Has economics become a unified science or a collection of specialties?
A philosopher called economics “inexact and separate.” The methodologists say it finally has one shared method. The Austrians and the MMT people say they were never invited. All three are describing the same discipline — and all three are right about a different part of it.
以辩论图谱查看Where this lands: Economics is method-federated, substantively specialty-fragmented, and heterodox-peripheral. Since roughly 1990 the discipline has converged on a genuine shared method — optimization-under-constraints plus the credibility revolution’s standards for what counts as a publishable empirical claim. That convergence is real, and it is recent. But the substantive frames — what objects exist, what variables matter, what question is even worth asking — diverge enough across behavioral, development, finance, macro, and mechanism design that “loosely federated specialties sharing a method” describes the discipline better than “one unified science.” And a heterodox periphery — Austrian, Post-Keynesian, MMT, modern Marxian — uses different methods entirely and does not get absorbed. The honest one-word answer is neither “unified” nor “collection of specialties.” It is federated.
The unification thesis at its strongest
“The credibility revolution in empirical economics … has shifted the focus of work toward research designs that can deliver convincing causal estimates. What unites the field now is less a shared theory than a shared standard for what counts as evidence.”
— Joshua Angrist & Jörn-Steffen Pischke, Journal of Economic Perspectives, 2010
Here is the strongest case that economics is one science: not because economists agree about anything substantive, but because they finally agree about what a real answer looks like. A labor economist, a development economist, and a finance economist can now read each other’s papers and judge them by the same standard. In 1970 that was not true. The question is whether sharing a method is enough to make a discipline one.
The machinery underneath the unification claim has two parts, and both are recent. The first is a shared modeling grammar: optimization under constraints. Agents have objectives, they face constraints, and the model solves for what they choose. Game theory generalized this to strategic settings — what you do best depends on what I do — and information economics extended it to the cases where the constraint is what you know. By the 1990s this grammar had spread so widely that a paper in almost any subfield could be read as some agent optimizing something subject to something. That is not a small fact. It is the closest thing economics has to a common language.
The second part is the credibility revolution — the shift, from the late 1980s onward, toward research designs that can defend a causal claim: natural experiments, instrumental variables, regression discontinuity, randomized trials. Before it, an empirical economics paper argued about which regression to run; after it, the argument moved to whether the design isolates a real causal effect. This standardized what counts as a finding. A development RCT, a labor-supply estimate from a policy discontinuity, and a finance event-study now answer to the same evidentiary bar — which is exactly what makes cross-subfield reading possible. The historian’s version of this story, from the post-2008 vantage where the convergence became visible as a convergence, lives in History of Economic Thought Ch.17 (Modern pluralism).
What was true methodologically in 1970 is not true in 2025. The discipline that publishes today shares a method-floor that the discipline of fifty years ago did not — and that floor was poured deliberately, by people who can be named and dated. The formal optimization apparatus itself sits in Economics Ch.11 (Advanced microeconomics); this walkthrough takes the apparatus as intellectual history, not as something to derive.
The case for one science, made by the people who believe it
“The methodology of economics … is far more unified than the variety of its subject matter would suggest. The unity lies in a shared commitment to deriving implications from models of optimizing agents and testing them against data.”
— Kevin Hoover, The Methodology of Empirical Macroeconomics
Hoover’s claim, argued at full strength, is that the appearance of fragmentation is superficial. Behind the different subjects sits one method: build a model in which agents optimize, derive what the model implies, take the implications to data, and discipline the model against what you find. The credibility revolution did not invent this commitment — it standardized it, by fixing what a convincing test looks like. Before standardization, two economists looking at the same correlation could disagree forever about what it meant. After it, they share a vocabulary for adjudicating the disagreement. That shared vocabulary is what makes the work of one subfield legible to another, and legibility across subfields is exactly what you would want a unified science to have. The discipline cohered around a method even where it could not cohere around a theory.
“The notion that institutions … are a fundamental cause of economic growth has now moved to the center of how economists across many fields think about long-run development.”
— Daron Acemoglu & James Robinson, in the spirit of Why Nations Fail
Acemoglu pushes the unification claim further than Hoover — not just method, but substance. On his reading, institutions-as-variables now appear in growth theory, in development, in comparative politics, in finance, and the underlying framework — agents optimizing under institutional constraints that themselves emerge from agents optimizing — holds across all of them. The same modeling instinct that explains why a firm under-invests when property rights are insecure explains why a country does. If that is right, the convergence is not only about evidentiary standards. It is about a shared picture of what drives outcomes: incentives, constraints, and the institutions that set them. The unification is real, it is growing, and you can watch it happen in the topics the top journals choose to publish.
What the thesis actually establishes
Take the method-unification claim first: it is strong and largely correct as a description of how economics publishes after 1990. The shared evidentiary bar is genuine, it is enforced, and it does make subfields legible to one another. That much should be conceded fully — it is the real achievement the rest of this walkthrough has to reckon with, not explain away. Acemoglu’s further claim — that the substance has also converged, around institutions and optimization — is more contested. It reads coherence into a federation that may be more loosely connected than the institutions-as-variables framing suggests: a finance theorist and a development economist can both write down an optimization problem and still be studying objects that have almost nothing to do with each other. So the thesis at its strongest is method-unification plus a growing-but-partial substantive convergence. That is the best case for “one science.” The next stage pressure-tests it from inside the specialties, where the federation looks very different.
But the unification claim is made from above — from the methodologist’s vantage, where every subfield resolves into the same grammar. From inside any single subfield, the view is different. What does “one economics” look like to a behavioral economist who has spent thirty years showing that the assumptions the macro modeler makes about people are empirically false?
For the methodological convergence as its own thread — what the credibility revolution actually delivers and what it leaves open — see the deeper sibling walkthrough The credibility revolution: what does economics actually know?.
The specialties seen from the inside
“Economists tend to use the word ‘rational’ to mean ‘logically consistent.’ … The definition of rationality that economists use is much too narrow … The world makes much more sense than standard economics would predict, but only because people are predictably irrational in ways the standard model leaves out.”
— Richard Thaler, Misbehaving: The Making of Behavioral Economics, 2015
You just heard the case that economics is one method-unified science. Now stand inside a single subfield and ask whether it feels that way. Behavioral economics spent four decades documenting that the rational agent at the center of mainstream models does not exist — and won a Nobel for it — while macro models kept assuming the rational agent anyway. Sharing a method, it turns out, does not mean sharing what you think is true.
Behavioral economics is the cleanest case of a specialty that built its own apparatus and never fully merged. It grew out of cognitive psychology, not economics: Herbert Simon’s bounded rationality, then Kahneman and Tversky’s prospect theory, which replaced the expected-utility curve with a value function kinked at a reference point and weighted by how people actually treat probabilities. For thirty years this evidence accumulated using experimental and survey methods that mainstream economics did not share. When it finally arrived in the mainstream — Thaler’s 2017 Nobel — it arrived as absorption by citation, not as a method merger. Prospect theory is now taught in micro; it is still mostly absent from the macro models central banks run. A field can be honored and quarantined at the same time.
Development economics is the second case, and it cuts the other way: it shares the mainstream’s newest method while asking questions no other subfield asks. The 2019 Nobel to Banerjee, Duflo, and Kremer was the credibility revolution’s endpoint inside development — randomized trials run in villages to test which interventions actually move outcomes. By the unification logic, development should therefore feel maximally integrated. It does not. The questions development cares about — how do you escape a poverty trap, when does aid work, what coordinates an economy that the market alone cannot — do not appear in finance or macro at all, and the deep lineage it argues with runs through Lewis’s dual-sector structuralism, Prebisch-Singer dependency, and Sen’s capabilities, not through the optimization core. Shared method, separate world.
The structural point is the same in both cases: even where subfields share method — and they increasingly do — they do not share objects or questions. Behavioral cares about cognitive biases; development cares about poverty traps; finance cares about no-arbitrage pricing; macro cares about aggregate stabilization. The shared method is a floor, not a ceiling. It tells you how to argue; it does not tell you what to argue about. And what you argue about is most of what a field is.
The federation, named by the people who live in it
“Economics is best viewed as a separate and inexact science. … Its laws hold only ceteris paribus and only inexactly, and its various branches are unified more by a shared theoretical vision than by a single body of established theory.”
— Daniel Hausman, The Inexact and Separate Science of Economics, 1992
Hausman’s title does the work: separate. He argued in 1992 — when it was far less obvious than it is now — that economics is a loosely federated discipline whose branches share a theoretical vision rather than a single established theory, and whose laws hold only approximately and only other-things-equal. The credibility revolution, on this reading, did not refute him; it proved him more right, by making each subfield’s own evidentiary standards so dominant that they, rather than any shared substantive theory, became what governs translation between fields. The method-floor Stage 1 celebrated is, in Hausman’s frame, a federation norm, not a unification one: a rule for how the provinces talk to each other, not evidence that they have become one country. A federation can have an excellent common language and still be a federation.
“Experimental economics had to develop its own methods, its own conventions about control and replication, because the questions it asked could not be answered by the tools the rest of the field was using.”
— in the tradition of Vernon Smith and Colin Camerer’s Behavioral Game Theory
From inside a specialty, the unified-discipline frame reads as aspirational rhetoric, not lived practice. The experimental economist publishes in experimental journals, presents at experimental conferences, cites experimental peers, and reaches for tools the macro modeler has never used. The numbers bear this out: within-subfield citation vastly exceeds cross-subfield citation in every field of economics, and it has not converged even as the method-floor rose. The day-to-day texture of doing economics is the texture of working inside a specialty whose boundary is a wall of accumulated convention, vocabulary, and shared assumptions about what is interesting. The method you share with the rest of the discipline is real. It is also, most days, beside the point.
Where the specialties leave us
The specialty-fragmentation reading is also strong, and also largely correct as a description of how the discipline actually works. Note what has just happened: Stage 1’s verdict still stands — the subfields really do share a method-floor — and Stage 2’s verdict is equally solid, that they do not share substantive frame, object of study, or normal-science problem set. These are not competing claims to be adjudicated. Hoover’s method-unification and Hausman’s substantive-separation are both true, at different scales of description: zoom out to the level of evidentiary standards and you see one discipline; zoom in to the level of what economists actually study and argue about and you see a federation of specialties. Calling that federation “one science” overcounts the shared commitments. But there is still a third position neither stage has touched — held by traditions that never agreed to the federation in the first place.
Everything so far has been an argument among mainstream economists about whether the mainstream is one or many. But there are traditions inside economics whose disagreement is more fundamental: they reject the shared method itself, not just the claim that it unifies. Are they part of economics, or are they doing something else?
For the behavioral case as its own full thread — did behavioral economics change the discipline or get absorbed by it — see Did behavioral economics change economics?. For rationality as a concept moving across economics, psychology, and philosophy, see What is rationality, across the disciplines that use it?.
The traditions that never agreed to the unification
“The project of mathematising economics … presupposes that social reality is everywhere constituted by systems of isolated atoms in closed systems. But social reality is not like that. The insistence on a single method excludes, in advance, the possibility that the world is open.”
— Tony Lawson, Economics and Reality, 1997
The mainstream split in Stages 1 and 2 was a family quarrel — everyone agreed on the method, they only disagreed about whether the method unified. Lawson is making a deeper move. He says the shared method is itself a substantive philosophical commitment, made before any evidence is gathered, and that it excludes whole ways of seeing the economy. The traditions that refuse the method are not failing at economics. They are doing a different economics, on purpose.
The Austrian tradition is the cleanest parallel lineage, because its break is methodological rather than substantive. From Menger’s founding through Mises and Hayek, the Austrians built economics on praxeology — reasoning deductively from the axiom that humans act purposefully — and on subjective marginalism, where value lives in individual minds, not in measurable aggregates. Mises rejected econometrics on principle: you cannot, he held, run a controlled experiment on a one-time historical event driven by purposeful choice, so statistical inference over economic data answers the wrong kind of question. Part of the tradition did get absorbed — Hayek’s knowledge problem flowed into information economics — but the methodological core, the insistence on deduction-from-action over statistical inference, did not, and could not, because the two methods disagree about what economic knowledge even is.
The Austrians are not alone on the periphery, and the periphery is not a dead end. The post-2008 reckoning re-energized a whole cluster of parallel traditions, each with an apparatus that does not reduce to the federated core. Modern Monetary Theory starts from the accounting of a sovereign currency issuer rather than from an optimizing agent; Post-Keynesians build on fundamental, non-probabilistic uncertainty, which optimization-under-constraints cannot represent because there is no probability distribution to optimize over; Piketty’s distributional dynamics, the Minsky revival’s financial-fragility hypothesis, and a modern Marxian revival treating class as an analytic category each carry their own machinery. None of these translate into the mainstream’s optimization grammar without losing precisely what makes them distinctive. The chapter that stages this whole cluster — and explicitly asks whether the discipline is converging or fragmenting — is History of Economic Thought Ch.17.
Here is the structural point this walkthrough wants to make plainly: the fact that the federated core does not absorb the heterodox periphery is not a defect to be fixed. It is a correct description of what the discipline is. The honest posture toward these traditions is to engage them on their own terms — to read MMT as monetary accounting rather than as a broken IS-LM, to read the Austrians as a deductive program rather than as bad econometricians — not to translate them into the core and declare them refuted by the translation. Non-absorption is the data, not the problem.
The periphery, in its own voice
“The ontological presuppositions of mainstream economics — that the social world is closed, that it consists of isolated atoms whose relations are external — are rarely defended because they are rarely noticed. They are built into the choice of method, and the method is treated as neutral. It is not.”
— Tony Lawson, Reorienting Economics, 2003
Lawson’s argument, taken at full strength, is the sharpest version of the parallel-lineages position, and it is not a complaint about results — it is a claim about what the method assumes before it runs. To model the economy as agents optimizing in a closed system is to assume the social world is closed: that its parts have stable, externally-related properties, that the future is a probability distribution rather than something genuinely open, that macro phenomena reduce to micro ones. Each of those is a contestable claim about the nature of social reality, and the mainstream commits to all three the moment it picks the method — not as a conclusion drawn from evidence but as a precondition for the evidence to be gathered the way it is. The traditions that hold different ontologies — that the future is open, that social structures are real and causal, that emergence is genuine — cannot be translated into the closed-system method without being deleted. They are not bad economics. They are economics that declined a metaphysical bet the mainstream took without noticing it was a bet.
“A country that issues its own currency … can never run out of money. It can’t go broke. … The federal government is not like a household. Once we accept that, the entire conversation about deficits has to start over.”
— Stephanie Kelton, The Deficit Myth, 2020
MMT is the most publicly visible parallel lineage, and Kelton’s framing shows why it does not reduce to the core. The apparatus starts from accounting identities of a sovereign-currency monetary system: government spending credits bank accounts, taxation debits them, bond sales drain reserves rather than fund expenditure, and the real constraint on spending is the inflation that excess demand produces, not the revenue taxes raise. Descending from Lerner’s functional finance and the chartalist account of money as a creature of the state, the framework is internally coherent and its operational accounting is, on its own terms, correct. The standard attempts to refute MMT by translating it into a loanable-funds or IS-LM frame systematically miss what it is doing, because those frames already assume the financing story MMT denies. You cannot defeat a framework by restating it in the vocabulary it rejects — you can only engage it where it actually stands.
What the periphery establishes — and overclaims
The parallel-lineages reading is partly right and partly overstated, and the distinction matters. It is right that the federated mainstream genuinely does not absorb Austrian methodology, MMT’s accounting frame, Post-Keynesian fundamental uncertainty, or modern Marxian class-analysis without losing what makes each distinct; engaging them on their own terms, rather than translating-and-dismissing, is the honest posture. Where the position overstates is in reading the mainstream as philosophically committed to the closed-system ontology. Many working economists treat optimization-under-constraints as a useful approximation they would cheerfully abandon if a better-predicting tool appeared — an “as-if” modeling stance, not a metaphysics of agents. The federation is more pragmatic than the ontological critique credits. So the structural answer is this: the heterodox periphery is real, plural, live, and not going to be absorbed — and whether you call that “economics is fragmented” or “economics has a federated core plus a non-absorbing periphery” is a labeling choice. This walkthrough takes the second framing, because it keeps the core’s real cohesion and the periphery’s real separateness both in view.
So we now have three readings of the discipline, each made at full strength by people who believe it: method-unified (Hoover, Acemoglu), substantively separate (Hausman, Thaler), parallel lineages (Lawson, Kelton). All three are inside the legitimate conversation. None is a strawman. What, then, is the synthesis — and is “synthesis” even the right word for what holds them together?
For the live public debates these traditions drive, see the controversial-question walkthroughs What did the Austrians get right?, 货币到底是什么？, and Is MMT crank or serious?.
What the verdict actually is
“Economics is not a set of universal truths but a collection of models. … The right answer to almost any question in economics is: it depends. The discipline’s strength is the variety of its models; its weakness is the temptation to crown one of them king.”
— Dani Rodrik, Economics Rules, 2015
Rodrik’s “one economics, many recipes” is the synthesizing move that tries to hold both poles at once — a single discipline whose unity is precisely its plurality of models. It is the right instinct. But it is not quite enough, because it collapses three different kinds of plurality into one. This stage separates them, and lands the verdict.
History of Economic Thought Ch.17 stages exactly this question at the post-2008 vantage: is the discipline converging on a new synthesis or fragmenting into specialties that no longer share a paradigm? Its honest answer is that both are happening at once, and the verdict this walkthrough reaches is one notch more precise. The discipline is not converging-or-fragmenting along a single axis. It is doing different things at three different layers: converging at the level of method, fragmenting at the level of substantive frame, and carrying an unabsorbed periphery alongside both. Collapsing those three into a single “converging vs. fragmenting” question is what makes the debate feel unresolvable. Separate the layers and it resolves.
The discipline has been here before, which is worth remembering before treating the current configuration as a crisis. In the mid-twentieth-century Cambridge capital controversy, the two Cambridges fought over whether the dominant capital-theoretic apparatus was a real unification or a parochial choice dressed up as necessity. The technical points were largely conceded by the side defending the apparatus — and mainstream practice changed almost not at all. That precedent is instructive: a discipline can absorb a deep internal challenge to its core apparatus, concede the logic, and carry on, because method-unification and substantive-fragmentation are not the same thing and do not rise and fall together. The marginalist formalization that set up that whole controversy is History of Economic Thought Ch.5 (Marginalist revolution and formalization).
Three readings, three scales, all correct
Hoover and Acemoglu saw the method-floor. What was scattered in 1970 is shared in 2025: a common standard for what counts as a credible empirical claim, and a common grammar of optimization under constraints. This is real, and it is correctly described as unification — but at the methodological level only. They are not wrong; they are reporting the layer they stand on, where the discipline genuinely is one.
Hausman and Thaler saw the substantive walls. Even a shared method does not produce shared questions, shared objects, or shared frames. Behavioral and macro share a method-floor and disagree about whether the agent at the center of the models is real. This too is real, and it is correctly described as fragmentation — but at the substantive level. They are reporting a different layer than Hoover, not contradicting him.
Lawson and Kelton saw the non-absorbing periphery. Some traditions reject the shared method outright, carry coherent apparatus of their own, and persist outside the federation by design. This is real too, and the federation’s correct posture toward it is engagement on its own terms, not absorption. They are reporting a third layer — the boundary of the federation, seen from just outside it.
The integration is not picking a winner. It is recognizing that all three are seeing real structure at different scales, and that the structure has three layers, not one. “Method-federated, substantively specialty-fragmented, heterodox-non-absorbing” is the single description that holds each reading exactly where it is correct — and it is more precise than Rodrik’s “one economics, many recipes,” which is true but folds the method layer, the frame layer, and the periphery into one undifferentiated “plurality.”
The verdict
Neither pole alone is the honest answer. Economics is not “a unified science,” and it is not merely “a collection of specialties.” It is method-federated: the subfields share enough method for genuine cross-subfield communication, disagree about objects and frames and questions enough to look like a federation of specialties, and sit alongside a heterodox periphery that uses different methods and stays peripheral by design. This is a split located at the frame and method level, not at the level of a disputed parameter — which is the rarer and more demanding kind of split to report, and the reason “both views have merit” would be a failure here rather than a verdict. The calibration is the position: method = unified; substantive frame = specialty-fragmented; periphery = non-absorbing.
Four things follow for a reader trying to use this:
- When someone asks “what does economics say,” the honest reply is “which subfield.” Different subfields make different substantive commitments even when they share a method. There is no single economic answer to most interesting questions, because there is no single economic frame.
- The method-floor is a genuine achievement. Empirical claims in economics are far more comparable across subfields than they were in 1970. Cross-subfield reading is possible in a way it once was not. Do not let the fragmentation story erase how real this is.
- The unabsorbed periphery is not a failure of the federation. That Austrian methodology, MMT, Post-Keynesian uncertainty, and modern Marxian class-analysis do not fold into the core is the federation working as it should. Engage these on their terms; do not translate-and-dismiss.
- The discipline is less unified than its method-floor suggests and more coherent than its specialty-walls suggest. “Federated, with a non-absorbing periphery” is the most honest description of economics in 2025.
Two caveats keep the verdict honest. The “method-unified” claim is historically recent — it reads cleanly only back to about 1990, and projecting it onto the longer, messier history of the discipline would flatter the present. And one question sits deliberately outside this walkthrough’s scope: whether economics is a natural science in the Popper–Lakatos–Kuhn sense. That is a philosophy-of-science question one level up from “what is the discipline’s self-understanding,” and it deserves its own treatment rather than a paragraph here.
Where this leaves us
We started with three voices describing the same discipline and apparently contradicting each other — a philosopher calling economics “inexact and separate,” methodologists calling it newly unified, heterodox economists saying they were never invited. The walkthrough’s claim is that none of them is wrong. They are standing on different layers of the same structure:
- Method-unification is real. Optimization-under-constraints and the credibility revolution’s evidentiary standards give the discipline a shared method-floor it did not have in 1970. (Hoover, Acemoglu — Stage 1.)
- Substantive fragmentation is also real. Shared method does not make shared objects, questions, or frames. From inside a subfield, economics is a federation of specialties. (Hausman, Thaler — Stage 2.)
- The heterodox periphery is real and unabsorbed. Austrian, Post-Keynesian, MMT, and modern Marxian traditions use different methods and stay peripheral by design. (Lawson, Kelton — Stage 3.)
- The synthesis is all three at once. Economics is method-federated, substantively specialty-fragmented, and heterodox-peripheral — the single description that holds each reading where it is correct. (Stage 4.)
What do you do with this view? First, distrust the phrase “economics says.” It almost always means “a particular subfield, working in a particular frame, says” — and the next subfield over may study objects so different that the two barely speak. Asking “which economics” is not pedantry; it is the first honest question. Second, hold the method-floor and the fragmentation in your head at the same time without forcing a choice. The discipline really did converge on a shared standard for evidence, and it really did not converge on a shared picture of the world — and once you see those as two different layers rather than two answers to one question, the apparent paradox dissolves.
If this view is right, the next questions almost ask themselves. Given a federated core with a non-absorbing periphery, is there a coherent left-economics inside it, or a coherent right-economics — explored in Is there a coherent left-economics? and Is there a coherent right-economics?. Did 2008 break the federated macro core in particular, or merely bolt a missing part onto it — the question of Did 2008 break macroeconomics?. And why did the federated mainstream take the shape it did at all — a question of historical contingency taken up in How did the Cold War shape economic thought?. The discipline is not one science. It is a federation that learned to share a method without agreeing on a world — and that turns out to be a stranger, more interesting thing to be.