1 Can human behaviour be studied scientifically?

The human sciences — psychology, economics, sociology, anthropology, political science — aspire to apply the methods of science to human behaviour. The ambition is justified: if we can know the world through systematic observation and experiment, why shouldn’t this apply to the most important thing in it, which is us? But the application raises a fundamental problem that distinguishes the human sciences from the natural sciences in ways that go beyond technique.

1.1 The Object/Subject Problem

In physics, the objects of study — particles, waves, fields — do not know they are being studied. They cannot change their behaviour in response to the physicist’s attention, cannot understand the theory being used to describe them, and cannot adjust their properties in light of the conclusions drawn. The physicist studies a world that is genuinely independent.

In the human sciences, the object of study is human beings. Human beings are conscious, self-interpreting, theory-reading agents. They know they are being studied. They can read the published findings of sociology and adjust their behaviour accordingly. They can understand what experimenters expect and either conform to or subvert those expectations. The sociological law that “rising incomes reduce crime” can be rendered false if people read it, believe it, and use it as an excuse not to invest in crime prevention. There are no such feedback loops in physics.

The feedback loop is not a technical problem to be solved by better experimental control; it is a structural feature of the relationship between the human sciences and their subject matter.

1.2 Weber’s Verstehen

Max Weber, in The Methodology of the Social Sciences (essays published 1904–1917), argued that social science requires a distinct kind of understanding that is irreducible to the causal explanation used in natural science. He called it Verstehen — understanding.⁴

To explain why a Catholic population has lower rates of economic productivity than a Protestant one (Weber’s question in The Protestant Ethic and the Spirit of Capitalism, 1905),⁵ it is not sufficient to identify statistical correlations. You need to understand the meaning that economic activity has for agents shaped by different religious frameworks — the sense in which the Protestant doctrine of calling made mundane labour a spiritual vocation, while Catholic frameworks provided different routes to salvation. Causal explanation without Verstehen produces correlations without understanding.

Weber’s claim is epistemological: understanding human action requires grasping the subjective meaning it has for the actor, and this kind of grasping is different from the third-person causal knowledge that natural science produces.

1.3 Reasons and Causes

Weber’s Verstehen leaves an open question that the 20th century took half a century to answer. If understanding human action requires grasping the agent’s reasons — beliefs, desires, intentions — and explaining a natural phenomenon requires identifying its causes, are reasons and causes two different categories of explanation, or two descriptions of the same thing?

Donald Davidson’s “Actions, Reasons, and Causes” (Journal of Philosophy, 1963) is the canonical answer. Davidson argued, against the then-dominant Wittgensteinian view that reasons-explanations and causal explanations belong to logically distinct categories, “the ancient — and commonsense — position that rationalization is a species of causal explanation.”¹¹ Giving an agent’s reason for an action — the belief and desire that together rationalise it — is not an alternative to identifying the cause of the action; the reason is the cause, under a particular description. The “primary reason” for an action, on Davidson’s account, “is its cause.”¹²

Take an everyday case. “She shouted because she was angry.” The clause names the reason — what made the shout intelligible to anyone who knows her — and the cause — what produced the shout. There are not two events here, the reason-event and the cause-event; there is one event under two descriptions.

The implication for social-scientific method is structural rather than rhetorical. Reasons and causes do not mark off two epistemically independent domains; an interpretive social science that grasps an agent’s meaning has not thereby exited the space of causal claims.

Rosenberg draws the consequence: if Davidson is right, the categorial gap between Verstehen and Erklären (understanding and causal explanation) that Weber and the German hermeneutic tradition treated as foundational collapses into a methodological difference — reasons require interpretive access, causes do not — without a difference in the logical type of the explanation produced.¹³ Whether Davidson’s argument actually does the work Rosenberg asks of it is itself contested in the philosophy of action (Anscombe, Sellars, the more recent neo-Wittgensteinian programme of Hutto and others have all denied it).

This does not settle the Verstehen debate; it relocates it. Even if reasons are causes, the question remains whether the social sciences can identify them with the rigour the natural sciences identify their causes — and whether, as Winch insists, grasping the concepts that constitute a rule-governed practice is a precondition of identifying any reasons at all.

1.4 Ethnography in Action: Hochschild’s Flight Attendants

The Verstehen / Erklären dispute can be made concrete by considering what a properly Verstehen-shaped piece of social science actually looks like. Arlie Russell Hochschild, The Managed Heart: Commercialization of Human Feeling (University of California Press, 1983), is the standard exemplar. Hochschild trained with Delta Air Lines flight attendants — observing the company’s training programmes, riding flights as passenger and observer, interviewing trainees and senior staff over three years — and produced a sustained account of what she called emotional labour: the active management of one’s own felt and displayed emotions as a condition of paid employment. Her central concepts (surface acting, in which the displayed emotion does not match the felt one; deep acting, in which the worker actually persuades herself to feel what the role requires; the transmutation of private emotional capacities into a managed work product) are not produced by any econometric instrument; they are produced by sustained close observation of what trained people do and what they say about it. Forty years on, emotional labour is a load-bearing concept across labour economics, the sociology of work, organisation studies, and the contemporary literature on care work and burnout — and the path by which it entered all those fields was not statistical but ethnographic. The point for present purposes is not that ethnography replaces econometrics but that any account of “evidence in the human sciences” that treats the experiment and the regression as exhaustive misses where many of the field’s most-cited concepts came from. The same point can be made using Erving Goffman’s frontstage / backstage distinction (The Presentation of Self in Everyday Life, 1959), or Elinor Ostrom’s fieldwork on Maine lobster fisheries (Governing the Commons, 1990) — both produced through observational and interview methods that no quantitative reformulation captures without remainder.

1.5 The Hawthorne Effect

The Hawthorne Works was a Western Electric factory in Cicero, Illinois, where, between 1924 and 1932, a series of experiments were conducted to test the effects of working conditions (lighting levels, break schedules, pay structures) on worker productivity.¹⁴ The findings were puzzling: almost every change in conditions — better lighting, worse lighting; longer breaks, shorter breaks — produced improved productivity. The researchers concluded that the workers were responding not to the specific changes but to the fact of being observed and studied.

The Hawthorne Effect has become the name for this general phenomenon: humans change their behaviour when they know they are being watched. The epistemological implication is severe: any study of human behaviour that involves observation may be measuring something other than natural human behaviour — it may be measuring how humans behave when they know they are being measured.

Later researchers challenged the Hawthorne findings sharply: Levitt and List (2011), re-examining the original Western Electric records, were essentially unable to detect the classical Hawthorne effect at all in the lighting study — the productivity gains they did find could be explained by ordinary trends and incentives, not by observation per se.¹⁵ The “Hawthorne effect” survives as a pedagogical label more securely than as a clean empirical finding traceable to the original studies. But the general principle it stands for — that observation is rarely neutral in the human sciences, that the act of measuring can change what is being measured — remains methodologically important; it is just that the textbook anchor is shakier than the principle requires.

1.6 The Performativity of Social Theory

There is a deeper version of the observer problem that goes beyond the Hawthorne effect. When social scientists publish their findings about human behaviour, those findings enter the world — and change the behaviour they describe. This is the performativity of social theory: the human sciences do not merely describe a pre-existing social reality; they actively constitute it.

The clearest case is economics. When Black and Scholes published their option-pricing model (1973), traders began using it. The model became self-fulfilling: markets behaved according to the formula not because the formula captured a pre-existing truth about markets, but because market participants started using it as the basis for decisions. Donald MacKenzie (An Engine, Not a Camera, 2006) calls this Barnesian performativity: a theory that is used by enough actors in a system tends to make the system conform to the theory, because the actors reshape their behaviour to match the model’s assumptions.¹⁶

A weaker analogue may appear in psychology. After Kahneman and Tversky’s findings about cognitive biases became widely known, the bias literature was absorbed by behavioural economics, by trader-training, and by debiasing programmes — the Nudge programme (below) assumes that this absorption is incomplete and that biases survive awareness. The honest performativity claim here is narrower than for Black–Scholes: salient, high-stakes presentations can dampen specific biases for trained populations,¹⁷ but the claim that Thinking, Fast and Slow “changed the population it described” overstates the debiasing literature. Performativity in economics — where market participants act on the model and reshape prices — is structural; performativity in cognitive psychology is, at most, an unevenly successful intervention.

This is not the Hawthorne effect. The Hawthorne effect shows that the act of observation changes behaviour during the study. Performativity shows that the published theory changes behaviour permanently, restructuring what it was supposed to describe. In the natural sciences, publishing a paper about black holes does not change how black holes behave. In the human sciences, publishing a paper about how humans make decisions changes how humans make decisions. The feedback loop between theory and phenomenon is structural, not incidental.

1.7 Questions to Argue About

• If humans can change their behaviour in response to theories about them, does this mean the human sciences are fundamentally different from the natural sciences — or just more complicated?
• Weber’s Verstehen implies that explaining human action requires grasping its subjective meaning. Can this be done rigorously? Or does it introduce unacceptable levels of interpretive latitude?
• The Hawthorne Effect suggests that any observational study of human behaviour is contaminated by the observation itself. Is there any way out of this problem? Or must the human sciences accept it as a permanent limitation?
• If social-scientific theories are performative — if publishing them changes the phenomena they describe — does this make the human sciences less objective, or just differently objective from the natural sciences?
• Davidson says reasons just are causes (under a particular description). Weber, Winch, and the Verstehen tradition treat reason-explanations and causal explanations as categorially distinct. If you side with Davidson, what is left of Verstehen as a distinctive method? If you side with the Verstehen tradition, what do you say to the charge that you are simply giving causes a different name?

2 Are humans rational?

Mainstream economics, much political and moral philosophy, and most of the law are built on the assumption that humans are rational. Decades of empirical research have shown the assumption to be systematically wrong as a description of how people actually make decisions.

2.1 Homo Economicus

Adam Smith’s analysis in The Wealth of Nations (1776) rests on the assumption that individuals act in their own self-interest in ways that, through the mechanism of the market, produce collective benefit — the “invisible hand.”¹⁹ This is not quite the same as assuming people are fully rational, but it became the basis for the model of Homo economicus developed in 19th and 20th century economic theory: an agent who knows their preferences, has consistent beliefs, and makes choices to maximise expected utility.

The model has real theoretical virtues: it is tractable, it generates precise predictions, and it captures something genuine about human motivation in competitive market contexts. It also happens to describe a kind of being that does not exist — not because human beings are irrational in random ways, but because they are irrational in consistent, predictable, and thoroughly documented ways. This distinction matters: if irrationality were random, it would cancel out. The disturbing finding is that it does not.

Two further caveats before Homo economicus gets used as a punching bag. First, working economists rarely treat the rational-agent model as a literal description of psychology — Milton Friedman’s 1953 “as if” methodology argued explicitly that the question is not whether agents are rational but whether their behaviour can be predicted as if they were, and Becker’s economic-imperialism programme extends the model to crime, family, and discrimination on the same instrumentalist grounds. Second, Smith himself was more behaviourally complex than the textbook caricature: the Wealth of Nations grounds market exchange not in maximisation but in “the propensity to truck, barter, and exchange” — and locates motivation in an interplay of self-love, fellow-feeling, and status, not in expected-utility calculation.²⁰

Amartya Sen’s “Rational Fools” (1977) is the canonical philosophical critique that calibrates rather than dismisses rational choice — pointing out that an agent with a single preference ordering, indifferent to commitment and sympathy, is “a social moron.” The behavioural-economics work below refines the rational-choice apparatus; it does not refute it wholesale.

2.2 Kahneman and Tversky

Daniel Kahneman and Amos Tversky spent decades documenting the systematic ways in which human judgement deviates from the predictions of rational choice theory. Their key findings, summarised in Kahneman’s Thinking, Fast and Slow (2011):²¹

Anchoring. When asked to estimate a quantity (the population of Chicago, the length of the Amazon), people’s estimates are heavily influenced by arbitrary numbers they were exposed to just before — even when those numbers were generated by a roulette wheel in front of them and they know the numbers are random. Rational agents should not be influenced by irrelevant anchors. (Original demonstration: Tversky and Kahneman, “Judgment Under Uncertainty: Heuristics and Biases,” Science 185, 1974.)²²

Loss aversion. People weigh losses approximately twice as heavily as equivalent gains. A prospect of losing £100 is more aversive than a prospect of gaining £100 is attractive. This violates the symmetry required by standard utility theory and produces a range of irrational choice patterns (status quo bias, endowment effect, risk aversion in the domain of gains, risk-seeking in the domain of losses). (Original account: Kahneman and Tversky, “Prospect Theory: An Analysis of Decision Under Risk,” Econometrica 47.2, 1979.)²³

Availability bias. People estimate the probability of events by how easily examples come to mind. Plane crashes are overestimated relative to car accidents because they are more memorable; deaths by lightning are underestimated relative to deaths by falling furniture. Rational Bayesian agents should estimate probability from base rates, not from availability. (Original demonstration: Tversky and Kahneman, “Availability: A Heuristic for Judging Frequency and Probability,” Cognitive Psychology 5, 1973.)²⁴

Framing effects. Whether a decision is framed as a gain or a loss changes what people choose, even when the options are logically equivalent. A medical treatment described as “90% survival rate” is chosen more often than the same treatment described as “10% mortality rate.”

A caveat that matters. The four findings above — products of Kahneman and Tversky’s 1970s heuristics-and-biases programme — have weathered the replication crisis comparatively well. What has not survived is the social priming literature that grew on top of them in the 1990s and 2000s: Bargh’s “elderly priming” walking-speed studies, the “Florida effect,” lexical primes shifting downstream behaviour. Kahneman gave the stratum a full chapter in Thinking, Fast and Slow under the heading “The Marvels of Priming.”

Doyen and colleagues’ 2012 failure to replicate Bargh, the wider replication crisis, and Kahneman’s own 2017 open letter conceding that the priming chapter rested on shaky evidence have together pulled that stratum apart.²⁶ On the standard reading of the resulting picture, the judgment-under-uncertainty findings sit on firmer ground than the priming findings, and the two strata should be evaluated separately. On a more sceptical reading (Gigerenzer, parts of the small-effects literature), the same selection pressures that made priming look strong also touch the heuristics-and-biases material; how far the doubt spreads is itself contested.

2.3 The 2008 Financial Crisis

The 2008 financial crisis is often cited as a real-world stress-test of rational-agent models. The crisis was produced by mortgage-backed securities rated AAA by agencies whose models assumed US housing prices would not fall nationally (they had not done so since the 1930s); banks holding far more risk than their capital reserves could absorb; and a global financial system in which closely related models were being used by nearly every major institution at the same time. The acute phase ran from the failure of Bear Stearns (March 2008, sold to JPMorgan with Federal Reserve assistance) through Lehman Brothers’ Chapter 11 filing on 15 September 2008 — the largest bankruptcy in US history at over $600 billion in assets — to the $85 billion Federal Reserve loan to AIG the following day, which prevented a chain failure of credit-default-swap counterparties.²⁷

A careful description matters here. The operative models did not, strictly, assume agents would “behave rationally” in the philosophical sense. Ratings models assumed historical default correlations would continue to hold — a statistical assumption about the data, not a psychological assumption about agents. Bank risk models (Value-at-Risk, Gaussian copula pricing) used distributional assumptions that systematically underweighted tail events. The misjudgement that mattered was about the joint distribution of housing-market outcomes, not about any trader’s rationality. Some economists — Raghuram Rajan in 2005, Nouriel Roubini from 2006 — warned publicly, using broadly orthodox tools, that the system was accumulating tail risk; the failure was as much institutional and incentive-driven as it was a failure of the rational-agent abstraction.²⁸

Kahneman: “The illusion that we understand the past fosters overconfidence in our ability to predict the future.”²⁹ The traders, bankers, and economists who believed their models were, in many cases, genuinely overconfident — not dishonest. The models were seductive because they were elegant, mathematically sophisticated, and had worked well for a period when the underlying assumptions happened to hold.

2.4 The Evolutionary Function of “Irrational” Behaviour

Framing cognitive biases as “errors” or “bugs” presupposes a benchmark of rationality — the rational agent of economic theory — and measures human cognition against it. But this framing has an alternative: what if the biases are adaptations, behaviours that were functionally rational in the environments in which they evolved?

A warning before any example. Reverse-engineering specific cognitive biases to specific ancestral selection pressures is the most overreached move in evolutionary psychology. “Ancestral environment” is not a single thing; it covers hundreds of thousands of years across radically different ecologies. The evidence that any particular contemporary bias was selected for a particular function in that period is almost always indirect, and the field’s track record of confidently reverse-engineering specific adaptations has been poor. The points below are plausibility sketches, not established causal stories. Hold them accordingly.

Loss aversion is the usual example. A hypothesis: in a foraging context where losing your food cache can cost you your life and missing an extra cache cannot, a weighting of losses over gains would be functionally rational rather than a cognitive error — and the bias is only plainly irrational when applied to financial decisions in environments where the payoffs are symmetric. This is a plausible story and no more; direct empirical support that loss aversion was specifically selected on this pressure is limited.

The availability heuristic — judging the probability of events by how easily examples come to mind — is accurate in environments where memory is a good proxy for frequency, and arguably fails in modern media environments where salience and frequency are systematically decoupled (plane crashes are rare but memorable; car accidents are common but unremarkable). Again: this is a functional story about current performance across environments, not a verified account of what was selected for.

Anchoring — adjusting estimates insufficiently from an initial reference point — may reflect a general principle of efficient computation: start from the best available estimate and adjust, rather than constructing every estimate from scratch. The failure is not the principle but its application in contexts where the anchor is arbitrary or deliberately manipulated.

Robert Sapolsky (Behave, 2017)³⁰ argues that to understand any human behaviour, you must understand it at multiple levels simultaneously: the immediate neurological cause, the hormonal context of the past hours, the developmental history of the individual, and the evolutionary history of the species. Sapolsky’s analogy — calling a behaviour “irrational” without asking what selection pressure produced it is like calling a leg “irrational” for being poorly designed for swimming — is more rhetorical than evidential, since it licenses the conclusion only when the evolutionary etiology of the bias is established, which (per the warning above) it usually is not. What survives is a methodological prescription: do not call a behaviour an “error” without specifying the benchmark of correct functioning, and on what grounds that benchmark is taken to be relevant.

This does not vindicate the biases. Loss aversion still produces bad financial decisions in modern markets. The availability heuristic still leads to poor risk assessment in media-saturated environments. But understanding that they are adaptations rather than mere failures tells us something important: they will not be easily corrected by information or argument alone, because they are not produced by information-processing failures. Changing the context of decision-making (Thaler and Sunstein’s “nudge”) is more effective than changing the content of reasoning, precisely because the biases are structural rather than informational.

2.5 Questions to Argue About

• Kahneman distinguishes “System 1” (fast, intuitive) and “System 2” (slow, deliberate) thinking. Is irrational behaviour just System 1 operating in the wrong context? Or is it something more fundamental?
• Are cognitive biases like loss aversion and anchoring “bugs” in human cognition — features that should be corrected — or “features” that served adaptive purposes in our evolutionary environment?
• The 2008 crisis was partly caused by economists’ models of rational behaviour. Does this mean those models are simply false? Or were they appropriate in some contexts and misapplied in others?
• If people are systematically irrational, should governments use “nudges” to steer them toward better decisions? Or is this paternalistic in a way that undermines autonomy?

3 What can experiments tell us about human nature?

Ordinary people, under specific contrived laboratory conditions, do things they would have been confident they would never do — or so a famous run of mid-twentieth-century experiments was taken to show. The laboratory method, applied to human behaviour, isolates variables and (in principle) supports causal rather than merely correlational claims. Its findings have been treated as some of the most disturbing in social science. They are also contested: Le Texier’s 2018 archival reconstruction has substantially impeached the canonical reading of Zimbardo’s prison experiment, and the canonical reading of Milgram has been challenged by Perry’s archival work and by the Haslam–Reicher “engaged followership” reanalysis (though Burger’s 2009 partial replication cuts in the other direction). The popular take-aways from this literature have routinely outrun what the underlying experiments can support; how far the underlying findings themselves survive is what the next generation of historians of psychology is sorting out.

3.1 Milgram’s Obedience Experiments

Stanley Milgram designed an experiment in 1961–1962 in the basement laboratory of Linsly-Chittenden Hall at Yale University, published in 1963, that became one of the most replicated and discussed studies in psychology (“Behavioral Study of Obedience,” Journal of Abnormal and Social Psychology 67.4, 1963).³⁵ Participants — forty male volunteers per condition, recruited by newspaper advertisement and paid $4.50 — were told they were part of a Yale study on the effects of punishment on learning. Each played the role of “teacher,” while a confederate (a mild-mannered accountant named Mr Wallace) played the role of “learner,” strapped to a chair in an adjacent room with electrodes attached to his wrist. The teacher was instructed to administer electric shocks to the learner using a fake shock generator with thirty switches — labelled in 15 V increments from “Slight Shock” (15 V) through “Danger: Severe Shock” (375 V) to an unmarked “XXX” (450 V) — whenever the learner gave a wrong answer on a paired-associates word task. The shocks were fake; the learner’s protests, screams, and pounding on the wall were tape-recorded scripts standardised across participants; at 300 V the learner pounded on the wall and demanded to be released, and from 315 V onward fell silent — a silence the experimenter instructed the teacher to treat as a wrong answer.

When the experimenter (dressed in a grey laboratory coat) instructed the teacher to continue in the face of the learner’s apparent distress, 65% of participants in Milgram’s first variation administered what they believed were potentially lethal shocks at the maximum voltage. Milgram expected this rate to be around 1–2%.³⁶ The 65% figure is the one most often quoted, but Milgram ran eighteen variations of the experiment and obedience rates ranged from 0% (when two confederates rebelled) to 92.5% (when the teacher only had to relay orders rather than press the shock lever). Gina Perry’s 2013 archival reconstruction shows that Milgram and his collaborators were selective in which conditions they emphasised. The phenomenon is real; the canonical figure is one slice of it.

What does this prove? Several interpretations are in play:

• That ordinary people will commit atrocities when instructed to by legitimate authority (the “situationist” interpretation).
• That people in ambiguous situations defer to those who appear to know what they are doing.
• That human moral psychology is more fragile and context-dependent than we like to believe.

Milgram himself connected his findings to the Holocaust — ordinary Germans obeying orders. Hannah Arendt’s phrase “the banality of evil” (from Eichmann in Jerusalem, 1963, published the same year as Milgram’s study)³⁷ addresses the same phenomenon.

3.2 The Stanford Prison Experiment

Philip Zimbardo’s Stanford Prison Experiment, run over six days in August 1971 and funded by the US Office of Naval Research,³⁸³⁹ assigned twenty-four male student volunteers — selected from seventy-five applicants screened as the most psychologically stable — randomly to the roles of “guards” (n=12, working three at a time in eight-hour shifts) and “prisoners” (n=12) in a mock prison constructed in the basement of Jordan Hall, the Stanford psychology department. Prisoners were “arrested” at their homes by the Palo Alto police, booked, and dressed in numbered smocks; guards were issued khaki uniforms, mirrored sunglasses, and wooden batons (which they were told not to use). The experiment was scheduled to last two weeks; it was terminated after six days when it became clear that the “guards” were engaging in increasingly sadistic behaviour toward the “prisoners” — including forced exercise, sleep deprivation, and humiliation — and the “prisoners” were showing signs of genuine psychological distress, with five released early after acute breakdowns.

What Zimbardo claimed. The situation — the roles, the uniforms, the institutional context — produced the behaviour, not the individual characters of the participants. The same people, randomly assigned to different roles, behaved differently. For decades this was the canonical demonstration of situationism: character is weak; context is strong.

What the evidence actually supports. The canonical reading does not survive scrutiny. Jonah Lehrer’s investigations and later researchers (notably Thibault Le Texier’s archival work, published in French in 2018 and in American Psychologist in 2019)⁴⁰ revealed that Zimbardo had coached his “guards” to be tough at the experiment’s start; that he himself occupied the role of “prison superintendent” that made him an active participant, not a neutral observer; that the “prisoners” who broke down most dramatically may have been performing the distress they took to be expected of them; and that the experiment’s protocols were not remotely adequate for extracting a causal claim about situational power. The study is now regarded as methodologically flawed and ethically compromised. The weaker conclusion it can still support: strong institutional roles, plus an active authority figure shaping behaviour, can produce cruelty faster than participants expect — which is worth knowing, but is a much narrower claim than “situation determines character.”

3.3 The Replication Crisis

Beginning around 2011, it became clear that a significant proportion of landmark psychology experiments could not be reproduced. The Open Science Collaboration’s systematic replication attempt (2015) reproduced only 39% of 100 published psychology studies.⁴¹ Among social psychology studies — including the kind of high-profile situationist work represented by Milgram and Zimbardo — the replication rate was lower still.

The structural causes of the replication crisis are well understood: p-hacking (analysing data in multiple ways and reporting only the one that reaches statistical significance); small samples that produce unstable estimates; publication bias (journals publish positive results, not failed replications or null findings); and questionable research practices (deciding to stop collecting data once a significant result emerges, adding or removing participants after data collection begins).

The epistemological implication: a published finding in a peer-reviewed psychology journal is not reliable evidence of a true effect. That does not mean psychology finds nothing — but it requires a more sceptical reading of the literature than was previously common.

A parallel methodological response in economics — what is sometimes called the credibility revolution — has tightened what counts as a defensible causal claim from observational data. Joshua Angrist and Jörn-Steffen Pischke, in Mostly Harmless Econometrics (2009), make the benchmark explicit: a randomised trial is what you would like to run; absent that, the researcher’s job is to find a “natural experiment” — some accident of policy, geography, or biology that assigns the variable of interest as if at random.⁴²

Their canonical example is Angrist and Krueger’s 1991 study. Compulsory-schooling laws set a minimum age at which children may leave school; depending on quarter of birth, two children with the same intended school-leaving age end up with slightly different total years of schooling — the calendar, not their ambition, decides the difference. That accident of birth-month is what econometricians call an instrument: a variable that nudges schooling without itself depending on family background, ability, or motivation, and so lets the researcher read off schooling’s effect on adult wages without those confounders contaminating the estimate. The credibility revolution’s defenders take this to specify the narrow conditions under which observational social science can establish causation. Its critics — Angus Deaton on the imperialism of randomised controlled trials, James Heckman on external validity, structural and ethnographic schools more broadly — argue that the conditions are too narrow to recover most questions the human sciences need to answer, and that the methodological prestige of the technique has narrowed what economists ask.

3.4 Questions to Argue About

• Milgram showed that 65% of ordinary people will administer apparent lethal shocks when instructed by an authority figure. What exactly follows from this? Does it tell us something general about human nature, or something about behaviour in highly specific contexts?
• The Stanford Prison Experiment has been criticised as methodologically flawed and ethically compromised. But its findings — that role assignment shapes behaviour — are widely cited. Can flawed studies produce valid insights?
• The replication crisis shows that many landmark findings don’t hold up. Does this mean the human sciences have been producing false knowledge, or just that they are correcting themselves over time?
• What ethical constraints should govern psychological experiments? Is there any level of harm to participants that can be justified by the knowledge the experiment produces?

4 Can the human sciences make predictions?

If the human sciences cannot predict anything, it is unclear in what sense they are sciences rather than sophisticated storytelling about the past. The natural sciences make their reputation on prediction — general relativity tells us in advance where starlight will bend; quantum mechanics tells us in advance what particles will do. If the human sciences can also predict, the predictions should be checkable. If they cannot, we should be honest about what they are instead. The record, when examined honestly, is mixed at best — and in some cases, actively instructive about the difficulty of the enterprise.

4.1 Economic Forecasting

The discipline of economics has a poor record of predicting economic crises. The 2008 financial crisis was not predicted by most mainstream economists; the 2020 COVID-19-induced recession was genuinely unpredictable, but the depth and duration of the subsequent recovery was again misjudged by most forecasters. The IMF, the World Bank, and national treasury departments routinely produce GDP forecasts that are significantly wrong within two years.

This failure has been explained in several ways. Complexity: economies are systems with billions of interacting agents, each responding to expectations about what others will do; small changes in initial conditions produce large differences in outcomes (sensitivity to initial conditions). Self-defeating prophecy: economic forecasts affect the behaviour of agents who read them, changing the conditions being forecast. Model inadequacy: standard macroeconomic models assume rational agents, smooth adjustment processes, and equilibrium — assumptions that fail catastrophically during crises.

The physicist-turned-financial-theorist Jean-Philippe Bouchaud has argued that economics suffers from “physics envy” — the desire to achieve the precision and predictive power of physics using models that are not accurate descriptions of economic reality.⁴⁷ The mathematical sophistication of the models can be inversely related to their real-world accuracy.

4.2 Malthus and the Demographic Prediction That Failed

Thomas Malthus, in An Essay on the Principle of Population (1798),⁴⁸ argued that population growth inevitably outstrips food production (because population grows geometrically while food production grows arithmetically), leading to periodic catastrophes of famine, disease, and war that restore population to sustainable levels. His prediction: given 18th-century trajectories, mass starvation was mathematically inevitable.

The prediction failed. Why? Malthus did not anticipate the Agricultural Revolution, the Industrial Revolution, the Green Revolution of the 1960s, or the demographic transition (the consistent finding that as societies industrialise, birth rates fall). Each of these involved non-linear changes that were not extrapolations of existing trends. Historical prediction based on extrapolating current trends fails when the conditions generating those trends change.

4.3 Cliodynamics

Peter Turchin, in Ages of Discord (2016) and earlier work, proposes a mathematical approach to historical dynamics he calls cliodynamics.⁴⁹ His model identifies two cycles of political instability: a roughly 50-year cycle driven by generational dynamics, and a longer “secular cycle” of 200–300 years driven by elite overproduction (growing numbers of elite aspirants competing for a limited number of elite positions) and popular immiseration (the declining share of income going to labour). When both cycles peak simultaneously, political violence increases.

In 2010, Turchin predicted that the US would experience severe political instability around 2020.⁵⁰ The 2020 racial justice protests, the COVID-19 political crisis, and the January 6, 2021 Capitol attack have been cited as confirmations.

The sceptical objections: the prediction is broad enough to encompass a range of outcomes; the model has few genuine out-of-sample tests; the categories (what counts as “political instability,” “elite overproduction”) involve significant interpretive choices; and there is a risk of selection bias — selecting historical cases that fit the model while ignoring those that don’t. Turchin’s response is that his model makes quantitative predictions (with confidence intervals) that can be tested, not just post-hoc rationalisations.

4.4 Popper’s Challenge

Popper argued that the social sciences should be judged by the same falsifiability standard as natural science: produce falsifiable predictions, test them, and revise or abandon theories that fail. He argued, in The Poverty of Historicism (1957),⁵¹ against any social science that makes claims about inevitable historical trajectories (he was targeting Marxist historical materialism). Grand historical laws, he argued, cannot be tested and should be abandoned.

The question is whether Popper’s criterion is appropriate for the human sciences, given their subject matter. History does not repeat; controlled experiments are rarely possible; the “ceteris paribus” conditions required for a clean test are almost never satisfied. Perhaps the human sciences require different standards of rigour that are adapted to their subject matter rather than imported from physics.

4.5 Questions to Argue About

• Economic forecasting has a poor record. Does this mean economics is not a science? Or that it is a science of a different kind from physics?
• Malthus’s prediction failed because he didn’t anticipate structural change. Is structural change in principle predictable? Or does this mark a fundamental limit on social science prediction?
• Turchin predicted US political instability around 2020. The events of 2020–21 seem to confirm this. Is this genuine scientific prediction, or is the model flexible enough to accommodate any outcome?
• Popper argues that social sciences should make falsifiable predictions or stop claiming to be sciences. Is this the right standard for the human sciences? What alternative criteria of rigour might be more appropriate?

5 How do culture and society shape what we know — and what we study?

The human sciences study human beings. But human scientists are also human beings — members of particular societies, shaped by particular educational systems, funded by particular institutions. This sounds like a truism, the sort of thing you could put on a poster with a sunset photograph. It is not a truism. It is a specific claim with specific consequences: their research is conducted on particular populations, using methods developed in particular cultural contexts, and this creates a systematic risk of mistaking culturally specific findings for universal truths — then building entire disciplines on the mistake.

5.1 WEIRD Bias

In a landmark 2010 paper, Joseph Henrich, Steven Heine, and Ara Norenzayan documented what they called the WEIRD problem:⁵⁶ in a survey of papers in six top psychology journals between 2003 and 2007, 96% of subjects came from Western, Educated, Industrialised, Rich, and Democratic societies — populations that represent roughly 12% of the world. (The figure is sample-specific; the broader pattern is well-attested across reviews.)

More importantly, WEIRD populations are outliers on many psychological dimensions, not typical humans. The Müller-Lyer illusion (see Core) has a much larger effect on WEIRD populations than on populations who live in environments without rectangular rooms and straight lines. Concepts of fairness, as measured by standard economic games (the Ultimatum Game), vary dramatically across cultures in ways that the WEIRD literature had not captured. Self-construal (the degree to which people define themselves as independent agents vs. members of groups) varies substantially across cultures and shapes fundamental cognitive patterns.

The implication Henrich and colleagues draw: psychological universals claimed on the basis of WEIRD samples may not be universal at all but features of a particular type of society. The implication critics of the WEIRD argument draw (Norenzayan in his subsequent work; Apicella and others) is more limited: WEIRD samples bias the generality of the claims that can be made, but not every finding is equally affected, and several of the cases the critics cite (basic perceptual constancies, fundamental memory architecture) do hold up cross-culturally. How wide the WEIRD problem actually is remains an empirical question being worked through.

5.2 Margaret Mead and Derek Freeman

Margaret Mead’s Coming of Age in Samoa (1928) claimed that adolescent sexual behaviour in Samoa was largely free of the storm and stress characterising Western adolescence, demonstrating that what Americans took to be natural features of puberty were culturally determined. The book was a landmark of cultural anthropology and of progressive social thought: it provided evidence that human nature was malleable, that culture rather than biology determined behaviour.

In 1983, Derek Freeman published Margaret Mead and Samoa: The Making and Unmaking of an Anthropological Myth, arguing that Mead had been systematically misled by her informants, who told her what she wanted to hear, and that Samoan adolescence was not in fact unusually harmonious. Freeman’s critique was itself contested: he was accused of political motivation (he was a critic of cultural determinism), of selectively using his own evidence, and of misrepresenting Mead’s actual claims.

The controversy is epistemologically instructive not because it proves Mead wrong (the jury is still out on specific factual claims) but because it illustrates the structural vulnerabilities of anthropological fieldwork: the observer’s expectations shape what they observe; fieldwork is conducted under conditions that make independent verification difficult; and the theoretical commitments of the field shape what kinds of findings get published and cited.

5.3 Relativism and Universalism

The question that the Mead-Freeman debate frames: is there a universal human nature, or are humans sufficiently plastic that culture determines everything important?

Relativism: human behaviour, values, and ways of knowing are culturally specific. What counts as rational, what counts as beautiful, what counts as a good life — these vary so much across cultures that they cannot be measured by a single universal standard. The attempt to apply universal standards is typically an exercise of cultural imperialism.

Universalism: the anthropological evidence for cultural variation is real but overstated. Some things do not vary — basic emotional expressions, certain cognitive capacities, a range of social structures and prohibitions. The debate is not between variation and universality but about where the limits of variation lie.

Paul Ekman’s research on facial expressions of emotion (1971) claimed to find six universal basic emotions (happiness, sadness, fear, disgust, anger, surprise) expressed in culturally invariant ways.⁵⁷ This has been challenged by Lisa Feldman Barrett’s theory of constructed emotion, which argues that emotional expressions are more culturally variable than Ekman claimed.⁵⁸ The dispute is ongoing.

A note on how live this binary actually is. Working anthropology has largely moved past the relativism-versus-universalism stand-off that TOK textbooks treat as the central question. After Geertz’s interpretive turn, after Said’s Orientalism, and after the reflexive critique of “the native point of view,” most contemporary fieldworkers practise something closer to calibrated humility — a methodological assumption that observed differences require careful explanation before judgement, without thereby committing to incommensurability or to the claim that no cross-cultural comparison is possible. Webb Keane’s Ethical Life (2016) and the late work of Marshall Sahlins are representative; the Cultural Anthropology journal has run a long line of “Beyond the Native Point of View” debates in the same direction.⁵⁹ The binary is still useful as a TOK exam frame, because it sharpens the issues; but treating it as the live state of the discipline overstates how cleanly the field divides.

5.4 Questions to Argue About

• The WEIRD bias shows that psychology has been generalising from a non-representative population. Does this invalidate existing psychological findings, or only make them less general?
• Margaret Mead found evidence for cultural determinism because she was looking for it. Does this mean anthropological fieldwork is inevitably circular? Or are there methods that can discipline it?
• Is anthropological relativism — the principle that cultures should be understood on their own terms — an epistemological principle (the most accurate method) or an ethical one (cultures deserve respect)? Does the distinction matter?
• If some psychological universals (like the Müller-Lyer illusion’s effects) vary across cultures, what remains of claims about universal human nature? How would you distinguish genuine universals from WEIRD generalizations?

6 Is there a human nature?

Aristotle argued that “the city belongs to the class of things that exist by nature, and that man is by nature a political animal” — a zōon politikon, a creature of the polis whose flourishing is constituted in political community rather than added to it.⁶¹ Nobody has successfully refuted him, though a great many have tried, often with the implicit corollary that if we could just change the social arrangements, we could change the animal. The question of human nature is simultaneously empirical (what are we actually like?) and political (what does the answer justify?). Different answers have been used to justify very different social arrangements, and the political stakes have made the empirical debate hard to conduct dispassionately. We have a powerful interest in the answer — which is precisely why we should be suspicious of whoever gives it with the most confidence.

6.1 The Bucharest Early Intervention Project

In the late 1990s, Romania’s transition from the Ceaușescu regime — whose pronatalist policies had institutionalised over 100,000 children — left orphanages whose conditions were severe by any measure. From 2000, a research team led by Charles A. Nelson (Harvard / Boston Children’s Hospital), Nathan A. Fox (University of Maryland), and Charles H. Zeanah (Tulane) ran the Bucharest Early Intervention Project (BEIP), a randomised controlled trial that recruited 136 institutionalised children aged 6–31 months from six Bucharest orphanages.⁶⁶ Children were randomly assigned either to continued institutional care (the existing standard for Romanian orphans) or to high-quality foster care that the project arranged and supervised. Assessments at 30, 42, and 54 months showed that children placed in foster care before about 24 months had substantially better outcomes on cognitive testing, language development, and brain electrical activity (EEG and event-related-potential measures) than those who remained institutionalised; children placed after 24 months showed measurable but smaller gains.⁶⁷ Sixteen-year follow-up data showed persistent EEG, attachment, and psychiatric-symptom differences between the institutional and foster-care groups; the foster-care group remained substantially better off but neither reached the typical-development comparison group.⁶⁸ The research raised acute ethical questions — the team had to argue that the control arm was the existing standard of care and that the trial was defensible only because the project was simultaneously producing the foster-care alternative the study would later show was better. The BEIP is the contemporary, randomised, ethically-supervised counterpart to the Verdingkinder (info box above) and to the case of Genie (treated below) — three different evidential structures (RCT, historical state cohort, single tragic biography) addressing the same question of what becomes of children deprived of normal developmental conditions.

6.2 The Genie Case (1970)

In November 1970, a thirteen-year-old girl known to researchers as Genie was discovered in a house in Arcadia, California, where she had spent her entire childhood in isolation, strapped to a potty chair in a darkened room, forbidden to make noise, beaten if she did. Her father believed she was brain-damaged and enforced near-total sensory and linguistic deprivation. When Genie was found, she could not speak and had no language. Linguist Susan Curtiss spent years attempting to teach her English; Genie acquired a vocabulary and a rudimentary ability to communicate. She never acquired grammar. Curtiss argued that Genie’s case was direct evidence for Chomsky’s theory of a critical period for language acquisition:⁶⁹ the capacity for language is biological and time-limited, and if the critical window closes without exposure, some aspects of linguistic competence are permanently unavailable. The case sits at the intersection of Hobbes, Locke, Rousseau, and Pinker. It is not a clean illustration of any of their positions: Genie was not a blank slate written on by culture, but nor was her biological endowment sufficient to produce full language without environmental input at the right time. Human nature, on this reading, is neither a fixed essence nor an entirely social construction but a set of developmental potentials that require specific environmental conditions to actualise. The Bucharest Early Intervention Project (treated above in the body) is the methodologically rigorous controlled successor — randomised, large-sample, ethically supervised — to the question that Genie’s biography could only test through a single tragic life; the Verdingkinder archive (info box above) provides the third evidential register, a Swiss state-record cohort across two centuries.

6.3 The Philosophical Background

Three foundational positions, simplified but not caricatured:

Hobbes (Leviathan, 1651): in the “state of nature,” human life is “solitary, poor, nasty, brutish, and short.”⁷⁰ Without the coercive apparatus of the state, humans will fight each other for scarce resources. Human nature is competitive, self-interested, and requires external constraint to produce social order.

Locke (Two Treatises of Government, 1689):⁷¹ humans are rational, cooperative, and capable of governing themselves through consent. The state exists not to suppress natural brutishness but to protect natural rights that exist prior to it. Human nature is not benign, but it is not primarily aggressive either.

Rousseau (Discourse on the Origin and Foundation of Inequality Among Men, 1755):⁷² humans are naturally good and peaceful; it is society — the introduction of property, inequality, and competition — that corrupts them. “Man is born free, and everywhere he is in chains.”⁷³

These are not just historical curiosities. They shape contemporary debates about criminal justice (are criminals bad people or products of bad circumstances?), economic policy (do people need incentives and constraints to work productively, or do they work from intrinsic motivation?), and political organisation (do democracies require educated, engaged citizens, or can they function with strategic self-interested voters?).

6.4 Pinker’s Case for Human Nature

Steven Pinker, in The Blank Slate (2002),⁷⁴ argues that the dominant view in social science — what he calls the “blank slate” (no innate human nature), “noble savage” (natural goodness corrupted by civilisation), and “ghost in the machine” (mind separate from brain) — is empirically refuted by the evidence from evolutionary psychology, genetics, and neuroscience.

Humans have a nature shaped by millions of years of evolution: cognitive architecture, emotional systems, social instincts, and behavioural tendencies that are part of what we are rather than products of socialisation. This does not mean these tendencies are unchangeable or morally justified; it means they must be understood rather than denied.

Pinker’s The Better Angels of Our Nature (2011)⁷⁷ extends the argument: despite our evolved tendencies toward violence, rates of violent death (per capita) have declined dramatically over millennia, and the decline can be explained by the growth of states, trade, literacy, and the expansion of moral empathy. Human nature is malleable in important ways — not infinitely, but enough to produce a measurable decline in violence.

6.5 Sartre’s Counter

Jean-Paul Sartre’s existentialist counter begins with a slogan:

“Existence precedes essence.” — Jean-Paul Sartre, Existentialism is a Humanism (lecture 1945, published 1946)⁷⁸

For Sartre, there is no fixed human nature that precedes individual existence. We are not made with a predetermined essence, the way a paper knife is made with a predetermined function. We exist first, define ourselves through our choices, and create our own nature in the process. This is the source of both human freedom and human anguish: there is no nature to appeal to, no pre-given function to fulfil. We are “condemned to be free.”⁷⁹

The empirical critique of Sartre, advanced most forcefully by Pinker and the evolutionary-psychology programme, is that the blank-slate view of human nature is implausible: there are documented species-typical cognitive architectures, emotional systems, and developmental constraints (the Verdingkinder, the Bucharest Early Intervention Project, and the Genie case, all treated above, are examples on three different evidential registers). Whether this refutes Sartre depends on the reading of “essence precedes existence” one starts with. The strongest reading — that humans have no species-typical cognitive endowments and are infinitely malleable — is hard to defend after a half-century of cognitive science. The weaker and more interesting reading — that whatever endowment there is underdetermines what an individual human becomes, leaving genuine choice in the gap — is compatible with the evidence and is closer to what Sartre actually argued in the 1945 lecture. The load-bearing claims are about the phenomenology of agency, the experience of choosing without a script, and what it means to take responsibility for one’s choices rather than blaming them on nature, society, or God — not a metaphysical denial of biology.

6.6 Questions to Argue About

• Is the debate between “human nature is fixed” and “humans are blank slates” a genuinely empirical dispute? Or is it partly a political one, with empirical findings used to support pre-existing commitments?
• If evolutionary psychology shows that humans have evolved tendencies toward violence, dominance, and tribalism, what follows politically? Does knowing this change anything — or does it just redescribe what we already knew?
• Sartre says there is no human nature — we define ourselves through choices. But if our choices are themselves shaped by neurology, genetics, and socialisation, is the “freedom” Sartre describes real?
• What is at stake, politically, in the claim that humans are naturally good (Rousseau) versus naturally brutish (Hobbes)? Can the empirical question be answered without addressing the political stakes?

7 What is the relationship between the human sciences and power?

Every human science category is a political act as well as a scientific one. To define “mental illness” is to decide who needs treatment and who needs policing. To measure “intelligence” is to rank people and justify the ranking. To classify societies as “developed” and “developing” is to establish a hierarchy with specific beneficiaries. Human science knowledge is not produced in a vacuum. It is funded by governments and corporations; it is applied in clinical, judicial, and policy settings; and its categories — mental illness, race, intelligence — have shaped who gets included in and excluded from social goods. The relationship between knowledge and power in the human sciences is not incidental. It is structural.

7.1 The 1973 Removal of Homosexuality from the DSM

Until 1973, the Diagnostic and Statistical Manual of Mental Disorders classified homosexuality as a psychiatric disorder — specifically, as a “sociopathic personality disturbance” in DSM-I (1952), retained through DSM-II. The classification was used to justify criminal prosecution, involuntary institutionalisation, electroconvulsive therapy, and chemical castration of gay and lesbian individuals. The relevant scientific evidence had been available for over a decade: Evelyn Hooker’s 1957 study had given the Rorschach, Thematic Apperception Test, and Make-A-Picture-Story tests to thirty homosexual and thirty heterosexual men, matched for age, IQ, and education, and asked expert clinicians (blinded to orientation) to identify pathology and to pick the homosexual subjects from the ratings. They could not — performance was at chance.⁸³ The 1973 removal did not result from new scientific evidence Hooker’s work had overlooked. It resulted from a sustained campaign by gay-rights activists — including Frank Kameny’s confrontational disruption of the 1970 and 1971 APA conventions in San Francisco and Washington — combined with the internal work of psychiatrist Robert Spitzer, who in 1973 drafted a reformulation that removed “homosexuality” from DSM-II while introducing a new diagnosis, “sexual orientation disturbance,” for individuals distressed by their orientation. On 15 December 1973 the APA Board of Trustees adopted Spitzer’s reformulation 13–0 with two abstentions; opposing psychiatrists then forced a 1974 membership referendum, which ratified the Board’s decision by 5,854 to 3,810 — about 58 per cent in favour of removal.⁸⁴ The episode is the clearest illustration of Foucault’s argument about the relationship between psychiatric knowledge and power: the diagnostic category had not been produced by neutral scientific inquiry but had reflected, and reinforced, the dominant sexual norms of mid-century American society. Its removal did not resolve the philosophical question — it opened it, by showing that the category’s production was a social and political process that could be reversed by social and political action. The 2013 DSM-5 reclassification of gender dysphoria and the 2018 ICD-11 reclassification of gender incongruence (info box above) are the same dispute, two generations on, with a different category in the dock.

7.2 Foucault’s Archaeology

Michel Foucault, in Discipline and Punish (1975)⁸⁵ and The History of Sexuality (1976),⁸⁶ argues that the human sciences are not neutral descriptions of human nature but instruments of what he calls power/knowledge (pouvoir/savoir) — the intertwining of knowledge production and social control. To classify, describe, and categorise human behaviour, on Foucault’s account, is to exercise power over it: to define what counts as normal and what counts as pathological, who counts as a subject and who as an object.

Foucault has critics. Ian Hacking accepts much of the historical analysis but distinguishes “interactive kinds” (categories that affect the people they classify) from “indifferent kinds” (natural-scientific categories that do not), arguing that Foucault’s blanket claim about all human-scientific classification overshoots the data — some psychiatric categories track real biological substrates regardless of how the category is socially deployed.⁸⁷ Nancy Fraser presses a different objection: if all knowledge is power, the framework cannot distinguish knowledge that liberates (the homosexuality declassification) from knowledge that oppresses (the original classification), since both involve power, knowledge, and their interweaving.⁸⁸ These are friendly criticisms within the broadly post-Foucauldian literature; they bound the claim rather than refuting it.

The psychiatric categorisation of homosexuality as a mental disorder is the case Foucauldians treat as paradigmatic. It appeared in the DSM from 1952 to 1973 (procedural detail in the info-box above). Hooker’s 1957 evidence had been on the record for over a decade before the removal without moving the institutional apparatus; what moved it was the combination of that evidence, the work of internal psychiatric reformers (Kameny, Silverstein, Spitzer), and sustained political pressure from the gay rights movement.

The Foucauldian reading: the classification reflected and enforced social norms about sexuality, and its dissolution required political work. The non-Foucauldian reading: the APA corrected an empirical error that its members were professionally and politically slow to correct, and the relevant evidence was scientific rather than political. Hacking’s interactive-kind framework offers a third synthetic option — the category was both an empirical mistake and a social act whose effects on those classified shaped what the category came to mean — though Foucauldians and realists about psychiatric kinds dispute whether Hacking’s middle position is genuinely a synthesis or a concession to the side opposite to the one each of them defends.

7.3 IQ Tests and Immigration Restriction

In the early 20th century, the American eugenics movement used psychometric research on intelligence to support immigration restriction. Henry Goddard administered IQ tests (translated versions of the Binet-Simon scale) to immigrants arriving at Ellis Island in 1912 and concluded that 83% of Jews, 80% of Hungarians, 79% of Italians, and 87% of Russians were “feeble-minded.”⁸⁹ These findings were presented as objective science and were cited in Congressional debates that led to the Immigration Act of 1924, which dramatically restricted immigration from Southern and Eastern Europe.

The methodological failures were spectacular: the tests were given in English to recent immigrants who spoke no English; the tests measured cultural familiarity with American middle-class life, not cognitive capacity; the samples were not representative. But the framing of the results as objective scientific measurement gave them a rhetorical power that more evidently political claims would not have had.

7.4 Decolonising Anthropology

Anthropology developed as a discipline in the 19th and early 20th centuries, while the major European powers were administering colonial empires. Colonial administrators needed to understand the societies they governed; anthropologists often produced that understanding. Talal Asad’s edited volume Anthropology and the Colonial Encounter (1973) made the entanglement explicit, and the field has been working with the resulting questions ever since.⁹¹ The historical record does not, however, support a simple “anthropology was a colonial tool” reading: Franz Boas and his American students (Mead, Benedict, Hurston) used the discipline to argue against the racial hierarchies the European powers were imposing, and several British social anthropologists (Evans-Pritchard among them) were treated with suspicion by the colonial administrations they were supposed to serve.⁹²

The decolonisation programme — the attempt to reconstruct the discipline’s methods, categories, and institutional structures in ways that do not replicate colonial hierarchies — is an active research agenda associated with Linda Tuhiwai Smith (Decolonizing Methodologies, 1999), the Subaltern Studies collective, and more recent work by Faye V. Harrison and Eduardo Viveiros de Castro. It has both wide internal acceptance (most contemporary fieldworkers grant the basic point about positionality and consent) and contested edges: whether all Western analytic categories are colonial in the relevant sense (Adam Kuper argues they are not, in The Reinvention of Primitive Society, 2005); whether alternative “indigenous methodologies” can sustain cross-cultural comparison without collapsing into the relativism the WEIRD literature criticises; and whether the institutional reforms proposed (community veto over research outputs, indigenous co-authorship requirements) help or hinder the production of knowledge that travels.⁹³

The “tribe” example is the canonical case for the decolonisers: the term was used by colonial administrators to impose fixed ethnic categories on more fluid social formations and was then absorbed into academic anthropology as a natural unit of analysis.⁹⁴ The example is real but less universal than the strongest decolonisation arguments imply: kinship-based segmentary organisation is a documented social form (Evans-Pritchard’s Nuer, Marshall Sahlins’s analysis of Polynesian chiefdoms) whose existence does not depend on the colonial label. The methodological lesson is narrower than the political slogan: imported categories must be tested against local data rather than assumed; whether they survive that test is a case-by-case question, not a general one.

The deeper methodological tension that decolonisation surfaces — whether any social science can be both cross-culturally generalisable and culturally specific — is unsolved and is one of the central methodological tensions in contemporary anthropology, sociology, and development economics.

7.5 Questions to Argue About

• Foucault argues that the human sciences produce “power/knowledge” — that scientific classification is always also an exercise of power. Does this mean human science knowledge is less valid than natural science knowledge? Or does it just add a dimension of analysis that natural science lacks?
• Homosexuality was classified as a mental disorder until 1973, then reclassified by a political process rather than new scientific evidence. What does this tell us about the DSM as a scientific instrument?
• The IQ tests administered to Ellis Island immigrants produced systematically biased results that were presented as objective science. How should this history affect our assessment of current psychometric research?
• Can anthropology be “decolonised” while remaining a rigorous academic discipline? Or does decolonisation require changes that would undermine the conditions for rigorous inquiry?

8 Can the human sciences be value-free?

Weber had a word for it — Wertfreiheit, value-freedom — and he was honest enough to admit that he was describing an aspiration, not a description. The ideal holds that the scientist’s job is to describe and explain, not to prescribe or judge. Facts about how societies work should be kept separate from values about how they ought to work. This seems methodologically reasonable. The problem is that every act of measurement is also an act of valuation — you cannot choose what to count without already having decided, at some level, what matters. The ideal turns out to be much more difficult to achieve than it looks — and, as the Swedish economist Gunnar Myrdal argued, the attempt to pursue it actively conceals the values that are already operative.

8.1 Weber’s Wertfreiheit

Weber articulated the ideal clearly: the social scientist must keep their value judgements out of their scientific work.⁹⁷ Not because values are unimportant, but because the scientist’s authority comes from their command of facts and methods, not from their moral commitments. A sociologist who lets their preference for socialism or liberalism distort their empirical findings has abused their professional role.

Weber was not naive about the difficulty. He knew that values inevitably shape which questions social scientists ask, which phenomena they find worth studying, and which categories they use. But he thought that rigorous self-discipline could manage these influences sufficiently for empirical findings to have objective standing.

8.2 Myrdal’s Critique

Gunnar Myrdal, the Swedish economist and sociologist, argued in The Political Element in the Development of Economic Theory (1930)⁹⁸ that Weber’s ideal is systematically unreachable, and that the attempt to pursue it actively hides the values that are actually operative.

Economic categories are not neutral. “Unemployment” as a category defines some people as being in an economically relevant relationship to the labour market and others as not; the choice of who counts as “unemployed” (not those in unpaid care work, not the “voluntarily unemployed,” not the long-term sick) is a political choice. “Growth” as the primary measure of economic success embeds a valuation of certain kinds of economic activity over others. Choosing GDP as the measure of national welfare is itself a value judgement — it says that welfare consists in material production and consumption, not in leisure, community, environmental quality, or spiritual life.

The values are not introduced by individual scientists making bad choices; they are built into the apparatus of measurement and the categories of analysis. Making them explicit — asking “whose interests does this framework serve?” — is not a corruption of social science; it is a form of scientific rigour.

8.3 Positive and Normative

The standard distinction in economics: positive economics describes how the economy works (“a minimum wage increase of £1 reduces employment by X%”); normative economics prescribes how it ought to work (“the minimum wage should be increased”). The positive/normative distinction is supposed to demarcate the scientist’s domain from the politician’s.

The difficulty: even if positive findings are value-free, the choice of which positive questions to investigate is value-laden. If economists predominantly study the effects of minimum wage increases on employment and rarely study the effects of low wages on worker health and family stability, that selection reflects values even if each individual study is “positive.” The choice of what to measure, what to call a “cost” and what to call a “benefit,” and what causal pathways to investigate are all value-laden choices embedded in apparently neutral analysis.

8.4 Questions to Argue About

• Weber says the social scientist should keep value judgements out of their scientific work. Is this achievable? Or does every choice of category, method, and research question already involve value commitments?
• Myrdal argues that the attempt to achieve value-free social science hides the values that are actually operating. Is explicit value commitment in social science more honest, or more dangerous?
• GDP measures national welfare in a specific way that reflects specific values about what matters. What would it mean to replace GDP with a different measure? Who should decide what to measure?
• The positive/normative distinction in economics is supposed to separate facts from values. But can they ever be cleanly separated in practice? Give an example where you think they bleed into each other.

9 Media

• Adam McKay, The Big Short (2015) — An unusually epistemologically sophisticated film about the 2008 financial crisis; it explains financial instruments accurately while dramatising the gap between what the models assumed and what the world did. The device of having characters break the fourth wall to explain concepts is unusual and effective.
• Daniel Kahneman, Thinking, Fast and Slow (2011) — Part IV (“Choices”) examines the gap between rational choice theory and actual human decision-making and remains the strongest material in the book. The Part I social-priming chapters have largely failed to replicate; Kahneman himself acknowledged in 2017 that he had “placed too much faith in underpowered studies.” Read Part IV as the load-bearing argument and Part I as a cautionary tale about how recently we have learned which findings will hold up.
• Hannah Arendt, The Origins of Totalitarianism (1951) — The third section, “Totalitarianism,” examines how political violence is produced not by uniquely evil individuals but by institutional and ideological structures. The connection to Milgram’s later findings is direct and profound.
• Joshua Oppenheimer, The Act of Killing (2012) — perpetrators of the 1965 Indonesian massacres re-enact their killings for the camera in the style of Hollywood genres. The film raises questions about what human beings are capable of, and how they understand themselves, that no experimental psychology can fully address.
• Asghar Farhadi, A Separation (2011) — A divorce in contemporary Tehran becomes a sustained meditation on testimony, class, religious obligation, and what it costs to discover the truth about another person’s actions. The film makes visible how human-scientific knowledge — sociological, legal, psychological — is produced under conditions of conflicting interest and incomplete information.
• George Orwell, The Road to Wigan Pier (1937) — Part I is a documentary account of working-class life in northern England; Part II is an essay on socialism and class that is methodologically self-aware about the observer’s position. A model of engaged social science that does not pretend to be value-free.
• Studs Terkel, Working (1974) — Oral history interviews with 133 Americans about their experience of work. The methodology is explicitly qualitative and testimonial; the book is both a rich empirical source and an argument about what matters in the study of economic life.
• Robert Sapolsky, Behave: The Biology of Humans at Our Best and Worst (2017) — A neurobiologist traces any human behaviour through its immediate neural causes, its hormonal context, its developmental history, and its evolutionary origins. The most important demonstration that biology and social science must be integrated rather than sequenced: you cannot understand human rationality, morality, or cooperation without all the levels simultaneously.

10 Bibliography

11 Notes