Institutions and Growth

Frontier If institutions really are the first cause of national income — which is the claim this literature exists to defend and which its own instrument cannot yet carry — then the prize is a lever no other development input offers: rewrite the rules and the income follows. Acemoglu, Johnson and Robinson (AJR) put a number on that lever. In their 2001 sample of 64 former colonies, moving a country from Nigerian institutions to Chilean ones is estimated to raise income “in the long run — as much as a 7-fold increase in Nigeria’s income,” against an observed gap in which Chile is over 11 times as rich. No vaccine, no crop, no road delivers a multiplier of that size in the cross-country data.

Established The magnitude problem it addresses is real and large. The richest 20% of countries are now around 30 times richer than the poorest 20%, and by the Nobel committee’s framing per-capita income in the United States runs roughly 50 times that of the poorest countries. Established The correlation the whole field is built on is not in dispute: across the AJR sample, over 50% of the variation in income per capita is associated with a single index of protection against expropriation. What is in dispute — and it is the whole ballgame — is whether that association is a lever you can pull or a mirror of prosperity already achieved.

Frontier Who collects the upside is a policy question that only has an answer if the causal claim survives. If institutions cause growth, the beneficiaries are the world’s poorest states, because the input is cheap in principle — rules, not capital — and the counterfactual is enormous. If instead institutions are a proxy for human capital (the Glaeser line) or for a disease-free geography (the Sachs line), the same reforms deliver little, and a generation of governance conditionality attached to aid was priced off a coefficient that measurement error may have manufactured.

Frontier The condition on all of it is one the record states against itself: the causal estimate rests on the settler-mortality instrument, and David Albouy’s published comment shows that 36 of the 64 mortality observations are assigned from countries other than the ones they are used for. The 2024 Nobel settled the prize, not the identification. Section 5 holds the data problem in full, and section 7 names the one exercise — a pre-registered, blind reconstruction of the series — that would tell you whether the 7-fold number is a finding or an artefact.

Established The subject is a single empirical claim, not a school of thought: that the quality of a country’s economic and political institutions is a deep, causal determinant of its long-run income. This is distinct from institutional design — the engineering of specific mechanisms, auctions and constitutions, covered in Institutional Design. Here the object is the macro correlation and the fight over whether it is causal.

Frontier The identification problem is the reason the field exists. Rich countries have good institutions and good institutions appear in rich countries, so a raw correlation cannot tell you which way the arrow runs, or whether a third thing — geography, culture, human capital, colonial luck — produces both. The standard fix is an instrumental variable (IV): find something that moves institutions but touches income through no other channel, and the two-stage regression recovers the causal effect. The entire causal claim is only as good as that instrument.

Established The instrument that made the field is settler mortality. AJR’s 2001 paper, The Colonial Origins of Comparative Development, argued a three-link chain: the mortality Europeans faced in a colony determined whether they settled; whether they settled determined whether they built inclusive (settler) or extractive institutions; and those early institutions persisted to shape income today. Settler mortality, being a fact of 18th- and 19th-century disease ecology, is assumed to affect modern income only through the institutions it seeded — the exclusion restriction. Established The mortality figures come largely from Philip Curtin’s compilations of death rates among European soldiers in overseas garrisons, supplemented for Latin America by Gutierrez’s data on the mortality of bishops.

Frontier Three live objections define the territory, and this brief steelmans each rather than picking a winner. First, Albouy: the mortality data are mis-assigned and incomparable, and once corrected the instrument is too weak to identify anything. Second, Glaeser, La Porta, Lopez-de-Silanes and Shleifer: what settlers carried was human capital, not constraints on government, so the instrument — if it works at all — may identify the effect of schooling, not of institutions. Third, Sachs: geography, especially malaria, affects income directly, which if true breaks the exclusion restriction outright. Frontier AJR and allies (Rodrik–Subramanian–Trebbi, Auer) have answered each, and the answers are serious. The result is not a settled science but a decade-long, unresolved argument that a 2024 Nobel honoured without closing.

Established The correlation is robust and large; every serious party to the debate agrees on it. In AJR’s baseline, the ordinary-least-squares coefficient of the expropriation-risk index on log GDP per capita is 0.54 (standard error 0.04), and over 50% of income variation tracks that one index. Whether you call the index “institutions” and whether the relationship is causal are the questions; the association itself is not one of them.

Established The first-stage correlation, in the original data, is also strong. Regressing institutions on log settler mortality gives a coefficient of −0.61 (standard error 0.13), and mortality faced more than a century ago “explains over 25 percent of the variation in current institutions” (first-stage R² 0.27). Frontier The 2SLS estimate of institutions on income is 0.94 (standard error 0.16) — larger than the OLS estimate, which AJR read as evidence that measurement error, not reverse causality, dominates the raw correlation. This is the single most-cited result in the deep-determinants literature and the one the rest of this section interrogates.

Established The number that anchors the policy imagination is the Nigeria–Chile counterfactual. AJR write that improving Nigeria’s institutions to Chilean levels could, in the long run, yield as much as a 7-fold increase in income, adding that the 7-fold figure “is an upper bound” inflated by measurement error. Reported as an upper bound in the paper, it is routinely cited as a point estimate elsewhere — a drift worth flagging because it is where a hedged academic result becomes an unhedged policy claim.

Frontier The independent strand is the “reversal of fortune.” AJR’s 2002 companion paper shows that among colonised regions, those most urbanised and densely populated in 1500 are relatively poorer today — a reversal that runs opposite to any simple geography-is-destiny story, since a fixed geographic endowment cannot flip sign, and that occurred specifically around industrialisation. The Nobel committee treats this as the load-bearing evidence precisely because it does not depend on the settler-mortality instrument. Frontier It is genuine independent support for the institutions story, though it identifies a correlation with 1500 conditions rather than a clean causal effect, and rival readings (which crops and technologies the New World could adopt; who migrated where) remain in play.

Frontier The pro-causal case does not rest on AJR alone. Rodrik, Subramanian and Trebbi’s Institutions Rule horse-races institutions against geography and trade using settler mortality among its instruments and reports that “institutions trump geography and openness”: a one-standard-deviation rise in institutional quality is associated with a 2-log-point (about 6.4-fold) income difference, the geography coefficient turns small and insignificant (−0.70), and the first-stage F-statistic on their instruments reaches 41.7 — far above conventional weak-instrument thresholds. Frontier That F is computed on essentially the disputed data, so it certifies instrument strength conditional on the data being right, which is exactly what Albouy denies.

Established The 2024 Sveriges Riksbank Prize in Economic Sciences went to Acemoglu, Johnson and Robinson “for studies of how institutions are formed and affect prosperity.” The award is a fact; it is also, read against the record below, an endorsement of a research programme whose central identification remains contested in the same journal that published its founding paper. The two “colonial origins” and “reversal of fortune” papers have together drawn over 25,000 citations, which measures influence, not correctness.

Frontier Question one: is the settler-mortality instrument strong enough to identify anything once the data are cleaned? This is Albouy’s charge and it is the sharpest of the three. In his AER comment (published 2012) and the working paper behind it, he documents that “36 of the 64 countries in the sample are assigned mortality rates from other countries, often based on mistaken or conflicting evidence.” Frontier He builds two corrected series — a “high” revision (64 countries, 47 changes) and a “low” revision (43 countries, 34 changes) — and reports that the significance of mortality in the first stage “is never very strong, and in fact can be rejected at a significance level of 5% in all cases,” so that the 95% Anderson–Rubin confidence regions become unbounded, in some specifications spanning the entire real line. If that holds, the IV identifies nothing: any effect size from minus to plus infinity is consistent with the data.

Frontier Question two: even granting a strong first stage, does the instrument satisfy the exclusion restriction? Sachs says no. He shows that a malaria-ecology index enters the income equation with a large, significant negative coefficient (about −1.43, t ≈ −4.0 in AJR data) even after institutions are controlled, while the institutions coefficient stays large — evidence that disease burden depresses income through channels (labour productivity, human-capital accumulation, investment) other than institutions. Frontier If a colony’s historic settler mortality is correlated with its present disease environment, then the instrument affects income directly and the exclusion restriction fails, biasing the 2SLS estimate by an unknown amount and sign.

Frontier Question three: is “institutions” even the right label for what the instrument moves? Glaeser and co-authors argue that settler mortality predicts today’s years of schooling (correlation −0.73) more tightly than it predicts constraints on the executive (−0.54), and that log population density in 1500 correlates with schooling at −0.93 versus only −0.35 with executive constraints. Frontier Their inference: what Europeans brought was “themselves, and therefore their know-how and human capital,” so the instrument — if valid at all — may be identifying the return to human capital wearing an institutions label. When they run IV with both predicted schooling and predicted executive constraints, schooling is the significant determinant of income and executive constraints is not.

Frontier Question four: are the standard institutional measures even measures of institutions? Glaeser et al. press a conceptual point that survives independent of the settler-mortality quarrel: expropriation risk and Polity executive-constraint scores are outcomes — what governments did — not rules that constrain what they can do. Frontier They note the measures are about twice as volatile as human-capital measures and mean-revert sharply; Haiti’s executive-constraints score swings from 1 to 6 to 1 to 7 to 3 within a single decade. A variable that moves that fast is recording political weather, not the durable “rules of the game” the theory needs.

Speculative The synthesis question nobody has answered: how much of the 2SLS estimate would survive a joint attack? Each critique has been rebutted in isolation. What has never been done is a single specification that cleans the data to a pre-registered standard, admits a direct geography channel, and separates human capital from institutions at once. Handwave The intuition of partisans on both sides — that their preferred adjustment is the decisive one — is exactly the belief the four-flag discipline exists to hold at arm’s length until such a specification is run.

Established The binding bottleneck is not theory or computation; it is the provenance of 64 numbers. The settler-mortality series is a hand-assembled compilation from heterogeneous 18th- and 19th-century sources, and the entire causal edifice sits on it. Albouy’s central documentary finding — that 36 of the 64 observations are assigned from other countries — is a claim about a spreadsheet, checkable against archives, and it is the rare bottleneck in social science that is a bookkeeping problem rather than a conceptual one.

Frontier Bottleneck one: incomparable populations. Albouy shows the series mixes “mortality rates from populations of laborers, bishops, and soldiers — often on campaign — combined in a manner that favors the hypothesis.” Troops in barracks die at very different rates from troops on campaign; forced labourers and clergy are different populations again. Frontier The Latin-American figures rest on Gutierrez’s bishops, whose estimated death rates carry standard errors of roughly 7.8 to 8.6 per thousand — wide enough, Albouy argues, that the regional differences the instrument relies on are not statistically distinguishable.

Frontier Bottleneck two: coding choices with degrees of freedom. Where a source gives a range, which figure enters? For the Congo, Albouy notes AJR use a peak rate of about 240 per thousand rather than an average nearer 100. Frontier AJR’s own robustness move — capping mortality at 250 per thousand (justified by Tulloch’s 19th-century assessments) — actually strengthens the first stage, raising the coefficient from −0.61 to −0.94, which they present as evidence the result is not driven by outliers. The same fact reads two ways: robust to capping, or sensitive to which capping rule you pick.

Frontier Bottleneck three: researcher degrees of freedom on both sides. AJR reply that Albouy’s own recoding is not clean either — that he codes New Zealand and Hong Kong as non-campaign against Curtin’s explicit discussion of battles and a standing field force — and that a campaign dummy is not statistically significant in either stage (coefficients around −0.72 with standard errors near 0.46). Frontier They report that dropping Gambia from Albouy’s preferred 28-country subsample returns results “very similar to those in AJR (2001),” and that alternative naval-station benchmarks for Latin America keep the first stage “robustly significant.” The disagreement is not about a formula; it is about which hand-coded judgements are defensible, which is precisely the kind of dispute that pre-registration and blind coding exist to remove.

Speculative The deepest bottleneck is unobservable and permanent: you cannot randomise a country’s institutions. Every estimate in this literature is observational, and the settler-mortality instrument was celebrated because it looked like a natural experiment that substituted for the impossible randomised trial. Handwave If the instrument fails, there is no obvious replacement of comparable historical reach, and the causal question may simply not be answerable to the precision the 7-fold counterfactual pretends to.

Frontier The result depends first on the primary demographic archives that Curtin drew from. Regimental medical returns, colonial office statistics, actuarial life tables and 19th-century medical geographies are the raw material; the instrument is a summary statistic over them. Any reassessment depends on getting back to those sources rather than re-litigating the summary.

Frontier It depends on a settled measure of “institutions,” which the field does not have. The expropriation-risk index is a commercial country-risk rating; Polity executive-constraints scores are hand-coded from political histories. Glaeser et al.’s critique — that these are outcomes, not constraints — means the dependent-side variable is as contestable as the instrument. Frontier Progress depends on rule-based measures (constitutional provisions, statutory constraints) that are, in their data, only weakly correlated with income at all.

Frontier It depends on separating three confounded deep determinants: institutions, human capital and geography. These co-vary so tightly in the cross-section — settler mortality predicts schooling almost as well as it predicts institutions — that no single historical instrument can cleanly attribute income to one of them. Speculative The dependency here is on a source of variation that moves institutions while holding human capital and disease ecology fixed, and no one has produced one.

Frontier It depends on the econometrics of weak instruments, which matured after the original paper. Anderson–Rubin confidence sets, robust to weak instruments, are what turn Albouy’s critique from a quibble into a demonstration that the confidence region can be the whole line. Established The methodological tools to adjudicate instrument strength now exist and are standard; the dispute is over which dataset to feed them.

Frontier Finally it depends on the wider deep-roots literature it sits inside. Reversal-of-fortune evidence, the persistence of pre-colonial and colonial institutions, and the human-capital-persistence work are all partial substitutes and partial checks. The institutions-cause-growth claim is one node in a network of contested history, and it borrows credibility from, and lends fragility to, the rest.

Frontier The decisive experiment is a data exercise, not a new theory, and it is unusually well-specified because both sides already agree what the object is. Reconstruct the settler-mortality series from the primary archives — Curtin’s regimental returns, actuarial life tables, colonial medical reports — under a pre-registered coding protocol, with each country’s mortality assigned blind to that country’s modern income and institutions, and then re-run the first stage and the instrumental-variable regression. The point of pre-registration and blinding is to kill the one charge no rebuttal can answer post hoc: that the coding choices, on either side, were made in a manner that favours a conclusion.

Frontier What makes it decisive is that it targets the exact quantity in dispute. If a blind, pre-registered series still yields a first stage with an F-statistic above conventional thresholds and a bounded Anderson–Rubin interval, Albouy’s critique is answered on its own terms and the instrument stands. If it yields the unbounded confidence regions Albouy reports, the causal claim — not the correlation — falls, and the 2024 Nobel honours a research programme whose flagship identification does not replicate. Speculative Nobody has run it to that standard, which is the single most surprising fact in the subject.

Frontier Partial versions have been run, and they cut toward AJR — but none is blind or pre-registered. Karti Subramanian (2007) rebuilt mortality from British life-insurance actuarial tables (the Assurance Magazine and Journal of the Institute of Actuaries, 1848–1907) and a French medical treatise, on the argument that insured European civilians are closer to “settlers” than soldiers are, and found the instrument survives: first-stage coefficients of −0.74, −0.64 and −0.45 across three constructions, all significant at 1%, with second-stage effects of institutions on income of 0.66 to 0.74. Speculative This is real evidence the instrument is not a pure artefact, but it is an undergraduate thesis, not peer-reviewed, and it explicitly reports no pre-registration and no blind coding — so it cannot settle a charge that is precisely about undisclosed coding latitude.

Frontier A second family of attempts replaces the disputed data rather than cleaning it. Auer (2007) instruments the noisy historical mortality rates with a geographic model of disease determinants and reports that the instrumented rate remains “a highly significant predictor of institutional quality” regardless of which historical dataset is used. Frontier Gooch, Martinez-Vazquez and Yedgenov go further and propose “malaria endemicity” as a “superior instrument,” reporting effects of institutions on income larger than the original. Frontier But this route buys strength at the cost of the exclusion restriction: malaria is exactly the variable Sachs argues affects income directly, so an instrument built on malaria endemicity is strong precisely where it is most vulnerable to the Sachs objection. The two critiques are not independent, and fixing one can worsen the other.

Speculative The complementary test is a direct-channel horse-race with a defensible present-day disease control. Add a well-measured contemporary malaria-ecology index to the second stage and see whether the instrumented institutions effect survives; Sachs’s results predict it should shrink, AJR’s that it should not. Handwave Neither this nor the blind reconstruction requires new hardware, new fieldwork or a single new observation of the present — only archival labour and the discipline to commit to a coding rule before seeing the answer, which is why the absence of the experiment is a fact about incentives, not about feasibility.

Frontier The “engineering” here is archival and statistical infrastructure, not apparatus. A credible reconstruction needs systematic digitisation of the primary demographic sources — regimental returns held in national archives, the actuarial journals Subramanian used, colonial medical geographies — and a documented crosswalk from each raw figure to each country-observation, with the population type (barracks, campaign, civilian, labourer, clergy) tagged rather than silently pooled.

Frontier It needs a pre-registration and blind-coding pipeline that development economics does not routinely run for observational work. Registered Reports and pre-analysis plans are now standard for experiments; they are rare for historical cross-country regressions, where the data are fixed and the temptation to iterate specifications is strongest. The engineering task is procedural: a registry, a locked coding manual, an independent coder blind to outcomes, and a published crosswalk. This is the same discipline that Institutional Design identifies as the missing ingredient across the wider field — measurement committed to before the result is seen.

Established The statistical tooling is already built. Weak-instrument-robust inference (Anderson–Rubin, conditional-likelihood-ratio and related tests), clustered and heteroskedasticity-robust standard errors, and first-stage diagnostics are standard software. No estimator needs inventing; the inputs need cleaning.

Frontier It needs a second, independent instrument to over-identify the model. With one instrument the exclusion restriction is untestable by construction. A genuinely independent second instrument for institutional quality — not another disease proxy — would allow an over-identification test and is the piece of “equipment” the field most lacks. Speculative Candidates (legal-origin, linguistic or precolonial-institution variables) each carry their own exclusion problems, so the requirement may be unmeetable, which would itself be an informative result.

Frontier The nearest sibling is institutional design, and the contrast is the point. Institutional Design covers the sub-field where mechanisms are engineered and outcomes measured — matching markets, spectrum auctions — and it finds that designs which remove a decision work and are measurable, while designs that move a decision fail and are not. This brief is the macro counterpart: whether the aggregate stock of good institutions causes national income, a claim that cannot be randomised the way a school-choice algorithm can.

Frontier The deep-roots family of empirical history is the immediate neighbourhood: reversal-of-fortune and persistence studies, the legal-origins literature, colonial-institution persistence, and the human-capital-persistence work that Glaeser’s critique feeds into. Each shares the identification problem and the shared dependence on hand-assembled historical data.

Frontier The econometrics of weak instruments is a methodological adjacency that did as much to shape this debate as any new data. Anderson–Rubin inference and the Stock–Yogo weak-instrument thresholds are what let Albouy state his critique as a confidence region rather than a hunch.

Frontier On the policy side, the debate touches growth diagnostics, development finance and governance conditionality. It is adjacent to Economic Resilience, Innovation Ecosystems and Technological Sovereignty, each of which inherits a prior about how much rewriting rules can move outcomes — a prior this literature is supposed to supply and cannot yet defend to the precision those debates assume.

Frontier It is also adjacent to the meta-science of the field itself. Science of Science and Research Productivity Decline ask how disciplines know what they know; a Nobel-honoured result whose identification is openly disputed for over a decade is a case study those briefs would recognise.

Frontier The result has already moved real money, which raises the stakes on whether it is right. The proposition that institutions cause growth underwrote two decades of governance conditionality, rule-of-law programming and “good governance” benchmarking in development finance. If the causal coefficient is an artefact of mis-assigned mortality data, the intellectual warrant for a large slice of that spending was thinner than its confidence suggested.

Frontier The institutional failure that keeps the debate open is the absence of a registered-reports norm for observational history. The decisive experiment is cheap, needs no new data collection in principle, and would resolve a headline dispute — yet nobody has funded or scheduled it to a blind, pre-registered standard. Frontier The incentives explain the gap: a clean replication that confirms AJR adds little to a résumé, one that overturns them invites a decade of rebuttal, and neither is as publishable as a new instrument or a new dataset. The missing institution is a venue that rewards the boring, decisive check.

Established The scholarly institutions worked exactly as designed in one respect. Albouy’s critique was published in the American Economic Review — the same top journal, with AJR’s reply alongside — and the replication data are archived and open. The dispute is legible precisely because the field’s comment-and-reply machinery functioned; the failure is downstream, in the absence of the decisive follow-up.

Frontier The Nobel is itself an institutional act with epistemic weight. Awarding the prize “for studies of how institutions are formed and affect prosperity” confers a settledness the underlying identification has not earned. Speculative The likely effect is to reduce, not increase, the chance the decisive replication gets funded, because prizes signal that a question is closed. That is a governance-of-science problem the award cannot see about itself.

Frontier The first ethical hazard is a policy multiplier applied to real countries on a contested coefficient. A 7-fold counterfactual stated in a paper as an upper bound, and cited elsewhere as a target, licenses advice to low-income governments — restructure institutions and incomes will follow — whose evidentiary base is exactly what is in dispute. If Glaeser is right that human capital is the first cause, the same advice deprioritises schooling in favour of governance reforms that deliver less.

Frontier The colonial data carry their own ethical weight. The instrument is literally built from the death rates of European soldiers, clergy and colonial officials; the framing treats places where Europeans died as places that got “bad” institutions, which encodes a settler-eye view of history into the econometrics. Frontier That does not make the estimate wrong, but it means the variable’s meaning — and the populations whose suffering is and is not counted — deserves the scrutiny the technical debate has crowded out.

Speculative There is a fairness question in the determinism itself. A strong reading of the institutions-cause-growth claim implies that a country’s income is substantially set by colonial-era choices it did not make, which can read either as fatalism or as a case for reparative transfers. Handwave The brief takes no position on that; it notes only that a contested coefficient is doing heavy normative work in arguments well outside economics.

Frontier The meta-ethical issue is honesty about uncertainty. The correlation is real and the causal claim is genuinely open; presenting the second with the confidence of the first — which prize citations and secondary summaries tend to do — is the specific failure this brief exists to flag. The Frankel and Taylor Nobel commentaries, both of which present the instrument as vindicated without mentioning Albouy, are honest scholarship and also evidence of how fast an open question is narrated as closed.

Frontier If the causal claim is true, the civilizational implication is optimistic and specific: national poverty is not destiny, and the binding input is institutional rather than material. A poor country is then poor because of rules that can, in principle, be changed, not because of an immovable geography or a fixed resource base — the reading the Nobel committee foregrounds and the one that makes the prize feel consequential.

Frontier If the Glaeser reading is closer to right, the implication inverts: the deep input is human capital, and institutions are something a society grows into as it educates itself. That is the Lipset hypothesis, and Glaeser et al. report timing evidence consistent with it — initial schooling predicts subsequent institutional improvement, while initial institutions do not predict subsequent schooling. Speculative A civilization that believed this would invest in people first and expect governance to follow, roughly the East-Asian development sequence rather than the governance-conditionality sequence.

Frontier If the Sachs reading holds, geography reasserts a partial claim, and the policy frontier moves to disease control and physical endowment. The three readings are not mutually exclusive, and the honest civilizational statement is that we do not know their relative weights, which is an uncomfortable thing to say about the most-cited result in the economics of development.

Handwave The largest implication is epistemic. This is the cleanest available test case for whether social science can deliver a durable causal law about something as aggregated as national income. If a question this central, this well-funded and this instrumented cannot be resolved to better than an unbounded confidence interval after twenty-five years, that is itself a finding about the limits of observational macro-social-science — and a caution against every downstream claim that borrows its confidence.

These horizons track publication, replication and funding cycles, not any technology — every input already exists in an archive:

• 10 yr: Frontier The most likely development is more of the same — new instruments (malaria endemicity, precolonial variables) and new deep-roots datasets — rather than the decisive blind reconstruction, because the incentives described in section 10 still hold. Speculative A registered-reports replication of the settler-mortality first stage is entirely feasible within this window and would resolve the Albouy dispute; whether one is funded is the open question. Frontier Expect the 2024 Nobel to raise citations and lower, not raise, the appetite for the boring check. The measurement discipline that Institutional Design calls for across the field applies here with unusual force.
• 25 yr: Speculative The plausible split is that the correlation remains textbook-canonical while the causal identification is quietly downgraded in the technical literature to “suggestive,” the way many first-generation IV results have been. Speculative Machine-assisted digitisation of colonial archives may make a full, transparent reconstruction cheap enough that someone finally runs it, blind or not.
• 50 yr: Speculative If the human-capital and institutions channels are ever separated with genuinely exogenous variation — the one thing nobody has — the field either consolidates around a measured relative weight or concludes the two are not separable in the historical record. Handwave Which of those happens is not forecastable now.
• 100 / 250+ yr: Handwave Beyond useful forecasting. Whether “institutions cause growth” is even the right unit of analysis — against, say, a future data environment that tracks sub-national institutional variation continuously — is unknowable, and a claim about it would be exactly the overreach this brief flags in others.

• Depends on Nothing on this map blocks it, and that is the finding: the binding constraints are archival provenance and pre-registration discipline, both decades-old capabilities, not any result another brief is waiting to produce. Its nearest kin, Institutional Design, shares the diagnosis — that the field’s limit is measurement committed to before the answer is seen — but supplies no input this brief waits on.
• Requires (not on this map) A pre-registered, blind reconstruction of the settler-mortality series from primary archives, which is the one exercise that could answer Albouy on his own terms and which nobody has run. A second, genuinely independent instrument for institutional quality, without which the exclusion restriction stays untestable. Exogenous variation that separates human capital from institutions, the missing ingredient behind the Glaeser critique. A registered-reports track that rewards decisive replication of observational history rather than novelty. And the modest funding for the primary-archival and actuarial digitisation the whole reassessment rests on. All five are methodological or institutional capabilities, not discoveries.
• Enables A defensible prior for every downstream claim that assumes rewriting rules moves national income — growth diagnostics, governance conditionality, development finance. No typed enabling edge is asserted, because on current evidence the field cannot say how large the institutions-to-income effect is, and a typed edge claiming it would be exactly the overconfidence this brief documents.
• Adjacent The deep-roots empirical-history family; the econometrics of weak instruments; growth and development economics; and within this map Institutional Design, Economic Resilience, Innovation Ecosystems, Technological Sovereignty and Science of Science.

Frontier “The 2024 Nobel proved that institutions cause growth.” The prize was awarded “for studies of how institutions are formed and affect prosperity” and rewards a body of work; it is not a referee report certifying the settler-mortality identification. The founding paper’s instrument was challenged in the same journal in 2012 and the challenge has not been decisively answered. A Nobel settles a prize, not an exclusion restriction.

Established “Albouy showed the institutions-income correlation is fake.” He did not, and did not claim to. The OLS correlation — over 50% of income variation tracking the institutions index — is not in dispute. Albouy’s target is narrower and deadlier: the causal estimate that the IV is supposed to license, which he argues collapses into an unbounded confidence region once the data are cleaned.

Frontier “The reconstruction has been done and AJR won.” Reconstructions exist (Subramanian’s actuarial series, Auer’s geographic-model instrument) and they do cut toward AJR, but none is blind or pre-registered, and one leading alternative — malaria endemicity — buys first-stage strength at the direct cost of the Sachs exclusion objection. The decisive version has not been run.

Frontier “The critiques cancel out, so the result stands.” They do not cancel; they compound. Weak-instrument, exclusion-restriction and wrong-variable objections attack different links in the chain, and an answer to one (instrument malaria to fix strength) can worsen another (malaria has a direct income effect). No single specification has yet met all three at once.

Speculative “This means institutions do not matter.” That is the opposite overreach and the brief rejects it too. The reversal-of-fortune evidence, the persistence literature and plain historical narrative all point at institutions mattering somewhere in the causal story. Handwave What is unproven is the specific, large, cleanly-identified causal coefficient the field is famous for — and the honest position is that its size is currently unknown, not that it is zero.

Frontier “An instrumental variable removes endogeneity, so the 2SLS number is causal by construction.” Only if the instrument is both strong and exogenous. The last twenty-five years of this debate are a case study in how an IV can fail both tests quietly — strong on disputed data, exogenous only if geography has no direct effect — while the headline number keeps being quoted as though the method guaranteed the result.