.
We Need to Start Talking About Statistics:
Social Scientism, Individualism, and Paradigmatic Economics and Statistics
by Paulo dos Santos
.
In The Structure of Scientific Revolutions, Thomas Kuhn (1962) framed scientific development as a punctuated process shaped by the evolution of paradigms organising inquiry in specific disciplines. Paradigms can be understood as constituted by epistemic systems and institutional structures. Epistemic systems organise reasoning within a domain, including a discipline’s recognised canonical problems, its accepted methods of investigation and standards of explanation, background abstractions and metaphysics, and a shared sense among its researchers of what constitutes a good theory or solution. Institutional structures organise the social and material conditions for inquiry, including the instruments and modalities for training practitioners and researchers, and forms of engagement with the wider public about the discipline, its findings, and its material needs.
Kuhn offered a simple but evocative schematisation of how the historical and social evolution of paradigms shapes scientific progress. An emerging paradigm offers a discipline the ability to organise and pursue inquiry successfully over a certain period. Invariably its usefulness starts to fray as new observations and results appear to challenge its premises. First cast as empirical errors or anomalies, their persistence and accumulation soon beget explicit explanation. This typically ushers in an epicyclical moment as new facts are interpreted as special cases, explainable on the basis of increasingly elaborate and inelegant grafts onto paradigmatic conceptual structures [1]. The proliferation of epicyclical variants of a paradigm offering contradicting ways to accommodate “anomalies” within its terms is soon followed by explicitly foundational debates, interrogating a paradigm’s conceptual foundations, sometimes even identifying them clearly for the first time. This is a moment of crisis.
Revolution is not the inevitable consequence of a paradigmatic crisis. With institutions under the control of older, senior researchers committed to what is increasingly asserted as orthodoxy, alternative conceptual frameworks and their proponents are systematically marginalised. Control over authoritative journals, research appointments, the writing of canonical texts, and even the ability to define what is considered to be within the bounds of “the discipline” are all marshalled in defence of the status quo. In his celebrated discussion of the history of Statistical Mechanics and the substantive bases for its remarkable scientific success, mathematical physicist Edwin T Jaynes (1967) called those maximum-compression periods: Moments in the history of a discipline where the full range of institutional and career incentives are deployed to inhibit the development of new paradigms that may address the incongruences fuelling the crisis. A scientific revolution takes place when a non-paradigmatic conceptual framework proves capable of addressing some of those incongruences, shows the promise to address the remaining ones, and opens new lines of inquiry in spaces that may not even have been discernible under old orthodoxies. This typically involves work by younger or marginalised researchers who had not been fully socialised in the old orthodoxy and are willing to persist in the face of its strong, conservative headwinds.
Periods of maximum compression are typically short lived at the time scales of scientific progress. The epicycles prove too unwieldy, as the inventory of “exceptional cases” grows too big and conceptually disjointed; the explanatory failures, retractions, and even scientific scandals become too big to ignore; disciplines become parochial, focussed more on internal problems of their fraying conceptual frameworks rather than with crucial realities of their domains of inquiry. Fundamental change eventually happens, even if only generationally as “its opponents gradually die out,” as grimly put by Max Planck (1950, p 33).
Unless the discipline is Economics… or Statistics.
.
A Familiar Case of Paradigm Maintenance
Heterodox economists may not yet know the term maximum compression. But they certainly know what it is like to work in a discipline experiencing one across several generations.
Economics has a remarkable tolerance for zombie ideas, lingering in professional and public debates despite decades of accumulated empirical evidence and logical arguments refuting them; like the contention that money is valuable because and in the measure to which it is “scarce.” Much of the profession continues to be preoccupied by insular, epicyclical concerns: Diamond and Dybvig won the Sveriges Riksbank Prize in Economic Sciences in 2022 for work showing that bank runs observed as far back as the collapse of Bardi and Peruzzi in the 1340s and grasped in their essential logic and mechanics centuries ago are, in fact, the expression of equilibrium strategies in a mid-twentieth-century game-theoretic model of interacting utility-maximising rational agents.
The discipline has also shown a revealing resistance to error correction: When then-PhD student Thomas Herndon found tendentious methodological choices and coding errors in highly influential work by leading orthodox scholars claiming to have found statistical evidence that large fiscal deficits reduce economic growth, neither the scholars nor their underlying substantive claim saw their standing or credence diminish among economists. At a deeper conceptual level, conventional economic thinking continues to be substantively anchored on an individualist abstraction whose profound limitations have long been partly understood by unusually critical voices among leading figures like Arrow (1994), and more fully taken apart by critical voices outside of paradigmatic spaces [2].
While keeping ideas and scholars offering fundamental critiques and alternatives outside its recognised boundaries, the discipline has also evolved paradigmatic spaces where researchers can engage with empirical economic realities without challenging the paradigm’s conceptual foundations. In the years following Leamer’s (1983) call for a “credibility revolution” in econometrics, empirical work seeking statistical evidence of causal relations in natural or quasi-experimental re-search designs grew to the point where it has come to dominate research in orthodox Economics.
This “empirical turn” was not a response to a corresponding shift in the pressing questions facing economic analysis; it followed its own paradigmatic imperatives. By their very logical requirements and construction, studies based on natural and quasi-experimental identification yield findings internal to specific, localised settings and processes [3]. Their results carry no necessary implication for core conceptual tenets in the orthodox paradigm, enabling their epicyclical interpretation within those terms—much like the celebrated findings of Card and Krueger (1994) were interpreted in terms of “monopsony” in conventional approaches to the competitive behaviour of firms, not as pointing to the need to move beyond those approaches. As the absolute and relative measure of this work grew, Economics became steadily smaller and narrower in its scope of concern. All while climate change, the geographical reconfiguration of global production, and rising measures of income and wealth inequality posed difficult, potentially existential questions, demanding precisely the big, ambitious, and non-paradigmatic approaches to economic analysis the discipline has marginalised.
.
Compressed Statistics
Statistics has experienced its own version of “maximum compression,” defined over its own concerns and history, but showing crucial sociological and substantive parallels with the experience in Economics.
Between the 1920s and 1930s orthodox statistics had become clearly organised around a “frequentist” paradigm, founded on the view that the probability of an event or proposition is a property of the objects involved, given by the relative frequency with which the event or proposition occurs over large numbers of instances or “trials”. Probabilistic reasoning and statistical analysis became identified with the idea of grappling with “random” or “stochastic” processes in objects of inquiry. Heterodoxy has centred on the alternative view that probabilities are properties of an observer’s knowledge state about objects of inquiry, given by logically consistent measures of the plausibility that observer assigns to an event or proposition, given what they know. There is no such thing as “randomness” or “stochastic processes” under this view, at least not beyond quantum-mechanical scales. There is only incomplete knowledge about the objects confronted by different disciplines, which researchers work to reduce by considering new evidence and making inferences to update their beliefs. As Jaynes argued by drawing on the work of Pierre-Simon Laplace, Willard Gibbs, Richard T Cox, Harold Jeffreys, and John M Keynes, Probability Theory is the generalisation of logical reasoning to propositions of unknown veracity.
Differences between this latter epistemic characterisation and the conventional, frequentist interpretation of probability may seem subtle and abstract. But they ultimately resulted in starkly different approaches to the nature, explanatory goals, and instruments of statistical inference, with frequentist stances dominating the discipline’s substantive commitments and institutional spaces until the present.
The frequentist paradigm has been able to sustain itself despite mounting evidence of the analytical advantages of models and inference procedures anchored on epistemic principles, and the pitfalls in aspects of its approach to statistical inference. The wider availability of computing power since the early 1990s made implementation of computationally intensive Bayesian instruments practically feasible. Many of them have since become used widely in statistical inference, but overwhelmingly only as advantageous practical tools deployed with no reference or grounding in the epistemic interpretation of probability they presuppose.
More recently, evidence that large proportions of statistical results reported in medical, psychology, and social-science journals could not be replicated posed very serious questions going to the very heart of the frequentist paradigm’s approach to statistical inference [4]. In 2016 the Ameri-can Statistical Association issued a statement on the use of p-values in statistical significance tests, clarifying their logical content within the paradigm and identifying errors and abuses in many applications using them [5].The statement did not take on the deeper conceptual problems Bayesians have long recognised in frequentist hypothesis testing and how they contributed to the “reproducibility crises” raging across many disciplines. The general pedagogy, working culture, and the explicit or implicit background abstractions and metaphysical commitments anchoring most statistical work have remained frequentist.
.
Social Foundations and Content of the Frequentist Paradigm
The staying power of frequentist treatments of probability and statistical analysis has historical and social foundations that heterodox economists should readily recognise. Modern statistics was developed by scholars who were leading eugenicist theorists and advocates in late-nineteenth and early-twentieth-century Britain. Men like Francis Galton, Karl Pearson, and Ronald Fisher laid the discipline’s conceptual foundations as part of their efforts to establish “scientifically” that inherent and inherited individual human abilities are decisive determinants of social outcomes, and that society ought to encourage selective “breeding” favouring socially privileged, white, Anglo-Saxon men like themselves. Eugenicist objectives, presumptions, and metaphysical commitments left clear, substantive marks on statistical theory and practice.
Aubrey Clayton (2021) argues in this context that it was not the analytical merits but the discursive power of frequentist statistical analysis that led to its rise to dominance. The founders of modern statistics were looking for bases to make authoritative statements of “scientific” eugenicist “fact” to the general public. An epistemic characterisation of probability under which statistical analysis yields conclusions about an observer’s beliefs conditional on their explicit assumptions and knowledge offers comparatively poor bases for this ideological project. Frequentist abstractions fit the task much better, delivering seemingly authoritative, mathematically formal statements about aspects of objective “random processes” based on “impartial” inferences from “the data.”
Recent contributions have located a further, distinct eugenicist legacy at the heart of modern statistics; one concerning not the interpretation of probability but the explanatory individualism embedded in its basic abstractions (dos Santos, 2020; 2024). Fisher’s formative statistical work addressed problems of heredity in plants under controlled laboratory conditions. In that setting each individual can reasonably be treated as an instance of a single homogeneous reality. Accordingly, explanation can be taken to mean the identification of a general statistical relationships between individual outcomes and individual characteristics, holding uniformly and independently across all individuals. Transplanting this framework to more general analytical settings presumes that individual outcomes can be explained without reference to interaction, relational structures, or systemic effects between individuals. Doing so to social and economic settings mirrors the funda-mental, defining fallacy in eugenicist thought: The attribution of social differentiation exclusively to individual characteristics and behaviours presumed to be genetically given to individuals, with no reference to social hierarchies and interrelations, or to historical facts.
While the genetic contentions and explicitly racialist frames have been duly scrubbed from the framework since Fisher’s time, the underlying individualist reductionism has remained deeply ensconced in contemporary statistical analysis.
.
Affinities and an Anti-Paradigmatic Agenda
The persistence of maximum-compression states in both Economics and Statistics is conspicuous and reflects the same social realities. Both fields evolved and gained prominence in modern poli-ties as disciplines capable of presenting particular sociopolitical visions and discourses disguised as authoritative “science,” communicated to the public through mathematically formal statements of ostensibly “neutral” or “positive” fact. Their development became bound up with the rhetorical demands of this form of scientism. Both disciplines also came to reflect and reinforce conceptually pervasive social prejudices and analytical biases, most notably in a common commitment to individualist abstractions and explanatory frameworks where individual outcomes are interpreted in relation to individual characteristics and agencies, with no reference to relational structures, macroscopic interactions, or irreducibly systemic relations.
These affinities suggest a few implications for heterodox thought within both disciplinary domains. For heterodox thinkers in statistics, the analytical problems of socioeconomic systems may offer a uniquely rich terrain on which to illustrate the conceptual limitations of frequentist abstractions and the greater generality of Bayesian or epistemic approaches. These systems are irreducibly complex. Their individual members exhibit what philosophers call relational interiority: individual workers, enterprises, and national economies are inherently defined by their relations with other parts of the economic system. Their nature is constituted by their position and interactions within the system’s relational structures—by their class position, their integration into input-output systems and financial relations, and by their specific roles in the global division of labour, respectively. Many of the regularities enabling the development of successful theories of those systems are thus systemic or social in character. They are properties of the relevant relational structure as a whole rather than of any individual considered separately, straining the explanatory power of the naive abstraction of homogeneous, independent individual “trials” sustaining the frequentist paradigm. Further complicating the situation, analysis of these systems proceeds under conditions of chronically incomplete knowledge, with severe restrictions on what can be observed at all, and on the frequencies at which it can be observed. Latent relational structures shaping what may be observed are pervasive, ensuring individualist models defined over observables often misattribute to individual characteristics relationships that in fact involve systemic or social interactions. Analysis also demands careful specification of what is actually observed and what is unobserved, what is known and what is conjectured or assumed for convenience. The defining tenets of epistemic treatments of statistical analysis are more than mere philosophical nuances or preferences. They are often indispensable for effective and consequent analysis.
For heterodox economics, the nature, goals, and foundational abstractions of statistical analysis cannot be seen as extraneous or exoteric distractions. They are in fact a core concern of our shared intellectual project. Economic systems lend themselves to statistical analysis as aggregations of large numbers of constitutive parts, and conventional statistical inference can often be useful as we grapple with aspects of their functioning. But the conceptual foundations for most approaches to statistical inference embed strong and socially meaningful biases at the very heart of quantitative social inquiry [6].
Empirically oriented heterodox researchers thus need to be in a position to evaluate the potential bases and bounds of application of established statistical methods. Otherwise they risk sneaking, through the back door of statistical technique, substantive commitments and discursive thrusts most of them emphatically reject—like the presumption that the goal of social inquiry is to obtain statements of “fact,” or to arrive at individualist, atomistic “explanations” for social outcomes. The critique of statistical analysis and the development of statistical approaches capable of handling irreducibly social regularities and pervasive latent relational structures are crucial pedagogical and research tasks for contemporary heterodox economics.
.
Footnotes
[1] The term epicycle originated in astronomy, where the celestial movement of observable planets following elliptical orbits around the sun posed serious difficulties to ancient geocentric cosmologies as well as early heliocentric representations of the solar system.
[2] See Mohun 1994, Lawson 2012, or dos Santos 2024, for instance.
[3] See Reddy 2012, for instance.
[4] Gigerenzer 2004, Ioannidis 2005.
[5] Wasserstein and Lazar, 2016.
[6] dos Santos and Wiener, 2020.
.
References
Arrow, K. (1994). Methodological Individualism and Social Knowledge. American Economic Re-view, 84(2), 1–9.
Card, D., & Krueger, A. (1994). Minimum Wages and Employment: A Case Study of the Fast-Food Industry in New Jersey and Pennsylvania. American Economic Review, 84(4), 772–793. Clayton, A. (2021). Bernoulli’s Fallacy: Statistical Illogic and the Crisis of Modern Science. Columbia University Press.
dos Santos, P. and N. Wiener (2020). By the Content of Their Character? Discrimination, social identity, and observed distributions of income. The Journal of Mathematical Sociology, 44 (1), 12–41.
dos Santos, P. (2020). Statistical Equilibria in Economic Systems: Socio-Combinatorial or Individualist-Reductionist Characterizations? European Physical Journal — Special Topics, 229(9), 1603–1622.
dos Santos, P. (2024). Socio-Historical Ontology, Explanations, and Empirical Approaches. In M Ali and A Davis (eds), Radical Political Economics: Principles, Perspectives, and Post-Capitalist Futures. London: Routledge.
Gigerenzer, G. (2004). Mindless Statistics. Journal of Socio-Economics, 33(5), 587–606.
Ioannidis, J. P. A. (2005). Why Most Published Research Findings Are False. PLOS Medicine, 2(8), e124.
Jaynes, E. (1967). Delaware Lecture. In E. T. Jaynes: Papers on Probability, Statistics and Statistical Physics. New York City: Springer, 1989. 87–113.
Kuhn, T. S. (1962). The Structure of Scientific Revolutions. University of Chicago Press. Lawson, T. (2012). Ontology and the study of social reality: emergence, organisation, community, power, social relations, corporations, artefacts and money. Cambridge Journal of Economics, 36(2), 345–385.
Leamer, E. (1983). Let’s Take the Con Out of Econometrics. American Economic Review, 73(1), 31–43.
Mohun, S. (1994). Value, Value-Form and Money: Debates in Value Theory. New York City: St. Martin’s Press.
Planck, M. (1950). Scientific Autobiography and Other Papers (F. Gaynor, Trans.). Williams & Norgate.
Reddy, S. (2012). ”Randomise This! On Poor Economics” Review of Agrarian Studies, 2(2), 60–73.
Wasserstein, R. L., & Lazar, N. A. (2016). The ASA Statement on p-Values: Context, Process, and Purpose. The American Statistician, 70(2), 129–133.
.
.
.
.
.