The Mirror
There is a particular kind of conversation that repeats itself across medicine, public health, climate science, and economics. An expert, frustrated by public resistance to evidence, concludes that the problem is with the public — that people are too swayed by emotion to engage with facts, too attached to their existing beliefs to update them when presented with evidence. The solution, therefore, is better communication: clearer messaging, more accessible science, improved public understanding.
This diagnosis is not entirely wrong. But it has a blind spot so large that it undermines itself.
The same cognitive machinery that produces irrational belief in the general public operates inside expert communities. Education, credentials, and institutional affiliation do not neutralise it. In some respects, they amplify it. Understanding why this is true — and what it means for how we think about knowledge, expertise, and disagreement — is more useful than any number of campaigns to close the public's alleged knowledge deficit.
The Case Against Pure Rationalism
Research on human judgment over the past fifty years has established something uncomfortable for anyone who believes that reasoning is our primary decision-making tool.
Psychologists Daniel Kahneman and Amos Tversky spent decades documenting systematic, predictable errors in human judgment — not in uneducated populations but in trained statisticians, physicians, and economists. Their core finding was that fast, automatic, emotionally-inflected thinking operates as the default mode for most judgments, and that slow, deliberative reasoning is cognitively expensive, rarely engaged, and frequently recruited after a judgment has already been made — to justify it rather than to generate it.
Paul Slovic and colleagues demonstrated the affect heuristic: people consult an internal emotional tag attached to a concept — positive or negative, safe or dangerous — and use that tag as the primary input to risk judgments. Objective data about the same risk gets filtered through that emotional prior, not the other way around.
Social psychologist Jonathan Haidt pushed this further. His research on moral judgment showed that people typically reach conclusions first and construct reasons afterward. When their post-hoc reasons were systematically dismantled, they abandoned the reasons but kept the original verdict. The reasoning, in other words, was decorative.
None of this research was conducted on an unintelligent or uneducated sample. It describes baseline human cognition. The question is whether expertise provides immunity — and the evidence says it does not. The picture is more troubling still.
The Expert Who Knows More, Believes More Strongly
Expertise does not merely fail to neutralise identity-driven reasoning. Under certain conditions, it makes the problem worse.
Dan Kahan and colleagues at Yale Law School conducted a series of studies examining how scientific literacy affects attitudes on contested empirical questions — climate change, nuclear power, gun control. The intuitive prediction is that higher scientific literacy would reduce polarization: people who understand the evidence better would converge on the evidence.
The data showed the opposite. Higher scientific literacy and numeracy were associated with greater polarization, not less. The mechanism Kahan identified is identity-protective cognition — the tendency to evaluate evidence not on its merits but on whether accepting it would cost you standing with the people who matter to you. When a factual claim becomes aligned with a particular social or political group, accepting findings from the other side carries real social costs; it marks you as disloyal, credulous, or naive in the eyes of your own group. Skilled reasoners are better equipped to find counter-arguments, identify ambiguities, and selectively interpret data — and they deploy those skills in defence of positions their group already holds (Taber and Lodge, American Journal of Political Science, 2006).
The problem with information provision is therefore not just that it is insufficient. Under certain conditions, providing more information to a resistant audience makes the resistance stronger, because capable reasoners are better at motivated skepticism.
Expert communities are not exempt. This is particularly visible in academic and research institutions, where careers, funding lines, and professional reputation become tied to particular theoretical commitments over decades — making the cost of public dissent from established positions concrete and career-affecting, independent of what the evidence says.
The Replication Crisis as Evidence
Between roughly 2011 and 2020, the scientific community confronted evidence that a substantial proportion of published research findings in certain fields could not be reproduced. The fields where this problem is sharpest are not physics or structural engineering. They are social psychology, behavioural economics, nutrition science, and health behaviour research — precisely the fields whose findings most directly inform public health campaigns, government policy, and the kind of expert-to-public communication this article is about.
The Open Science Collaboration, publishing in Science in 2015, attempted to replicate 100 studies from leading psychology journals. Approximately 36 to 39 percent produced results consistent with the original findings, depending on the criterion applied. Effect sizes in replications were on average half those in original studies.
The pattern appeared elsewhere. The "backfire effect" — the claim that factual corrections increase belief in misinformation — was widely cited in public discourse as evidence of deep irrationality in how people process information. Wood and Porter, publishing a large-scale replication attempt across 52 issues in the American Journal of Political Science (2019), found no consistent backfire: corrections generally produced mild updating in the correct direction or no change at all. The effect as originally formulated appears to have been an artifact of small samples and specific experimental conditions, not a general feature of human cognition.
Ego depletion — the idea that willpower is a finite resource that runs out with use, a foundational concept in self-control research — showed inconsistent results. Power posing, the claim that adopting expansive physical postures raises hormone levels and confidence, failed similarly. Social priming, a body of findings suggesting that subtle environmental cues can unconsciously shape behaviour, and several findings in implicit bias research — studies measuring the unconscious associations people hold toward different social groups — followed the same trajectory.
John Ioannidis, in a 2005 paper in PLOS Medicine titled "Why Most Published Research Findings Are False," constructed a formal probabilistic model — not an empirical audit of a specific literature, but a mathematical argument — showing that the publishing system would systematically produce an overrepresentation of false positives. He identified four structural reasons: studies too small to reliably detect genuine effects; analytical flexibility, meaning the freedom researchers have to adjust their methods after seeing the data; publication bias, journals' preference for surprising positive findings over negative ones; and the simple fact that most hypotheses, when tested, turn out to be wrong. The model's assumptions have since found empirical support through the replication work described above.
The causes are not primarily about individual dishonesty. Researchers face career incentives — publications, grants, prestige — that reward novel positive findings and provide little return for replications or null results. Journals preferentially publish surprising results. These pressures operate regardless of individual integrity. Honest, conscientious researchers working within a system that rewards certain outputs will collectively produce a literature skewed toward those outputs.
This critique did not originate outside science. It emerged from within, and the response — pre-registration of studies, registered reports, open data requirements, replication initiatives — represents genuine effort at reform. Whether that effort has materially changed the underlying incentive problem remains an open question.
What Expert Communities Share With Everyone Else
Expert communities, like all human groups, develop consensus positions that then acquire social functions beyond their purely intellectual content. Challenging the consensus carries professional costs. Supporting it carries professional benefits. This is not corruption — it is ordinary group dynamics operating on a population of highly educated people who are not exempt from ordinary group dynamics.
Thomas Kuhn observed in The Structure of Scientific Revolutions (1962) that normal science operates within a paradigm, and paradigms are not primarily overthrown by evidence. They shift when accumulated anomalies become impossible to ignore and when a new generation of researchers arrives without the same investment in the old framework. Kuhn's full account has been contested — Laudan's Progress and Its Problems (1977) argued that scientific change is more continuous and evidence-responsive than Kuhn implied — but the sociological observation at the core survives the philosophical debate: community membership shapes what counts as acceptable evidence, and abandoning a framework is never a purely intellectual act.
Philip Tetlock's work on expert prediction (Expert Political Judgment, 2005) showed that domain experts performed modestly better than chance in their own fields, and that the most confident, media-prominent experts were among the least accurate. Experts who held a single strong explanatory framework performed worse than those who integrated multiple frameworks with explicit uncertainty. Tetlock's subsequent work is equally important and often left out of this discussion: the Good Judgment Project, documented in Superforecasting (2015), showed that structured training in probabilistic reasoning does produce meaningfully more accurate forecasters. The expert identity — confident, authoritative, committed to a framework — actively works against the specific habits that accurate forecasting requires.
Expert knowledge does outperform lay intuition — the evidence on medical diagnosis, engineering risk assessment, and epidemiological modelling is clear on this. But that performance advantage depends on the corrective infrastructure being present: pre-registration, open data, replication, honest uncertainty communication. Strip those away, or allow the incentives to work against them, and the gap closes in ways the replication literature has since documented. The relevant question is which conditions produce that performance gap — and whether those conditions are actually present in the field, institution, or study under discussion.
The Public's Resistance Is Sometimes Correct
This is the dimension that rarely surfaces in expert-led conversations about public distrust.
If a meaningful fraction of published research findings in the relevant fields are false positives, and if expert consensus can be shaped by career incentives rather than purely by evidence, then some public skepticism reflects a reasonable prior rather than ignorance. The skepticism may be poorly articulated. It may be directed at wrong targets. It gets mixed up with genuine misinformation. But the underlying intuition — that authoritative claims sometimes fail — is empirically supported.
Consider the pattern across the documented record.
The Tuskegee syphilis study, in which US public health authorities withheld treatment from Black men for decades while studying disease progression, is the canonical case for why distrust in medical institutions can carry a rational basis in some communities. This was not error. It was authorised, systematised betrayal by the very institutions responsible for care.
In Europe, the thalidomide disaster of the late 1950s and early 1960s offers a different failure: Grünenthal's widely marketed sedative, whose use during pregnancy caused severe birth defects, was approved and monitored by regulatory bodies across dozens of countries before action was taken. The manufacturer was culpable, but so were the systems supposed to catch exactly this kind of harm before it reached patients.
The decades-long suppression of tobacco industry research on smoking harms added a third pattern — deliberate manipulation rather than error. Internal documents showed company scientists knew the risks for years while public-facing statements denied them (as tobacco company records released under the 1998 Master Settlement Agreement later confirmed).
The successive revisions to dietary fat guidelines — including the removal of dietary cholesterol limits from the 2015 US Dietary Guidelines and parallel revisions in European nutritional guidance, after systematic reviews found the original evidence base weaker than decades of official advice had implied — completed the picture differently: not manipulation, but collective overconfidence, experts certain beyond what the evidence warranted.
A person who knows this history and applies elevated skepticism to new expert claims is not behaving irrationally. They are updating on a realistic base rate.
Expert consensus is one important input, not a verdict. Its reliability varies by field and by whether the conditions that make expertise trustworthy — transparency, replication, conflict-of-interest disclosure, honest uncertainty quantification — are actually present. Proportionate scrutiny scales with the stakes of the decision and with what is known about how the underlying research was produced.
What Intellectual Honesty Requires in Practice
Intellectual honesty has largely been domesticated into a rhetorical gesture — "acknowledge that you might be wrong" — inserted before asserting the same position with equal confidence. That is not what the evidence demands.
For anyone outside research institutions, the reforms below matter for a simple reason: they are what separates a scientific literature you can reasonably trust from one that systematically overstates its own certainty.
The most direct corrective is at the level of study design. Pre-registration — committing publicly to what you are testing and how, before collecting any data — closes the gap between hypothesis-testing and hypothesis-generating research, making it harder to present exploratory findings as confirmatory ones. Open data requirements allow independent verification, and the resistance to those requirements within some research communities is, in itself, informative. A literature where null results — studies that found no effect — disappear into desk drawers produces a systematically distorted picture of what the evidence actually shows. Publishing negative findings is not a consolation prize for failed research. It is part of the evidentiary record.
Reporting standards matter equally. A finding can clear the p < 0.05 threshold — the conventional standard meaning there is less than a five percent chance the result occurred randomly — while having an effect so small it carries no meaningful clinical or practical consequence. Effect sizes measure the actual magnitude of a finding, as distinct from its statistical likelihood. Reporting both is required for honest communication of what a study actually established rather than what it technically demonstrated. The cultural norm in several fields that treats replication attempts as attacks on original researchers directly undermines the self-correcting process that gives science its authority — replication is not criticism, it is the method.
Public communication of uncertainty is where these technical commitments meet the trust problem. Slovic's research on trust asymmetry (Risk Analysis, 1993) established that trust is lost faster than it is built — negative events are more salient, more credible when reported by independent sources, and carry disproportionate weight in how institutions are subsequently evaluated. An institution that oversells certainty and is later proven wrong loses more than it gained from the initial projection of confidence. Lewandowsky et al. (2012) and Ecker et al. (2022) confirm that misinformation and overclaimed certainty leave residues that corrections do not reliably clear. Honesty about limits is not just the principled position — it is the most reliable strategy for building trust capable of surviving the inevitable revisions that science produces. Whether institutions structured around projecting authority will accept that trade-off is a different question from whether they should.
The Mirror
The expert standing in front of a resistant public is looking at a problem that has a mirror.
The research documenting affect-driven reasoning, identity-protective cognition, and resistance to disconfirming information in the general public is well-grounded. The parallel research on expert communities is equally so. Where they diverge is not in the cognitive machinery but in whether the practices surrounding knowledge production correct for that machinery or quietly accommodate it.
Science, at its best, is a set of methods designed to work against human cognitive defaults — to make it difficult to confirm what you already believe and easy for others to check your work. The replication crisis is evidence that those methods have been inconsistently applied and that the incentive structures have worked against them in the fields that matter most for public communication.
Expertise is not broken. Some questions are genuinely settled. Some consensus is robust, well-replicated, and independent of the problems described above. What is broken, or at least bent, is the convention of presenting expert consensus as more certain than the conditions of its production warrant — and that convention is what produces the trust deficit that better messaging cannot fix.
For a reader outside those institutions, this knowledge is most useful not as a reason for cynicism but as a basis for calibration — asking not just what the evidence shows, but who produced it, under what incentives, and whether it has survived independent scrutiny.
The public, when it senses that certainty is being performed rather than earned, is often picking up on something real. Meeting that honestly requires something harder than better messaging — the kind of transparency about uncertainty that expertise, when it is working properly, actually makes possible.
Whether expert institutions will choose that path before the alternative becomes unavoidable is not a question the evidence answers.
Primary sources: Kahneman (2011, Thinking Fast and Slow), Slovic, Finucane, Peters and MacGregor (2002, Journal of Behavioral Decision Making), Slovic (1993, Risk Analysis), Haidt (2001, Psychological Review), Kahan, Peters, Wittlin et al. (2012, Nature Climate Change), Taber and Lodge (2006, American Journal of Political Science), Open Science Collaboration (2015, Science), Wood and Porter (2019, American Journal of Political Science), Ioannidis (2005, PLOS Medicine), Tetlock (2005, Expert Political Judgment; 2015, Superforecasting), Kuhn (1962, The Structure of Scientific Revolutions), Laudan (1977, Progress and Its Problems), Lewandowsky, Ecker, Seifert, Schwarz and Cook (2012, Psychological Science in the Public Interest), Ecker, Lewandowsky et al. (2022, Nature Reviews Psychology).
