When the Evidence Becomes the Offence

Roland Fryer’s 2016 research on police use of force produced a finding that unsettled a fashionable account of race and policing: racial disparities appeared in non-lethal encounters, but his data did not show racial bias in officer-involved shootings. The reaction to that result, and the wider record assembled by Vinay Prasad, raises a question more serious than whether one study was right: how often is research praised as “data-driven” only when it reaches the politically approved conclusion?

That question places Harvard psychologist Steven Pinker’s defence of science and professional departments under uncomfortable pressure. Pinker has argued that universities may be ideologically captured in parts of the humanities while remaining more faithful to evidence in science, medicine and related fields. Prasad, a haematologist-oncologist and epidemiologist, disputes the safe boundary. His catalogue of cases argues that the same pressures can operate inside medicine, public health, economics, and the research journals that claim to police standards of proof.

The dispute is not really about whether scientists have politics

Scientists have always had politics. They vote, campaign, join institutions, receive grants and form professional alliances. None of that, by itself, proves that a discipline has abandoned science. The more precise question is what happens when political commitments influence the treatment of evidence: which questions are worth asking, which results deserve headlines, which caveats are treated as responsible scholarship and which are described as dangerous.

That distinction matters because the modern research system does not rely on a single act of censorship. A result can be filtered at several points. A young researcher can decide not to pursue a question because the likely finding will damage a career. A supervisor can advise an author to publish only the politically comfortable half of a project. A journal can accept a paper whose methods are weak because its conclusion feels morally useful, then reject a better-controlled paper because its conclusion is unwelcome. A university can defend a scholar who repeats the approved position and condemn one who challenges it. A newspaper can convert an uncertain association into a moral fact before readers have seen the underlying design.

None of these decisions needs to be announced as “we are suppressing an inconvenient result”. Institutions use more respectable language. They speak of harm, context, sensitivity, responsible communication, community standards, structural violence or the need not to legitimise misinformation. Those concerns can be valid. Medical claims can cause harm; public-health advice can save lives; a misleading paper can distort policy. But such language can also become a way of ending an argument before the evidence has been examined.

Prasad’s central charge is not that every progressive claim is false. It is that the burden of proof changes with the direction of the claim. If the data appear to confirm an account of racial bias, the work is treated as a breakthrough even when the design is vulnerable. If the data complicate that account, the same standards are suddenly invoked to explain why the finding should not count. This logic turns methodological caution into a one-way gate.

The danger is not confined to one political side. Conservative institutions can perform the same trick with different conclusions. A government that suppresses climate findings, a company that buries an unfavourable drug trial or a public-health agency that hides uncertainty is also filtering evidence through power. The point is not that one faction has a monopoly on bias. Institutions claiming scientific authority must be judged by whether they apply their standards consistently when the result is politically inconvenient.

Fryer’s finding became a test of academic courage

Fryer’s paper remains the clearest case because the result was neither a simple vindication of police nor a simple indictment. The Harvard economist examined racial differences in police use of force. The paper reported that Black and Hispanic people were more than 50 per cent more likely to experience some form of force in police interactions, even after the researchers added controls for context and civilian behaviour. Yet, for officer-involved shootings, the paper reported no racial difference in the raw data or after contextual factors were taken into account.¹

The finding should not have been a test of academic courage!

That distinction is precisely what made the work useful. It did not say that racial disparities were imaginary. It did not say that policing was free of prejudice. It separated different forms of force and asked whether the same explanation accounted for all of them. A society that wants to understand policing should welcome that kind of separation. Slapping, pushing, restraining and shooting are not interchangeable events. They involve different decisions, different levels of threat and different evidential problems.

But the lethal-force finding collided with the public narrative established by a succession of heavily publicised shootings. Fryer later described the experience as a moment when “people lose their minds when they don’t like the result”. He said colleagues warned him not to publish and that doing so could ruin his career. After the paper appeared, he reported receiving threats and needing armed protection for more than a month.

The threats were indefensible. So was any attempt to intimidate a researcher into changing or concealing a result. Yet the existence of threats in either direction cannot settle the scholarly question. Fryer’s work was a National Bureau of Economic Research working paper, not a final verdict on every police shooting in America. The NBER itself identifies it as a working paper, and the study’s conclusions were limited by the jurisdictions and data available to the researchers. A serious response was to challenge the sample, the controls, the definition of a police encounter and the limits of the inference. A serious response was not to insist that the offending result should be kept from the public.

That is the difference between criticism and capture. Science advances by trying to break a result. Capture begins when the result must first pass a political loyalty test before anyone is permitted to inspect it.

Fryer’s case also exposes a common confusion about representation. People often assume that a researcher from a historically disadvantaged group will produce the politically expected conclusion, or that an unexpected conclusion is evidence of betrayal. That is not science. A Black scholar does not owe a movement a particular finding. Nor does a white scholar become neutral by disagreeing with a racial-justice claim. The researcher’s identity can be relevant to the questions asked and the experiences brought to a project, but it cannot substitute for study design.

The episode should therefore trouble both sides. Those who used Fryer’s result as proof that racial bias in policing is a myth were asking one paper to do more than it could. Those who wished to bury the lethal-force finding were asking a researcher to protect a narrative rather than report an analysis. The correct lesson is narrower and more demanding: a politically uncomfortable result must receive the same opportunity for scrutiny as a politically useful one.

When public health advice acquired a moral category

The Covid era supplied a more visible version of the same problem. In May 2020, more than 1,300 public-health officials and health professionals signed a letter supporting protests following the killing of George Floyd. The letter described anti-racist protest as a public-health imperative, while many public-health voices had previously condemned anti-lockdown demonstrations as dangerous mass gatherings. The New York Times reported the tension in unusually direct terms: was public-health advice dependent on whether experts approved of the cause represented by the crowd?²

There were real differences between events. Outdoor transmission risk was not identical in every setting. Crowds varied in size, density, duration and ventilation. Mask use differed. People travelled different distances. Participants’ behaviour, the timing of the events, and the prevalence of the virus also mattered. It would be foolish to claim that every gathering had precisely the same epidemiological consequences.(Claudine Gay, Harvard president, would say context is important!

But nuance cannot become a licence to deny an obvious symmetry. If a mass gathering is dangerous because people stand close together, shout, travel and return to their communities, those mechanisms do not disappear when the gathering carries a morally approved banner. They may be outweighed by another social concern; an expert may decide that a public-health risk is justified by the urgency of confronting racial violence. That is a political and ethical judgement. It should be presented as such, not smuggled into settled epidemiology.

The distinction is important because public health depends on trust. People can accept a recommendation that imposes a cost when officials explain the evidence, acknowledge uncertainty and admit the competing values. They become suspicious when the same conduct is condemned as reckless in one political setting and celebrated as protective in another. The problem is not that experts have moral commitments. It is that they sometimes present those commitments as if they were laboratory measurements.

The episode also shows how institutions protect themselves after a contradiction becomes visible. They add qualifications. They explain that one crowd resisted public-health measures while another protested a more serious threat. They say the groups were not morally equivalent. Some of those explanations may be persuasive. Yet morality is not a substitute for transmission analysis. The relevant question is whether the public was told clearly that the advice involved a trade-off rather than a simple reading of the science.

This is where Prasad’s argument becomes broader than a complaint about “wokeness”. Any public-health authority can be tempted to use science as a credential for a policy it has already chosen. Once that happens, contrary evidence is not treated as information but as a threat to compliance. The institution then faces a choice: admit that it made a judgement under uncertainty, or defend the original claim with increasingly elaborate language. The latter course protects prestige in the short term and damages credibility in the long term.

The Cass Review exposed the cost of confident medicine

Gender medicine provides another test because it combines vulnerable patients, disputed evidence and strong moral pressure. The Cass Review, an independent review commissioned by NHS England, examined services for children and young people questioning their gender or experiencing gender incongruence. Its final report warned that the evidence base for several interventions was weak, that the clinical picture was more complex than a single pathway could accommodate and that research had not kept pace with the rapid expansion of demand.³

The review did not say that every young person seeking help was confused, that every clinician acted in bad faith or that no form of medical intervention could ever be justified. It called for a more cautious model: fuller assessment, attention to mental health and neurodevelopment, better data, stronger research and services capable of dealing with uncertainty. Its significance lies partly in how ordinary those recommendations would sound in almost any other area of medicine.

Medicine normally treats uncertainty as a reason to improve the evidence. It does not mean that doctors must refuse all treatment until every question has been answered. It means that the strength of the recommendation should reflect the strength of the evidence, that patients should understand what is known and unknown, and that researchers should not confuse the absence of evidence with evidence of benefit.

In this field, however, caution was often treated as hostility. The dispute was framed as a moral contest between people who affirmed patients and people who wished to deny their existence. That framing left little room for the position the Cass Review eventually occupied: a patient can deserve dignity, protection from discrimination and careful medical attention at the same time. A clinician can oppose cruelty without claiming that every proposed intervention has been adequately tested.

Prasad’s criticism is strongest when it targets overconfidence rather than any particular political conclusion. A study that reports improvement after treatment may be valuable without proving that the treatment caused the improvement. A cohort of patients who remain in care may differ from those who leave. A comparison group may be poorly matched. Outcomes such as distress, self-harm, suicide and social functioning are affected by many forces beyond a single medical intervention. If researchers and professional bodies speak with certainty where the design supports only a qualified association, they use the authority of medicine to cover an evidential gap.

The opposite danger also exists. Critics may select the weakest studies in a field and imply that no patient can benefit from any intervention. That is not a serious use of evidence either. The Cass Review matters because it did not reduce the dispute to a slogan. It exposed a system in which clinicians had been asked to make decisions without the kind of long-term, comparative evidence that patients and families reasonably expected.

The lesson extends beyond gender care. A field becomes vulnerable when political urgency accelerates treatment, professional organisations repeat claims before they are secure and dissent is confused with harm. The language of compassion can then be used to avoid the hard work of measuring outcomes. Compassion is necessary in medicine. It is not a method.

Race, biology and the politics of categories

Prasad also points to a contradiction in the way universities and medical researchers speak about race. In one context, race is described as a social construct with no simple biological essence. In another, research institutions insist that genomic studies need participants from varied racial and ethnic backgrounds because samples from one population do not capture human genetic variation.

There is no logical contradiction in saying that racial categories are historically shaped and still useful, imperfect proxies for ancestry, environment, social treatment or genetic variation. The trouble begins when institutions slide between meanings without admitting the change. “Race” can mean a social classification in one sentence, a demographic label in the next and a biological proxy in a third. A researcher who points out the ambiguity is then accused of attacking the political purpose of the research.

The issue is not a minor dispute over vocabulary. Categories determine who is counted, what comparisons are made and which explanation appears plausible. If a health gap is described as evidence of structural discrimination, the claim may be correct, partly correct or false depending on how the groups were formed and what variables were measured. If the same gap is attributed to biology, the claim faces a different set of tests. Neither explanation can be established by selecting the category that produces the preferred moral conclusion.

Language policing becomes especially damaging when it enters peer review. A journal may demand one term because it signals the correct political orientation, even though another term is more precise in the study’s context. Readers then mistake the use of a preferred label for methodological competence. A paper can be praised for its vocabulary while its causal model remains weak. Another can be rejected for using an unfashionable expression even when its data are stronger.

Careful terminology is legitimate. Words can exclude patients, confuse participants or imply claims a study does not support. Researchers should explain their definitions and use language that matches the population being studied. But precision cannot mean obedience to an ideological glossary. A scientific term should be judged by whether it clarifies the question, identifies the population and permits the work to be replicated.

The public is entitled to ask which meaning is being used. Are researchers studying ancestry, self-identified race, perceived race, institutional treatment or a mixture of all four? Is the category stable across time and place? Does it predict the outcome because it measures a biological pathway, exposure to discrimination, access to care or something else? Those questions do not deny racism or celebrate biology. They are the minimum conditions for knowing what a result means.

In a healthy research culture, the answer to a difficult question is better measurement. In a captured culture, the question itself is treated as suspect if it might unsettle the preferred explanation. (Example: Nathan Cofnas seems to be suffering for not being afraid of the consequences of disclosing bad news about academic darling Professor Jason Arday).

The newborn-mortality paper and the seduction of a moral result

Few findings illustrate the power of a compelling narrative more clearly than the 2020 study of physician-patient racial concordance and newborn mortality. Researchers analysed 1.8 million hospital births in Florida between 1992 and 2015 and reported that Black newborns had better outcomes when the attending physician was also Black, with the apparent benefit concentrated in more difficult births and hospitals that treated more Black babies.⁴

The result was immediately legible. Black newborns face a higher risk of death than white newborns in the United States; a study suggesting that the racial identity of the attending doctor could materially narrow that gap offered a powerful story about bias, communication and representation. Headlines translated an association into a more direct claim: Black doctors save Black babies.

That translation is where journalism and research can part ways. The paper’s statistical association may matter, but it does not, by itself, prove that a doctor’s race caused the difference. Doctors are not assigned randomly to newborns. Hospitals differ. Patients differ. Births differ. Staffing patterns may relate to the seriousness of cases, clinicians’ experience, institutional resources, and the likelihood that a Black physician works in a hospital serving a particular population.

The authors were more careful than some of the coverage. Their findings suggested a relationship and offered possible explanations; they did not establish that race matching was a universal treatment or that white doctors were inherently less capable. An accompanying PNAS discussion later examined the claim that Black doctors save more Black babies and questioned how the evidence should be interpreted. That debate does not erase the original finding. It shows why a morally satisfying result must remain open to reanalysis.⁵

The article’s larger lesson is uncomfortable. A paper can be methodologically cautious and still be presented as a political fact because its conclusion fits an accepted account. The words “associated with” become “caused by”. A finding about a particular population becomes a general theory of medicine. A call to recruit more Black doctors, which may be justified on professional and representational grounds, is presented as if one study had settled the mechanism of neonatal mortality (Never forget correlation is not causation).

That does not mean representation is irrelevant. Communication, trust, prior experience and institutional culture can affect care. Medical work is not performed by disembodied algorithms. But a serious argument for a more representative workforce should not depend on exaggerating a single study. If the evidence is real, it can survive accurate description. If it cannot survive accurate description, the public is being recruited to defend a narrative rather than understand a result.

The same rule should apply if a study finds no benefit from racial concordance, or if it finds a benefit for one group and not another. Researchers should not be required to manufacture symmetry, and critics should not be permitted to dismiss an inconvenient result merely because it complicates an equality campaign. The honest question is what the data show, how certain that answer is and what further design would test the proposed mechanism.

When scientific journals cross into political advocacy

A research journal’s authority rests on a delicate separation. Its editors may have political opinions, but readers expect research findings to be assessed by methods rather than party. That separation became harder to defend when Nature endorsed Joe Biden in the 2020 US presidential election and explicitly tied the choice to truth, evidence, science and democracy.

Nature defended the decision as a response to a candidate and administration it believed had damaged scientific institutions. But a later randomised experiment involving 4,260 US adults found that exposing Trump supporters to the endorsement reduced their view of Nature as an informed and impartial source on science-related questions. It also reduced their willingness to trust information from the journal about Covid vaccines and increased reported mistrust in US scientists. The effect among Biden supporters was positive but smaller.⁶

The result should not be misread as proof that Nature should have endorsed Trump, or that political neutrality requires silence in the face of threats to science. The journal’s editorial board was entitled to make a political argument. The experiment did, however, identify the price of doing so. A publication that presents itself as a trusted scientific intermediary may lose part of its audience when it becomes a partisan actor.

There is a difference between reporting on a government’s treatment of science and endorsing a candidate. A journal can investigate research cuts, regulatory interference, climate policy or the misuse of evidence. It can publish criticism from scientists. It can editorialise against a policy. An electoral endorsement goes further: it asks readers to treat the publication’s political judgement as a reason to choose one person over another.

That choice can be justified in exceptional circumstances. But people who already agree with the choice often declare exceptional circumstances. If every election is framed as a final battle between reason and its enemies, scientific institutions will become permanent political organisations. Their opponents will stop hearing evidence from them, while their supporters will start treating agreement with the journal as evidence of scientific seriousness.

The consequences are visible in public-health communication. People do not receive evidence in a vacuum. They decide whether the messenger is honest, competent and willing to acknowledge inconvenient facts. A journal that tells one political group that it is morally aligned with science may gain applause from its allies and lose access to the people whose trust matters most. It may believe that the loss is deserved. It should not pretend that no loss occurred.

Scientific authority is not strengthened by pretending that political commitments do not exist. It is strengthened by marking the boundary: here is the research; here is the editorial judgement; here are the uncertainties; here is the argument that goes beyond the data. The more a journal blurs those categories, the easier it becomes for critics to say that the research itself is political.

The Scott Atlas censure turned disagreement into institutional discipline

Stanford’s treatment of Scott Atlas shows what happens when an institution decides that an individual’s public statements have crossed from error into misconduct. In November 2020, the Stanford Faculty Senate approved a resolution condemning Atlas, a Hoover Institution fellow who had served as a special adviser to President Donald Trump on Covid policy. The resolution passed with the support of 85 per cent of the senate membership and accused Atlas of promoting views that contradicted medical science, discouraging protective measures and damaging Stanford’s reputation.⁷

The context was severe. COVID cases were rising. Hospitals feared being overwhelmed. Faculty members believed that a prominent adviser was using his institutional affiliation and public position to lend authority to claims they considered dangerous. Their anger was not invented, and the public stakes were not trivial.

But emergency is precisely when due process matters most. Four years later, the senate debated whether to rescind its censure. The Planning and Policy Board found that Atlas had not been afforded due process and that the senate’s governing documents did not clearly establish its authority to issue the condemnation. The senate nevertheless voted 21 to 13, by secret ballot, against rescission.⁸

The arguments on both sides were revealing. Critics of rescission said reversing the censure could validate Atlas’s views and minimise the danger of his conduct during a health crisis. Supporters said the original action had sacrificed academic freedom and created a precedent in which a faculty body could punish speech without fair process. Jonathan Berk, a Stanford finance professor, put the principle bluntly: “Good, happy speech needs no protection. What needs protection is unpleasant speech.”

That sentence does not establish that Atlas was right. It identifies the institutional question. Universities do not protect academic freedom by defending only correct opinions. They protect it by creating procedures that can distinguish bad scholarship, professional misconduct, and unpopular speech. If a university can condemn a scholar whenever a majority believes his views are harmful, then academic freedom exists at the pleasure of the majority.

There is also an evidential problem. Scientific disagreement should normally be answered with better evidence, not a reputational sanction. A university may discipline fraud, fabrication, harassment or a failure to meet professional duties. It may not convert every disputed policy view into misconduct without explaining the rule, the evidence and the process. Otherwise, the institution teaches junior scholars that the safe route is not to be accurate but to be aligned.

Stanford’s later debate did not resolve whether every Atlas claim was scientifically defensible. It resolved something more limited and more important: the university acknowledged that the original process raised questions about authority and fairness, then chose to leave the censure in place. That is not proof of scientific capture. It is evidence that institutional confidence in its own political judgement can survive a finding that its procedure was defective.

The distinction should concern anyone who believes in evidence-based policy. If a claim is false, show why. If it is dangerous, explain the mechanism of harm. If a speaker abuses office, identify the duty breached. Punishment should not substitute for refutation.

Masks showed why uncertainty must be spoken aloud

Mask policy became another battle over whether evidence or symbolism should guide public health. The question was not whether masks could ever reduce transmission under any conditions. Fit, material, setting, duration, adherence and the virus itself all matter. The question was how confident officials could be that community mask mandates produced the promised reduction, especially when the available studies were mixed, and compliance was uneven.

A Cochrane review of physical interventions to interrupt the spread of acute respiratory viruses found uncertainty around the effectiveness of masks. The review reported no clear reduction in respiratory viral infection with medical or surgical masks in pooled randomised trials and no clear superiority of medical masks over N95 or P2 respirators for routine care among healthcare workers, while stressing limitations in the evidence and the need for better studies.⁹

That finding does not translate into “masks never work”. It means that policy claims must specify which masks, in what setting, for whom, with what adherence and against which outcome. A well-fitted respirator used correctly in a high-risk clinical setting is not the same intervention as a loose cloth covering worn intermittently in a crowded public place. Treating them as one category is a failure of analysis.

Yet public discussion often rewarded certainty. A mask became a sign of virtue or irresponsibility before it became a measure of effectiveness. Officials who had first presented masks as one tool among several sometimes spoke as if the tool carried an almost magical protective power. Critics who pointed to weak evidence were accused of wishing to infect vulnerable people. The argument moved from “this may reduce risk” to “opposition to this policy is immoral”.

That move has a predictable effect. When later evidence qualifies the original confidence, citizens do not hear a normal scientific update. They hear an admission that the moral campaign was built on a falsehood. The result is not greater compliance next time. It is scepticism toward the next recommendation, including recommendations that rest on stronger evidence.

Good public health communication can survive uncertainty. It can say that a measure is cheap, low-risk and biologically plausible, while admitting that randomised evidence is limited. It can explain why officials choose precaution in a crisis and what result would lead them to revise the policy. Such honesty may make the message less rhetorically powerful. It also makes it more durable.

The same principle applies to claims about school masking for diseases that spread through different routes, vaccine schedules, raw milk and public funding for new vaccine technologies. Prasad’s catalogue includes these disputes, but the underlying issue is not that every position he takes is correct. It is whether institutions have made room for proportionality. A precaution can be sensible without being a proven cure. A policy can be justified under uncertainty without being described as settled science.

The data do not speak without a system around them

The phrase “the data speak for themselves” is attractive because it suggests an escape from politics. Data do not speak by themselves. Someone decides what to count, how to define an event, which comparison group to use, what variables to control for and which outcome matters. Someone writes the paper, chooses the headline and decides whether an association is worth treating as causal.

That does not make data meaningless. It makes method indispensable. A political movement can choose a question for political reasons and still answer it well. A conservative researcher can produce rigorous work. A progressive researcher can produce rigorous work. The test is not whether the question arose from a political concern; it is whether the study exposes its assumptions and allows others to challenge the result.

Several safeguards follow from that principle. Researchers should pre-specify hypotheses where practical, report null results, make code and data available when confidentiality permits, distinguish exploratory from confirmatory analysis and disclose limitations without hiding them in a footnote. Journals should publish serious methodological criticism, not just commentary that agrees with the result. Universities should protect researchers from threats while refusing to treat criticism as harassment by default. Editors should ask whether they would apply the same standard if the conclusion pointed the other way.

Readers also have a role. A dramatic percentage is not an explanation. “Doubles the risk” may describe a change from one death in ten thousand to two in ten thousand, or from 10 to 20 per cent. “Associated with” is not “caused”. “No evidence of an effect” is not proof that an effect does not exist. A study can be important while remaining provisional.

These distinctions are not academic niceties. They determine whether a hospital changes treatment, whether a government restricts liberty, whether a university disciplines a scholar and whether a public trusts the next warning. When a claim is moralised before it is measured, every correction becomes a political defeat. When the original claim is reported with the proper limits, revision looks like science doing its job.

That is why Fryer’s story still matters. The paper did not settle the entire question of police shootings. The newborn-mortality study did not prove that racial matching is a universal medical intervention. The Cass Review did not answer every question about gender care. The mask evidence did not justify every policy or condemn every use. Nature’s experiment did not prove that journals must never endorse candidates. Stanford’s Atlas dispute did not establish which pandemic policies were correct.

These cases show the price of treating provisional findings as moral property. Once a result becomes a badge of group membership, people defend its weaknesses as attacks on the group. Once an institution links its authority to a political cause, criticism of the cause is heard as hostility to science. Once a researcher learns that an inconvenient conclusion carries a professional penalty, silence becomes a rational career strategy.

The price of getting the story right

Pinker’s confidence that science is less captured than the humanities may still be defensible if “captured” means that every laboratory and medical department has adopted one uniform ideology. Universities remain full of disagreement, and much important research is conducted by people who do not share the same politics. The stronger claim—that science is protected from the pressures shaping other parts of academic life—is much harder to maintain.

Scientific fields are not protected by the presence of equations, microscopes or prestigious journals. They are protected by habits. Researchers must be able to publish a result that disappoints their colleagues. Patients must be told when evidence is weak. Editors must distinguish an unpopular argument from a dishonest one. Universities must separate speech from misconduct. Public-health officials must say when they are making a moral judgement under uncertainty rather than reporting a settled measurement.

Those habits are fragile because the incentives run the other way. A result that confirms the prevailing story attracts attention, grants and institutional praise. A result that complicates it attracts suspicion. A scholar who uses the approved language is assumed to understand the subject. A scholar who challenges the language is required to prove that the challenge itself is safe. The resulting literature may look impressive while becoming less reliable at the edges where politics is most intense.

There is an irony here. Progressive institutions often claim to distrust neutrality because they understand that categories and institutions can reflect power. That insight is sometimes valuable. But rejecting neutrality does not justify replacing evidence with allegiance. It makes impartial procedure more important, not less. If researchers know that no question is free from social context, they should be more transparent about definitions, assumptions and trade-offs.

Nor should critics of progressive orthodoxies imagine that they are automatically on the side of science. A sceptical posture can become its own performance. Invoking “data” does not excuse selective citation, crude extrapolation or contempt for evidence that points the other way. The proper response to ideological pressure is not a new orthodoxy. It is a standard that survives contact with unwelcome results.

That standard is demanding because it denies everyone the comfort of a guaranteed conclusion. It says that a study can expose a real disparity without proving its cause, that a treatment can help some patients without being sufficiently tested for broad use, that a gathering can serve a just cause and still carry infection risk, and that an institution can defend science while damaging its credibility through partisan advocacy. It allows facts to remain inconvenient after they become politically useful.

The question Prasad poses to Pinker is therefore not whether universities contain good scientists. They plainly do. It is whether those scientists are free to follow evidence when the answer threatens the institution’s moral narrative. The answer isn’t in mission statements or diversity declarations. You can find it only by watching what happens to the researcher who publishes the wrong result.

1. Roland G. Fryer Jr., “An Empirical Analysis of Racial Differences in Police Use of Force,” National Bureau of Economic Research Working Paper 22399. https://www.nber.org/papers/w22399 ↩
2. Sheri Fink and Mike Baker, “Are Protests Dangerous? What Experts Say May Depend on Who’s Protesting What,” The New York Times, 6 July 2020. https://www.nytimes.com/2020/07/06/us/Epidemiologists-coronavirus-protests-quarantine.html ↩
3. Hillary Cass, Independent Review of Gender Identity Services for Children and Young People: Final Report, April 2024. https://cass.independent-review.uk/wp-content/uploads/2024/04/CassReview_Final.pdf ↩
4. Brad N. Greenwood et al., “Physician-patient racial concordance and disparities in birthing mortality for newborns,” Proceedings of the National Academy of Sciences, 2020. https://pubmed.ncbi.nlm.nih.gov/32817561 ↩
5. George J. Borjas and Robert VerBruggen, “Physician–physician racial concordance and newborn mortality,” Proceedings of the National Academy of Sciences, 2024. https://www.pnas.org/doi/10.1073/pnas.2409264121 ↩
6. Floyd Zhang, “Political endorsement by Nature and trust in scientific expertise during COVID-19,” Nature Human Behaviour, 2023, discussed by Nature. https://www.nature.com/articles/d41586-023-00789-5 ↩
7. “Faculty Senate condemns COVID-19 actions of Hoover’s Scott Atlas,” Stanford Report, November 2020. https://news.stanford.edu/stories/2020/11/faculty-senate-condemns-actions-hoovers-scott-atlas ↩
8. “Faculty Senate upholds 2020 Scott Atlas censure, discusses school initiatives,” Stanford Report, 21 November 2024. https://news.stanford.edu/stories/2024/11/faculty-senate-upholds-2020-scott-atlas-censure-discusses-school-initiatives ↩
9. Tom Jefferson et al., “Physical interventions to interrupt or reduce the spread of respiratory viruses,” Cochrane Database of Systematic Reviews. https://www.cochranelibrary.com/cdsr/doi/10.1002/14651858.CD006207.pub6/full ↩