In 1995, a man named McArthur Wheeler robbed two Pittsburgh banks in broad daylight without any disguise. When police showed him the surveillance footage after his arrest, he was visibly baffled. He had rubbed lemon juice on his face before the robberies, operating on the belief that lemon juice — which can be used as invisible ink — would make his face invisible to cameras. No one had told him this. He had concluded it himself, with complete confidence.
This anecdote provided the opening to one of the most widely cited papers in social psychology: David Dunning and Justin Kruger's 1999 study "Unskilled and Unaware of It," which documented the systematic tendency of people with low competence in a domain to overestimate their ability in that domain. The Dunning-Kruger effect, as it came to be called, entered popular culture with remarkable speed and has since been deployed in countless arguments about why people believe false things, why poor performers are oblivious to their shortcomings, and why social media seems populated by confident ignorance.
The popular understanding of Dunning-Kruger, however, has substantially missed what the research actually shows. The finding that incompetent people overestimate themselves, while real and replicable, is only one side of a more interesting and more practically important pattern. The other side — less frequently discussed, considerably more consequential for understanding how people navigate their lives and careers — is the consistent finding that highly competent people systematically underestimate themselves. The confidence gap runs in both directions. The direction that receives less attention is the one that affects the people most worth paying attention to.
What Dunning-Kruger Actually Found
Dunning and Kruger's original 1999 paper studied undergraduate students on tests of logical reasoning, grammar, and humour. Students in the bottom quartile of actual performance significantly overestimated their performance — they predicted scores well above their actual results. But the paper's finding about top performers was equally striking and is far less commonly cited: students in the top quartile consistently underestimated their performance. They predicted scores below what they actually achieved.
The explanation Dunning and Kruger offered for both findings rests on a single mechanism: metacognitive skill — the ability to accurately assess one's own performance — requires the same knowledge and cognitive tools as the performance itself. Someone who lacks the skill to perform a task well also lacks the skill to recognise what good performance looks like, and therefore cannot accurately locate their own performance relative to the standard. They fill the gap with a default assumption of competence. Conversely, someone who has developed genuine expertise has a clear and detailed internal model of what excellent performance requires. They are acutely aware of the gap between their current performance and that standard, and they systematically locate themselves closer to the gap than to the ceiling — even when, by any objective measure, they are performing at a high level.
Subsequent research has extensively qualified and refined the original findings. A 2020 meta-analysis by Gignac and Zajenkowski found that the magnitude of the Dunning-Kruger effect in the original paper was partly a statistical artefact of the methodology used — specifically, a regression-to-the-mean effect that would produce the characteristic pattern even with random data. The substantive finding — that people at the lower end of the skill distribution overestimate their ability more than people at the upper end — survived the critique, but with smaller effect sizes than the popular representation of Dunning-Kruger implies. More importantly for the present discussion, the finding of top-performer underestimation was not an artefact. It held across methodological refinements and has been replicated in multiple independent studies using different domains and populations.
The Inverse: Why Skilled People Underestimate Themselves
The underestimation of self by highly competent people is not merely a statistical curiosity. It is a robust psychological phenomenon with a set of contributing mechanisms that are now reasonably well understood.
The first and most fundamental mechanism is what Dunning and Kruger identified: expertise creates an accurate awareness of complexity. As a person develops genuine skill in a domain, they develop an increasingly fine-grained understanding of what the domain demands. A novice chess player does not know what moves a grandmaster would consider. A skilled chess player knows precisely which moves are available, which are theoretically best, and how their own play deviates from optimal. The novice, unaware of the distances they have not yet travelled, locates themselves at the edge of their known world. The expert, aware of all the distances remaining, locates themselves near the beginning of an apparently infinite road.
The second mechanism is social comparison calibration. Highly competent people tend to spend time with other highly competent people — their professional peers, their collaborators, the communities of practice in which their expertise developed. When your reference group consists primarily of people who are themselves exceptional, your own performance, measured against that group, appears less impressive than it would when measured against the general population. The Nobel laureate who works at an institution full of Nobel laureates is not comparing themselves to the average academic. The world-class surgeon who trained under other world-class surgeons calibrates their self-assessment against a reference group that would make any performance look modest.
A third mechanism involves what psychologists call the false consensus effect operating in reverse. People tend to assume that their own abilities, knowledge, and skills are more common than they actually are. An expert in a domain often finds the knowledge they have accumulated so natural and accessible — so available after years of practice — that they have difficulty recognising it as unusual. What is effortful for a novice has become automatic for them, and the gap between their effortful past self and their automatic present self has been crossed so gradually that they no longer see it as a gap. They project outward and assume that others, with reasonable effort, could access the same knowledge they have — which is both psychologically generous and systematically inaccurate.
Impostor Syndrome: The Clinical Face of the Confidence Gap
The confidence gap in high performers has a clinical cousin that has received considerable public attention in recent decades: impostor syndrome. First described by psychologists Pauline Clance and Suzanne Imes in a 1978 paper studying high-achieving women, impostor syndrome refers to an internal experience of intellectual fraudulence — the persistent belief, despite objective evidence of success, that one's achievements are not the result of genuine ability but of luck, timing, or the successful deceiving of others who would, if they knew the truth, revoke the recognition they have extended.
Clance and Imes originally described the phenomenon primarily in women, partly because their sample was exclusively female and partly because the cultural context of women in professional settings in the 1970s provided additional structural reasons for self-doubt. Subsequent research has found the pattern across gender, though with some differences in how it manifests and what triggers it. A review published in the International Journal of Behavioral Science in 2011 estimated that approximately 70 percent of people experience impostor feelings at some point in their lives — a prevalence that, if accurate, suggests it is less a pathology than a normal feature of navigating high-stakes domains without perfect information about how one measures up.
The cognitive structure of impostor syndrome is distinctive. Its defining feature is the attribution asymmetry: successes are attributed to external causes (luck, others' generosity, a fortunate performance on a day that was not representative) while failures are attributed to internal causes (a revelation of the underlying incompetence that was always there). This attribution pattern is the direct inverse of what psychologists call a self-serving bias — the tendency, more common in people without impostor syndrome, to attribute successes to ability and failures to external circumstances. Where self-serving bias protects self-esteem by claiming credit for good outcomes, the impostor pattern protects a particular self-narrative — "I am not as capable as people think" — by discounting good outcomes and counting bad ones.
The persistence of impostor feelings despite continued objective success is one of the phenomenon's most striking characteristics. Many people expect that accumulating evidence of genuine competence — promotions, awards, public recognition, measurable outcomes — will eventually silence the internal voice insisting on fraudulence. For those with entrenched impostor patterns, it typically does not. Each success is reinterpreted through the existing framework rather than updating it, in a pattern that resembles the confirmation bias documented in belief-perseverance research. The framework survives because it is protected from disconfirmation by the same attribution asymmetry that generates it.
The Role of Expertise Development in Self-Assessment
Understanding the confidence gap requires understanding how expertise changes a person's relationship to their own performance over time, and why the trajectory of skill development and the trajectory of self-confidence often run in opposing directions through the middle stages of learning.
A well-documented pattern in skill acquisition — sometimes described as the conscious competence ladder, though the model predates that label — describes how people move from unconscious incompetence (not knowing what you do not know) through conscious incompetence (knowing what you do not know), conscious competence (performing well through deliberate effort), and eventually to unconscious competence (performing well automatically). The painful middle stages of conscious incompetence and conscious competence are the stages at which the gap between current performance and perceived standard is most visible and most distressing. The novice is often more confident than the intermediate practitioner precisely because the novice cannot yet see the standard they are failing to meet.
Research by Erin Westgate and colleagues on skill development in writing found exactly this pattern: students' self-assessed writing ability dropped significantly between their first and second years of academic training, not because their writing had worsened but because their capacity to evaluate writing had improved faster than their writing itself. They could now see what good writing required, and this new awareness made their own work appear worse than it had before. Their writing had measurably improved; their confidence had measurably declined. The confidence nadir, in other words, can coincide with a genuine improvement in ability — a counterintuitive dynamic that has significant implications for how people interpret periods of self-doubt in developing careers.
Gender, Culture, and the Distribution of Underconfidence
The confidence gap between competence and self-belief does not distribute evenly across demographic groups, and the patterns are informative about the social construction of confidence rather than its individual psychological origins. Research across multiple countries and professional domains finds that women, on average, are more likely than men to underestimate their own competence, less likely to apply for opportunities unless they meet all stated criteria, and more likely to attribute success to luck and failure to ability rather than the reverse.
Katty Kay and Claire Shipman's book The Confidence Code (2014) synthesised a body of research suggesting that women in professional settings were generally operating with a confidence deficit relative to their competence — applying for fewer promotions, negotiating salaries less often, and speaking less in meetings — that was costing them professionally in ways that had no basis in actual ability differences. A widely circulated statistic from a Hewlett-Packard internal study — that men apply for jobs when they meet 60 percent of the listed criteria while women apply only when they meet 100 percent — captured the asymmetry in a form that resonated broadly.
The causes of this disparity are debated. Some researchers emphasise socialization patterns that reward confidence in boys and prescribe modesty in girls from early childhood, creating different internal reference standards for acceptable self-promotion. Others point to rational updating in the face of environmental signals: in environments where women's competence is more frequently questioned, where attribution of success to luck is applied to women more than men by external observers, and where the consequences of overconfidence are more severe, the apparently lower confidence of women may reflect accurate rather than distorted self-assessment of how their performance will be received and rewarded.
Cultural dimensions compound the individual and gender-based patterns. Research comparing self-assessments across cultures finds consistent differences between East Asian and Western populations, with East Asian respondents showing greater self-criticism and lower self-enhancement in self-assessments of ability. Whether this reflects genuine cultural differences in self-concept, different norms around modesty and self-promotion, or different degrees of accuracy in self-assessment is a matter of ongoing research. What is clear is that the confidence gap is not a fixed property of individuals but is shaped by the social and cultural environments in which people develop and operate.
The Cost of Chronic Underconfidence
If underconfidence in high performers were merely an internal experience with no external consequences, it would be primarily a matter of individual wellbeing. But the research on the downstream effects of self-assessed competence makes clear that it shapes behaviour in ways that have substantial professional and social consequences — consequences that fall disproportionately on the people with the most to offer.
Opportunity-taking behaviour is among the most directly affected. Research on negotiation consistently finds that people negotiate more aggressively when they believe they have high value to offer and more accommodatingly when they doubt their own value. Self-assessed competence predicts negotiated salary outcomes significantly, independently of actual competence — meaning that the same objective level of skill, evaluated differently by the person who possesses it, produces systematically different compensation over a career. The underconfident expert negotiates a lower starting salary, accepts the first offer more often, and advocates less vigorously for promotions, with effects that compound over decades.
Voice and visibility in collaborative settings follow similar patterns. Research on meeting behaviour, group decision-making, and team dynamics finds that people's willingness to contribute, disagree, and advocate for their positions is significantly predicted by their confidence in their own competence — independent of whether that confidence is calibrated to their actual ability. The high-ability, low-confidence person withholds contributions that would improve the group's decisions. The lower-ability, higher-confidence person contributes more than their ability warrants. Groups therefore make decisions that over-represent the less competent and under-represent the more competent, not because of any deliberate process but because confidence, rather than competence, is what drives vocal contribution.
The long-term career trajectory effects are well documented in research on mentorship and sponsorship seeking. Seeking out mentors, advocating for your own advancement, and making your work visible to people in a position to advance your career all require a degree of confidence in your own value that underconfidence systematically suppresses. The capable person who does not seek mentorship because they do not feel worthy of a senior person's time, who does not apply for a programme because they assume they will not be selected, who does not raise their hand for a visible project because they are not sure they can deliver — this person is not being modest. They are making decisions that shape a career trajectory, on the basis of a self-assessment that research suggests is more likely to be wrong than right.
What Calibrated Confidence Actually Looks Like
The goal of psychological work on the confidence gap is not to produce uniformly high confidence — that would simply invert the problem, replacing systematic underestimation with systematic overestimation. The goal is calibration: a self-assessment that is accurate relative to actual competence, rather than systematically biased in either direction.
Calibrated confidence has some counterintuitive properties. It involves the ability to hold high confidence in domains of genuine expertise simultaneously with genuine uncertainty about areas where one's knowledge is limited. The expert who is confident about what they know and honest about what they do not is exhibiting calibrated confidence. The person whose confidence does not vary with their actual competence — who is equally confident about everything — is not calibrated, regardless of whether their overall confidence level is high or low.
Research on expert forecasters, compiled by Philip Tetlock in his studies of geopolitical prediction, found that the best forecasters — those whose probabilistic predictions about real-world events were most accurate — were distinguished not by confidence level but by calibration quality. They were willing to say "I am 70 percent confident" rather than "I am certain" or "I have no idea." They updated their confidence when new information arrived. They knew the domains where their predictions were reliable and the domains where they were not. These cognitive habits — probabilistic rather than binary confidence, willingness to update, domain-specific rather than global self-assessment — are the markers of calibrated confidence rather than the markers of either high or low confidence per se.
Building More Accurate Self-Assessment
Research on improving the accuracy of self-assessment points to several practices that move people toward calibration rather than simply toward higher confidence. The distinction matters because interventions that simply boost confidence without improving accuracy produce overconfident people rather than calibrated ones — which is a different problem, not a solution.
Structured feedback from credible sources is the most consistently effective input for improving self-assessment accuracy. The challenge is that unsolicited feedback from the social environment is rarely structured for accuracy — it is shaped by social norms around politeness, by the asymmetry between positive feedback (easy to give) and critical feedback (requiring courage), and by the interests of the feedback-giver. Seeking out feedback from people who are both knowledgeable about the domain and willing to give honest assessments — and creating explicit permission for that honesty — provides the calibration data that normal social interaction tends to filter out.
Comparative performance data, when available, provides a more objective anchor than internal self-assessment. Knowing that your work was ranked in the top 15 percent of a peer group gives a calibration point that is harder to explain away than the subjective feeling of uncertainty. Professional certifications, competitive examinations, and performance reviews that include explicit benchmarking serve this function, though their effectiveness depends on how rigorously the comparison group is defined.
Cognitive reframing of the impostor experience itself has evidence support from ACT-based approaches. Rather than attempting to eliminate impostor feelings through reassurance or evidence-gathering — both of which tend to provide temporary relief without addressing the underlying attribution pattern — the ACT approach involves changing the relationship to the thoughts rather than their content. Noticing the thought "I am a fraud" as a thought, rather than treating it as a factual report on reality, reduces its behavioural influence without requiring the person to first convince themselves it is false. The thought can be present without being obeyed.
Perhaps most practically, developing a habit of asking "what would I advise a friend in this situation?" provides an outside-view corrective to the inside-view bias that drives underconfidence. Research consistently finds that people give more accurate and more generous assessments of others' abilities than of their own — the same information, evaluated from a third-person perspective, produces less distorted conclusions than the first-person perspective produces. Using this asymmetry deliberately, by explicitly adopting the external viewpoint when evaluating your own performance and decisions, does not eliminate the bias but measurably reduces it.
The Confidence We Earn and the Confidence We Deny Ourselves
The psychological literature on competence and confidence tells a story that cuts against several comfortable assumptions. It suggests that the people who speak most confidently in rooms are not reliably the people most worth listening to. It suggests that the people whose self-doubt leads them to prepare more carefully, to question their conclusions more rigorously, and to defer to others more readily are often the people whose competence most warrants the confidence they are not expressing. And it suggests that the relationship between knowing something and believing you know it is shaped by mechanisms — the development of expertise, the social context of comparison, the cultural and gender norms that govern self-presentation — that have no necessary connection to the truth of the matter.
Understanding this does not make confidence irrelevant. Confidence is a resource that enables action, and action is where competence becomes useful. The capable person who never applies, never advocates, never raises their hand is, from the perspective of the world that benefits from their capability, functionally equivalent to the incapable one. The confidence gap costs something real — not just to the individual who carries it but to every context that loses access to what that individual could contribute.
Calibrating the gap — moving toward a self-assessment that reflects actual competence rather than a distorted interior narrative — is not an exercise in self-promotion or arrogance. It is an exercise in accuracy. And accuracy, in this as in other domains, is something worth working toward.