Primer

What Is Metascience? The Study and Design of How Science Works

David Hilmer Rex · June 2026

Summary

Metascience is the study of how science works as a human system - and the deliberate design of better versions of it. It treats the machinery of research - funding, institutions, careers, publishing, incentives - as the object of inquiry, and asks how changing that machinery changes what can be discovered. The field carries two temperaments, one bent on making science more reliable and one on making it more productive, both resting on the premise that the social processes of science are contingent and therefore designable. This primer defines metascience, traces its lineage from Bush and Merton through the reproducibility crisis and progress studies, maps its strands, weighs the critiques of the movement, and explains how this resource is organised around it.

A working definition ↑

Michael Nielsen and Kanjun Qiu frame metascience as the search for “new social processes for science.”1

The field has a descriptive wing and a design wing. The descriptive wing - the quantitative “science of science,” or meta-research, institutionalised when John Ioannidis and Steven Goodman founded the Meta-Research Innovation Center at Stanford in 20142 and mapped at scale by the broader science-of-science research programme3 - studies research empirically: its methods, reporting, reproducibility, evaluation, and incentives. The design wing asks the normative question that follows: given how science works, what should we build or change? Metascience, as this resource uses the term, spans both - the study of how science works and the deliberate design of better versions of it.

Metascience as a systemic capability ↑

The definition above casts metascience as something researchers do to science: study its machinery, design better versions. There is a second register, increasingly useful, in which metascience names something the system itself has - the capacity of a research and innovation ecosystem to reform how it funds, organises, evaluates, and shares research, as deliberately as it revises its theories.16 On this reading the object is not only redesigned from outside; the system’s ability to redesign itself is the thing the word points to.

This capability is distributed. It can live at any node - a lab that changes how it credits contributions, a funder that pilots a new grant mechanism, a university that rethinks how it evaluates careers - and a system has strong metascience when the capacity is present across many nodes rather than in one place. Because the parts of a research system are coupled, rigidity in one part is felt by the rest: a funder free to experiment is limited by institutions that cannot, and a promising new organisation starves if the surrounding system cannot resource or absorb it.

Specialised institutions - agency metascience units, meta-research centres, science-of-science labs, science-of-science funders - are organs of this capacity, not its owners. They concentrate the work that benefits from depth, scale, or coordination: rigorous meta-research, large controlled trials of funding mechanisms, shared data and infrastructure. At their best they raise the metascientific capacity of every other node rather than hoarding it. The failure mode is inversion - when a single office becomes the only site of the capability, the system gains a point of specialisation but also a single point of fragility and capture.

This framing speaks directly to a tension noted below: that the movement is concentrated among a small, largely US and philanthropically funded group. Read as a field, that concentration is a limitation to be managed; read as a capability, it is the problem to be solved - the aim is a capacity distributed across the ecosystem, not a discipline owned by a few. The capability register is a normative and design claim, complementary to the study-and-design definition rather than a replacement: it names what the field’s design wing is ultimately trying to cultivate.

Understood this way, metascience is a whole-of-system ambition: that a research and innovation system, at every scale, retains the capacity to change and adapt. It is a regulative ideal - never fully reached, and not meant to be, since the arrangements worth having keep shifting as tools, questions, and conditions change. The aim is not that everything is always in flux, but that nothing is structurally unable to change when it should; a healthy system holds both the capacity to reform itself and the judgment of when to conserve. That capacity - distributed, and never finished - is what the word finally names.

A short lineage ↑

Metascience is older than its current name. Vannevar Bush’s Science, The Endless Frontier (1945) was a work of institutional design: it argued for sustained federal funding of basic research and led to the creation of the US National Science Foundation - building much of the very system metascience now interrogates.4 Robert Merton then treated science as a social institution with its own norms - communalism, universalism, disinterestedness, organised scepticism - making the culture of research a legitimate object of study.5 And Derek de Solla Price, in Little Science, Big Science (1963), helped found the quantitative study of science later called scientometrics.6

The modern revival has two triggers, two decades apart. In 2005 John Ioannidis published “Why Most Published Research Findings Are False,” which crystallised a methodological reckoning that became the reproducibility crisis.7 Then, around 2018, a separate conversation about scientific stagnation and institutional reform gathered force. These two impulses - reliability and ambition - are the field’s two temperaments.

Two temperaments: rigour and progress ↑

The first temperament wants to make science more reliable. It grew out of the reproducibility crisis and runs through open science, preregistration, and registered reports. Its institutional home is Brian Nosek’s Center for Open Science (founded 2013) and projects like the Reproducibility Project: Psychology, which found that only about a third of one hundred published findings reached significance when independently repeated.8 Its question is: how do we stop believing things that are not true?

Why do reliability problems persist despite decades of awareness? The clearest structural account comes from Smaldino and McElreath: poor methods are not primarily a product of ignorance or dishonesty but of selection pressure. Labs that prioritise publication output over methodological rigour produce more papers, attract more students, and propagate their practices through a Darwinian dynamic that selects for yield rather than accuracy. Reviews of statistical power in the social sciences found a mean of around 24%, and that figure did not improve over six decades of calls for reform.15 The implication is that the reliability agenda cannot be satisfied by education or better guidelines alone; it requires changing the incentive landscape that drives the selection.

The second temperament wants to make science more productive and ambitious. It grew out of the stagnation debate and runs through progress studies and the new-institutions movement - ARPA-model agencies, Focused Research Organisations, new funding mechanisms. Its question is: how do we discover more, faster, and attempt the things the current system cannot?

These can pull against each other - rigour can counsel caution and slower publication; ambition can counsel more risk and more failure. The most interesting metascience sits where they meet: how to make science simultaneously more trustworthy and more daring. This resource leans toward the institutions-and-progress side, while treating the reliability critique as a permanent constraint rather than an afterthought.

The stagnation question ↑

The empirical backbone of the progress temperament is the finding that research is getting less efficient. Bloom, Jones, Van Reenen and Webb, in “Are Ideas Getting Harder to Find?” (2020), show that research effort has risen sharply while research productivity has fallen across sectors: sustaining the historical pace of Moore’s Law now takes more than eighteen times the researchers it required in the early 1970s, with parallel declines in agriculture and pharmaceuticals.9

A second, independent line of evidence is the apparent decline in disruptiveness. Analysing 45 million papers and 3.9 million patents, Park, Leahey and Funk reported in 2023 that research has grown steadily less likely to break with the past and push a field in a new direction - the average disruption score fell by more than 90% between 1945 and 2010.10 That finding is itself contested - later analyses argue the measured decline is partly an artefact of citation inflation rather than a real loss of originality - which is instructive in its own right: even the core empirical claims of the stagnation thesis are live rather than settled.

Patrick Collison and Tyler Cowen turned this body of work into a public agenda with two essays - “Science Is Getting Less Bang for Its Buck” (2018) and “We Need a New Science of Progress” (2019) - that named progress studies and gave the institutional-reform conversation its broadest intellectual home.11 Whether science is genuinely slowing remains disputed; but the perception is itself one of the forces driving the field.

What metascience covers ↑

The field spans several layers: how research is funded and incentivised; how institutions, careers, and talent pipelines are structured; how results are published, validated, and communicated; how decentralized and AI-driven models are changing what is possible; and how the broader political and economic environment shapes what research gets done.

There is no single canonical taxonomy. The most widely used structural map, from Ioannidis and colleagues, organises meta-research around five domains: methods (how research is conducted), reporting (how findings are communicated), reproducibility (whether findings replicate), evaluation (how research is assessed), and incentives (what behaviours are rewarded). Peterson and Panofsky, reading the movement sociologically, identify three converging streams: the quantitative science of science, the open-science reform agenda, and the reproducibility movement. Watney and Gustetic add further resolution, mapping six communities that emphasise metrics, funding, institutions, culture, policy, and progress.12 The common thread: “metascience” names a coalition of overlapping projects, not a single programme.

Critiques and limits ↑

The movement has its critics, and taking them seriously is part of doing metascience well. A first objection is that science is not one thing: reforms that suit experimental psychology may not fit mathematics, field ecology, or history, and a search for universal “best practices” can flatten a productive disunity. A second is that the movement has leaned heavily on quantitative metrics and on the reproducibility-crisis reform agenda - preregistration, replication - while underusing the qualitative scholarship in science and technology studies that has examined how science works for half a century.13

The sharpest sociological account, from David Peterson and Aaron Panofsky, frames metascience as a “scientific social movement” - a coalition of methodologists, data scientists, and open-science activists - and presses two further points. First, the reproducibility “crisis” that licenses the movement is itself partly an actors’ construction, and some of its foundational evidence (an influential 2016 Nature survey, the Amgen cancer-replication study) has been criticised as methodologically thin. Second, by claiming authority over “science” as a whole, metascience presumes a unity the disciplines may not have - and, dressed in a lab coat and speaking in data, mounts a more direct challenge to disciplinary autonomy than science studies ever did.14

Three further worries recur. That prescriptions like preregistration can harden into dogma and themselves narrow what research is possible - a diversity cost of exactly the kind the institutional-diversity primer warns against. That the progress-studies framing justifies science largely instrumentally, as the fuel of material progress, when that is neither its only value nor an uncontested one. And that the movement is concentrated among a small, well-connected, largely US and philanthropically funded group - which raises the same question the new-institutions conversation faces: whose priorities set the agenda?

Finally there is the reflexive test. A field that studies how to improve science is obliged to hold its own prescriptions to the evidence it demands of others. That test is increasingly being run - randomised studies of funding mechanisms, empirical work on the reliability of peer review, the large reproducibility projects - but the honest status is that metascience is young: many of its proposed reforms are plausible and few are yet proven at scale. Its credibility will rest on meeting its own standard.

How this resource is organised ↑

This primer is the front door. Companion primers cover funding mechanisms, institutional diversity, enterprise foundations, focused research organizations, and the ecosystem forces map - held together by one question, in Nielsen and Qiu’s phrasing: what new social processes would let science do more? Treating the machinery of research as a designable thing, rather than a fixed inheritance, is how the next generation of discovery becomes possible at all.

Notes ↑

  1. Michael Nielsen & Kanjun Qiu, “A Vision of Metascience” (Science++, 2022). The “new social processes for science” formulation recurs throughout and frames much of this resource. ↩
  2. The Meta-Research Innovation Center at Stanford (METRICS) was founded in 2014 by John Ioannidis and Steven Goodman, with initial funding from the Laura and John Arnold Foundation. The agenda is set out in Ioannidis, Fanelli, Dunne & Goodman, “Meta-research: Evaluation and Improvement of Research Methods and Practices,” PLoS Biology 13(10): e1002264 (2015), which frames meta-research around five domains: methods, reporting, reproducibility, evaluation, and incentives. ↩
  3. Santo Fortunato et al., “Science of science,” Science 359, eaao0185 (2018); the field is synthesised in Dashun Wang & Albert-László Barabási, The Science of Science (Cambridge University Press, 2021). ↩
  4. Vannevar Bush, Science, The Endless Frontier (United States Government Printing Office, 1945). The report made the case for sustained federal funding of basic research and led to the creation of the National Science Foundation in 1950. ↩
  5. Robert K. Merton, “The Normative Structure of Science” (1942), reprinted in The Sociology of Science (University of Chicago Press, 1973). The norms are often summarised by the acronym CUDOS. ↩
  6. Derek J. de Solla Price, Little Science, Big Science (Columbia University Press, 1963), a founding text in the quantitative study of science. ↩
  7. John P. A. Ioannidis, “Why Most Published Research Findings Are False,” PLoS Medicine 2(8): e124 (2005). ↩
  8. Open Science Collaboration, “Estimating the reproducibility of psychological science,” Science 349 (2015): of 97 original studies reporting a significant result, about 36% of replications reached significance. The coordinating body, the Center for Open Science, was founded by Brian Nosek in 2013. ↩
  9. Nicholas Bloom, Charles I. Jones, John Van Reenen & Michael Webb, “Are Ideas Getting Harder to Find?” American Economic Review 110(4), 2020, 1104–44. Sustaining the historical rate of Moore’s Law is estimated to require more than 18 times the researchers needed in the early 1970s. ↩
  10. Michael Park, Erin Leahey & Russell J. Funk, “Papers and patents are becoming less disruptive over time,” Nature 613 (2023): 138–144, using a citation-based “CD index” over 45 million papers and 3.9 million patents. The result is disputed: subsequent analyses argue the measured decline is biased by citation inflation rather than reflecting a real loss of originality. ↩
  11. Patrick Collison & Tyler Cowen, “Science Is Getting Less Bang for Its Buck” (The Atlantic, 2018) and “We Need a New Science of Progress” (The Atlantic, 2019). ↩
  12. Caleb Watney & Jenn Gustetic, “The Six Camps of Metascience” (Macroscience, 2026). ↩
  13. On the disunity-of-science and STS objections, and the risk of reform-as-dogma, see the Center for Open Science symposium “Critical Perspectives on the Metascience Reform Movement” and the emerging strand of “critical metascience.” That a hub of the reform movement convenes its own critics is a sign of the field’s health, not its weakness. ↩
  14. Paul E. Smaldino & Richard McElreath, “The natural selection of bad science,” Royal Society Open Science 3: 160384 (2016). The 24% mean statistical power figure is drawn from their synthesis of reviews spanning six decades of social-science research. A minor coding error was identified in 2023 but did not affect the paper’s main conclusions. ↩
  15. David Peterson & Aaron Panofsky, “Metascience as a Scientific Social Movement”, Minerva 61 (2023). They read metascience as a transdisciplinary movement whose authority depends on treating science as a unified object - against the science-studies view that science is “not one, indivisible, and unified, but… many, diverse, and disunified” (Shapin) - and argue that its atheoretical, quantitative confidence is what gives it “the courage to intervene on science as a whole.” ↩
  16. The “capability” framing extends Nielsen and Qiu’s image of “a system that improves its own ability to discover.” Its organisational-learning analogue is Chris Argyris and Donald Schön’s double-loop learning - a system revising its own governing rules, not only its performance within them (Organizational Learning, 1978). Compare W. Ross Ashby’s law of requisite variety: a system needs sufficient internal variety to regulate the variety of its environment. ↩

Further reading ↑