid' rather than as domestic-government failure partial

Red team — the strongest counters

A freeze is not an RCT of the claim

Models of extra HIV deaths after a PEPFAR pause are scenarios. Recipient governments can buy generics. Treating a 2025 budget fight as proof that the prior USD 200 billion was a net good is the same stack-ranking error in reverse.

The cut might be sorting, not shrinking the good

If health shares were protected and consultants were not, a historic drop could be the Easterly update. OECD totals do not show which line items died. Directional panic is not a decomposition.

Sources

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●○
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · Foreign aid does more harm than good
Empirical — moderateP1

The growth regressions do not find that aid raises growth

Rajan and Subramanian (Review of Economics and Statistics, 2008) look at aid and growth in cross-section and panel, after trying to correct for the fact that poorer countries get more aid. They find little robust evidence of a positive — or negative — relationship. That is not a blog. It is the paper the profession still cites when someone claims the gift is in the GDP. Burnside and Dollar (American Economic Review, 2000) had given the industry its favourite sentence: aid raises growth in a good policy environment. Easterly, Levine and Roodman (AER, 2004) added data and the interaction disappeared. OECD DAC ODA from member countries was USD 223.7 billion in 2023 and USD 214.6 billion in 2024 on a grant-equivalent basis; preliminary 2025 figures are a further large drop. FOR's point is not the latest year's total. It is that sums of that order, repeated, still do not produce a robust growth coefficient. Emergency famine and earthquake relief is a third camp — a plane of grain after a harvest failure is not a growth programme, and this argument does not spend it. The claim is ODA and large development transfers. Those are the flows whose advertised product is development. The product is not in the panel.

Key assumptions

  • Cross-country and panel growth regressions are a fair test of whether aid, on net, helps partial
  • A null on growth is enough to ground 'more harm than good' once governance and Dutch-disease channels are admitted, rather than a mere 'we cannot see the good' untestable

Red team — the strongest counters

A growth null is not a harm

Rajan–Subramanian find little robust relationship either way. That is a miss on GDP, not a finding that children died who otherwise would not. Health RCTs and under-five survival are a different numerator. Stack-ranking 'harm > good' off a t-stat is the leap.

Aid still goes to the slow growers — the coefficient remains dirty

Their instruments try to fix endogeneity and still cannot turn a messy allocation rule into a clean experiment. A country in civil war gets humanitarian and reconstruction lines; a fast East Asian grower aged out of the DAC list. The panel is not a trial of the nets.

Sources

  • Aid and Growth: What Does the Cross-Country Evidence Really Show? Rajan & Subramanian, Review of Economics and Statistics 90(4):643–665, 2008. Little robust evidence of a positive (or negative) aid–growth relationship after addressing endogeneity. Pass-1: REST/MIT page and NBER w11513 exist. P1 checked
  • Aid, Policies, and Growth: Comment Easterly, Levine & Roodman, American Economic Review 94(3):774–780, 2004. Extending Burnside & Dollar (AER 2000) forward, the aid×policy growth result does not hold. Pass-1: NBER w9846 / AER comment exist. P1 checked
  • OECD DAC official development assistance statistics Final 2024: DAC members USD 214.6 billion grant-equivalent, −6% vs 2023. Preliminary 2023: USD 223.7 billion. Pass-1: oecd.org ODA pages exist. Worldwide DAC, not a US-budget exhibit. P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●●○
Representativeness●●●●●
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · Foreign aid does more harm than good
Empirical — moderateP1

Aid can price a country's factories out of the world

Rajan and Subramanian (Journal of Development Economics, 2011) exploit variation within countries and across manufacturing sectors. They find that aid inflows are associated with slower growth in manufacturing, especially in sectors that should be more sensitive to real-exchange-rate appreciation — the pattern you would write down for Dutch disease. The mechanism is not mystical. A government that can pay its bills in dollars without selling shirts does not need a competitive export sector as badly as a government that must. Tradable jobs are how poor countries have historically got rich; an aid windfall that cheapens imports and dearens labour is a tax on that path. FOR can grant that some aid is spent on ports and power that help factories. The steelman is that the exchange-rate channel is in the data, and it is a harm to the exact sector development ministries put on the slide. Humanitarian airlifts after an earthquake are not this argument. Budget support and multi-year programme aid are.

Key assumptions

  • Slower manufacturing growth is a welfare harm, not just a sectoral reallocation toward non-tradables that people might prefer partial
  • The within-country, across-sector pattern is aid causing appreciation rather than aid going to countries whose factories were already failing partial

Red team — the strongest counters

Factories are not welfare

A shrinking manufacturing share can be a rich-country path — services, public health, urban jobs — not a curse. Dutch disease is a relative-price story. It does not show that the people in the non-tradable sector were harmed.

Aid also buys the ports that factories need

The same inflow can fund power, roads, and customs. Rajan–Subramanian's sector test is clever; it is not a full general-equilibrium audit of the grant. Infrastructure ODA is in the pile this argument taxes.

Sources

  • Aid, Dutch disease, and manufacturing growth Rajan & Subramanian, Journal of Development Economics 94(1):106–118, 2011. Aid associated with slower manufacturing growth in exchange-rate-sensitive sectors. Pass-1: JDE page and PIIE working-paper PDF exist. P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●●○
Representativeness●●●●●
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · Foreign aid does more harm than good
Empirical — moderateP1

A cheque to the palace is oil, not a school

Djankov, Montalvo and Reynal-Querol (Journal of Economic Growth, 2008) treat aid as a resource curse: a windfall that, in their panels, is associated with worse democratic institutions, in the same family of results as oil. Moss, Pettersson and van de Walle (CGD, 2006) name the aid-institutions paradox for aid-dependent Africa: money that arrives from outside can substitute for the need to build a tax system and a civil service that survive when the donor leaves. Dambisa Moyo's Dead Aid (2009) is the public book of that case for Africa — dependency, inflation, and a political class that campaigns in European capitals. FOR does not need Moyo to be a growth paper. It needs the mechanism: if the ministry of finance eats from the World Bank, the farmer is not the principal. Fungibility makes the problem worse. A donor who 'funds education' frees the recipient to fund the presidential guard. Conditionality was supposed to stop that. Easterly's record is that conditions are bargained, missed, and rolled. The third camp still sits outside: a one-off famine shipment is not a substitute for domestic taxes. The claim is the repeated transfer that becomes part of how the state lives.

Key assumptions

  • Worse measured democracy and weaker tax effort are harms that outweigh whatever the earmarked project bought untestable
  • The oil analogy is fair — aid is unearned like a resource, not earned like exports partial

Red team — the strongest counters

Democracy indices are not a village

Djankov et al. move a Polity-style score. That is not the same object as a clinic that stays open. Aid also funds election commissions, journalists, and court projects — mixed, captured, and still not 'oil.'

Moyo is the vernacular, not the regression

Dead Aid is a 2009 polemic. African growth after 2000 happened while ODA was still large. A book that was already arguing in a boom is not a panel that survived the boom.

Sources

  • The curse of aid Djankov, Montalvo & Reynal-Querol, Journal of Economic Growth 13:169–194, 2008. Aid as a resource-curse analogue for institutions/democracy. Pass-1: Springer page exists (doi 10.1007/s10887-008-9032-8). P1 checked
  • An Aid-Institutions Paradox? Moss, Pettersson & van de Walle, Center for Global Development Working Paper 74, 2006. Aid dependence can undermine state-building. Pass-1: CGD page and PDF exist. P1 checked
  • Dead Aid Dambisa Moyo, 2009. Public case that aid to Africa fostered dependency and corruption. Pass-1: the book exists; treat as the vernacular FOR, not as a regression. P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · Foreign aid does more harm than good
Logically validP1

Remittances already dwarf aid — and they are not aid

World Bank / KNOMAD figures put officially recorded remittances to low- and middle-income countries at about USD 656 billion in 2023 — larger than FDI and larger than ODA, and they have been larger than ODA for years. Those dollars are not official development assistance. The scope of this page says so on purpose. FOR uses them as a contrast, not as a redefinition. A nurse in Madrid sending money to Dakar does not overvalue the real exchange rate in quite the same way as a budget-support grant, does not write the education plan, and does not keep a president from taxing. She also does not claim to have developed Senegal. The aid industry's reply is that remittances are private consumption, not public goods. FOR can grant the public-goods gap and still say: if the advertised product of ODA is 'help poorer countries,' and the flow that actually reaches households is already several times ODA, the burden of proof is on the official transfer to show it does something remittances cannot — without the Dutch disease, the palace windfall, and the planner. GiveDirectly-style cash is the closest official cousin; it is a large transfer, so it is in the fight, and it is the exception that names the rule. Emergency relief remains the third camp. This argument does not spend a famine airlift.

Key assumptions

  • Remittances are a fair contrast to ODA rather than a different object that cannot speak to whether aid helps partial
  • Household consumption from remittances is on net better for poor people than official projects of the same dollar amount partial

Red team — the strongest counters

Remittances are not a public-goods substitute

A nurse's transfer does not buy a cold chain or a bednet campaign. Contrast with household cash is real and still leaves the goods only a ministry or an NGO can bulk-procure.

Remittances can Dutch-disease too

Large private inflows also lift real exchange rates (the Dutch literature on that is old). If the mechanism is the windfall, the contrast does not rescue FOR — it names a cousin of the same harm.

Sources

  • World Bank Migration and Development Brief — remittances 2023 Remittances to LMICs estimated at USD 656 billion in 2023; surpassed FDI and ODA. Pass-1: World Bank 26 Jun 2024 press release and Peoplemove/KNOMAD brief exist. Remittances are cited as contrast, not as aid. P1 checked
  • OECD DAC ODA totals DAC ODA on the order of USD 200+ billion in 2023–24, well below LMIC remittances. Pass-1: already used on this page; here it is the denominator of the contrast. P1 checked

Confidence, decomposed

Logical validity●●●●○
Premise support●●●●○
Representativeness●●●●○
Source quality●●●●●

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.
FOR · Foreign aid does more harm than good
Logically validP1

Planners do not have to answer to the people they claim to help

William Easterly's White Man's Burden (2006) and his JEL survey 'Can the West Save Africa?' (2009) draw a line the FOR case needs: the West keeps trying to save a continent with a comprehensive plan, and the feedback that would kill a bad plan is missing. A firm that ships unsellable shirts goes bankrupt. A project that misses its indicators is redesigned, relabelled, and refunded. That is not a personality critique of aid workers. It is an incentive structure. Poor people do not vote on the DAC. They do not fire the task team. Sachs-style big pushes (The End of Poverty; Millennium Villages) are the other pole: enough money, simultaneous investments, a poverty trap broken. FOR's reply is Easterly's: we have been pushing. The accountability is still upward, to parliaments and NGOs in the donor country, which is why visible clinics photograph better than property rights. RCTs (Banerjee, Duflo, Kremer) are the honest admission that we did not know what worked. They are also not a defence of the rest of the portfolio — they are a method for finding the rare thing that does, at a scale that is not the World Bank's country strategy. FOR can grant every bednet trial and still say: the industry that needed those trials was not, on net, a help.

Key assumptions

  • Missing feedback is a structural fact of official aid, not a reformable reporting problem partial
  • RCT successes at the margin do not salvage the transformational portfolio partial

Red team — the strongest counters

The randomistas are the feedback Easterly asked for

J-PAL, GiveDirectly, and the 2019 Nobel are the industry answering the planner charge. Treating that as proof the rest of the portfolio is a net harm is how you refuse an update.

Searchers still cannot sell a vaccine campaign at retail

Public goods and epidemiological externalities are why some plans exist. A firm that must survive a market will under-provide nets. Easterly's dichotomy taxes the thing only a planner can buy.

Sources

  • The White Man's Burden William Easterly, 2006. Planners vs searchers; feedback and accountability as the missing piece of official aid. Pass-1: the book exists (Penguin). Dated dollar-stock figures in the text are not reused here. P1 checked
  • Can the West Save Africa? Easterly, Journal of Economic Literature 47(2):373–447, 2009. Survey contrasting transformational 'save Africa' plans with marginal approaches. Pass-1: AEA page exists (doi 10.1257/jel.47.2.373). P1 checked

Confidence, decomposed

Logical validity●●●●●
Premise support●●●○○
Representativeness●●●●●
Source quality●●●●○

Provenance

Generated by a paired steelman agent (single model family) · red-teamed by an independent adversarial agent · sources Pass-1 spot-checked (existence and rough fit) — framing-fidelity not independently verified. Judged on merit: per the founding rule of this project, AI authorship is disclosed at site level and arguments stand or fall on their content.