Start by separating layers. Proximate causes are capital, labour and technology — but saying "rich countries have more capital" only pushes the question back one step: why were they able to accumulate it? Fundamental causes have to explain the difference in accumulating behaviour itself.
Why this is worth being precise about: growth compounds. Income per head rising 2% a year versus 0.5% a year differs by roughly 4.4× after a century. Engineering analogy: this is not performance tuning, it is a difference in complexity class.
Today's gaps of tens of times over were also opened only in the last few centuries (the Great Divergence): they are a historical product, not a permanent condition.
| Comparison | Starting point | Outcome |
|---|---|---|
| South Korea vs Ghana | World Bank data for 1960: GDP per head about $159 and $175 (current dollars of the day) — Ghana slightly higher | By 2017, about $29,743 and $1,641 — a gap of roughly 18× |
| Botswana | At independence in 1966: about 12 km of paved road and 22 university graduates in the whole country | GDP per head grew about 7.8% a year over 1966–1991, the highest in the world at the time; yet it remains heavily dependent on diamonds, with high unemployment and inequality |
| Divided countries | The two Germanys, the two Koreas: same geography, same language and cultural tradition | Output diverged sharply once institutions split — the strongest single piece of evidence that geography and culture cannot explain everything on their own |
The "find the fundamental cause" camp: without going down to the fundamental layer, policy advice is just a description of what rich countries look like; only a deep variable explains why the same aid produces wildly different results across countries.
The "no master key" camp (Rodrik's growth-diagnostics approach): different countries are stuck on different binding constraints — capital, human capital, or insecure property rights. Chasing a single fundamental cause hands every country the same prescription.
Trade-off: the first travels well, at the cost of over-generalising; the second fits the individual case, at the cost of never becoming a general theory.
Mistaking correlation for recipe: what rich countries have today (generous welfare, an independent central bank, high tertiary enrolment) mostly arrived after they got rich, so copying the features is not copying the path. The other misreading is treating poverty as "nothing was tried" — most persistent poverty is an equilibrium: given the prevailing rules and expectations, each person's choice is rational, which is exactly why it is stable.
Three schools compete for the same slot: which factor is fundamental. Institutionalists (North's definition: institutions are the rules of the social game) locate the key in whether property rights are predictable and power is constrained — insecure property rights systematically suppress long-horizon investment. Geography focuses on disease burden, agricultural endowment, and landlockedness. Culture focuses on the radius of cooperation: can strangers contract with each other cheaply?
The difficulty is identification: all three are highly correlated, and all three may be effects rather than causes. The institutionalist workaround is an instrumental variable — a historical variable that affects income today only through institutions. Acemoglu, Johnson and Robinson (AER, 2001) used European settler mortality in the colonial period: where mortality was high, Europeans could not settle and tended to build extractive institutions that persisted, while mortality itself no longer acts directly on income today. The three received the 2024 Nobel Prize in economics for work on how institutions are formed and affect prosperity.
Strongest institutionalist argument: the natural experiment. Identical geography and cultural tradition, split by institutions, produce sharply different outcomes — the hardest case for the other two schools to answer.
Strongest opposing argument (Glaeser, La Porta, López-de-Silanes and Shleifer, 2004): standard institutional measures such as "constraints on the executive" are volatile and mean-reverting, which makes them look more like a record of outcomes than a deep constraint; human capital is far more stable. Their alternative narrative: poor countries first improve growth and education through policy, and institutions follow. Albouy separately questions the data itself: of the 64 countries in the AJR sample, 36 had mortality rates imputed from other countries.
Trade-off: the institutional view offers a lever you can act on, at the cost of an exclusion restriction that cannot be tested directly; the critics are more careful statistically, at the cost that "human capital first" faces the same reverse-causality problem.
Equating institutions with legal text, and so assuming that copying a constitution transplants the institution. An institution is an equilibrium in everyone's expectations: what the text says matters less than what happens if it is violated, who enforces it, and who constrains the enforcer. The other misreading is treating the three schools as mutually exclusive — they operate on different time scales: geography sets initial conditions, institutions set incentives over decades, policy sets output over a few years.
Chalmers Johnson, studying Japan's MITI in 1982, proposed the concept of the developmental state: a bureaucracy with relative autonomy that actively steers resources toward chosen industries. What later scholarship emphasises is not the subsidy itself — subsidies exist everywhere — but the discipline attached to it.
The core claim of Amsden and Wade is export discipline: credit and foreign exchange are conditioned on export performance. The elegance is that exports are a performance metric that is hard to game — a domestic market can be sustained by protection, but foreign buyers do not place orders because you are close to an official. Engineering analogy: attach an objective metric and an automatic circuit breaker to the subsidy, or it degenerates into pure rent distribution.
| Path | Approach | Cost |
|---|---|---|
| South Korea | Large business groups as the vehicle; credit rationing tied to export targets; rapid entry into heavy, chemical and electronics industries | Highly concentrated economic and political power; the dependence on external debt and the banking system was exposed in the 1997 Asian financial crisis |
| Taiwan | Small and medium firms predominate; public research bodies (such as the Industrial Technology Research Institute, founded 1973) carry technology absorption and spillover | Limited scale per firm, making entry into very capital-intensive industries slower |
| Latin American import substitution | Also used protection to nurture domestic industry, but the performance standard was import volume replaced, not export competitiveness | No external test, so protection was hard to end and industries stayed dependent on policy |
The market-fundamentals view: the World Bank's 1993 report The East Asian Miracle attributed these economies' high growth over 1965–1990 to macroeconomic soundness, investment in human capital and an export push, and warned countries without strong state capacity against copying industrial policy — the risk of picking the wrong industry and of capture often exceeds the gains.
The industrial-policy view: Amsden, Wade, Chang and others argue the report systematically understated targeted intervention. Wade's 1996 essay on "paradigm maintenance" further notes that the study was pushed and funded from the Japanese side, and that the product was a compromise between two positions.
A third objection: Krugman's 1994 "The Myth of Asia's Miracle", drawing on growth accounting, argued that East Asian growth came mainly from a surge in factor inputs rather than productivity gains, and so could not continue indefinitely; the rebuttal is that mobilising and allocating factors efficiently over decades is itself an institutional achievement.
"Industrial policy = government picking winners." The literature's emphasis is at the other end: how to exit — using a verifiable performance metric to stop supporting failures. Picking winners is hard; cutting losses in time is harder, and the latter is what kills most failed cases. The other misreading is treating East Asia as a template: it also had a relatively equal starting point after land reform, a particular geopolitical position during the Cold War, and open Western markets at the time — external conditions that policy does not determine.
Ask first: which constraint is aid trying to relax? Early theory assumed poor countries were stuck on a savings and capital gap, so an injection of funds would start growth. If the real constraint is implementation capacity or unstable rules, an injection is not only ineffective but can have side effects — large external funds resemble resource rents and weaken the chain running from "revenue depends on domestic taxation" to "therefore the government must answer to taxpayers".
A second mechanism is often overlooked: aid is disbursed by donors project by project, so the recipient government's attention shifts toward satisfying donors rather than its own voters. The objective function has been replaced from outside.
| Path | Representative research | Findings and costs |
|---|---|---|
| Cross-country macro regression | Burnside and Dollar (2000): aid raises growth in countries with a good policy environment | Easterly, Levine and Roodman (2004) redid it with expanded data and found the result not robust — small samples, mutually endogenous variables |
| Big push | Sachs argues for a large one-off injection to escape the poverty trap | Easterly's critique is "planners vs searchers": top-down designs lack feedback and accountability; village-level projects were also hard to attribute for want of a pre-specified control group |
| Randomised controlled trials | Banerjee, Duflo and Kremer received the 2019 Nobel Prize in economics for the experimental approach, which traces back to school experiments in Kenya in the mid-1990s | High internal validity — it can establish whether a given intervention works; the cost is limited external validity, since working in one place does not mean working in another setting |
Strongest argument for: evaluating "aid" as a single object is the wrong question. Vaccination, disease control and similar technical interventions have a clear causal chain and measurable returns; humanitarian relief was never aiming at growth in the first place. Using growth regressions to dismiss them is the wrong yardstick.
Strongest argument against: decades and trillions of dollars of aid have left no stable, identifiable growth effect in macro data, and that itself needs explaining. Aid also bypasses domestic fiscal and bureaucratic systems, so saving lives in the short run may postpone the building of state capacity.
Trade-off: humanitarian aid buys lives now, at the cost of not necessarily changing structures; development aid targets structural change, at the cost of being hard to verify and easier to capture politically. The two have different goals but are routinely conflated.
"This project works ⇒ aid works" is a level error: whether it aggregates to the national level depends on whether general-equilibrium effects appear at scale (price changes, talent drawn away). The reverse error is just as common: "macro regressions show nothing ⇒ aid is useless" — noise is enough to drown a real but moderate effect, and failing to detect is not the same as absence.