The Original Research: Kruger and Dunning (1999)
People with low ability in specific domains systematically overestimated their own competence; highly skilled participants tended to underestimate their relative performance.
Kruger and Dunning (1999), Journal of Personality and Social Psychology, pattern as described on this page; schematic, no data. Later work (Nuhfer et al. 2016; Gignac and Zajenkowski 2020) attributes part of the pattern to regression to the mean.
In 1999, Justin Kruger and David Dunning published "Unskilled and Unaware of It: How Difficulties in Recognizing One's Own Incompetence Lead to Inflated Self-Assessments" in the Journal of Personality and Social Psychology. The paper reported a series of experiments demonstrating that people with low ability in specific domains (logical reasoning, grammar, and identifying humorous content) systematically overestimated their own competence.
The mechanism they proposed was metacognitive: to be skilled in a domain requires not only the domain knowledge itself but also the metacognitive capability to assess one's own performance accurately. People who lack domain knowledge also tend to lack the metacognitive scaffolding needed to recognise their lack. They cannot see what they do not know, so they assume they know more than they do. It is the reason our new manager training measures what managers do in a rehearsed conversation rather than how confident they feel afterwards.
The paper also reported the mirror pattern: highly skilled participants tended to underestimate their relative performance, in part because they assumed that tasks easy for them were also easy for others.
Kruger and Dunning won the Ig Nobel Prize in Psychology in 2000 for the research, and the phenomenon they described entered popular culture under their names.
The Dunning-Kruger Effect in Workplace Training
The Dunning-Kruger effect has specific implications for workplace training design and measurement that are consistently underappreciated.
Post-training satisfaction surveys are unreliable. The most common training measurement instrument is the Level 1 satisfaction survey (in the Kirkpatrick framework): "How confident do you feel about X after this training?" Participants who have just been exposed to a topic for the first time often report high confidence, not because they have developed genuine capability but because they now know enough to recognise the topic as familiar. This is Dunning-Kruger: awareness produces inflated confidence that does not reflect actual capability.
Competence development produces a temporary confidence dip. As people develop genuine competence in a domain, they typically go through a phase where confidence is lower than it was immediately post-training. The learning curve produces not a smooth increase in confidence but a characteristic dip: initial exposure produces overconfidence, then genuine engagement with the difficulty of the skill produces more accurate (and therefore lower) self-assessment, then sustained practice produces both genuine capability and accurate confidence.
Self-reported training ROI is systematically inflated. If training participants are asked whether they will apply their learning, they typically say yes at high rates immediately post-training. If their actual behaviour is observed three months later, the application rate is typically much lower. The post-training confidence (Dunning-Kruger driven) inflates the self-predicted application rate relative to the actual application rate.
The Sidestream Research Connection
The 2020 UCL study behind our method observed a pattern it discusses through the lens of the Dunning-Kruger effect, in the domain of communication skills training.
Rehearsal scored higher than either passive format. The bars overlap: individual scores varied, and the difference lies in the group means. Slideshow and video did not differ significantly. The rehearsal condition was an online coaching session of at most 20 minutes with a role-play coach, so the study tests the principle behind the method, not today's programme.
Source: Laumann, B. (2020), Active vs. passive L&D: differences between modalities for effective L&D in communicational competence, PSYCH0056, University College London, unpublished. Randomised, n = 57. Cohen's d = 1.08 (practice vs video) and 0.75 (practice vs slideshow), Tukey-HSD with Bonferroni correction; slideshow vs video not significant, p = .838.
Participants who had attended passive-modality training (slide-deck, video e-learning) on communication skills reported high confidence in their post-training self-assessments. When all participants' delivery was rated blind by two independent assessors, the group that had rehearsed with a role-play coach performed significantly better than the slideshow and video groups. Yet confidence had risen in every group, with no significant difference between the groups, and no significant relationship was found between participants' final confidence and their assessed performance.
This finding had two implications for design. First, it validated the shift from satisfaction-survey measurement (Level 1) to behavioural-observation measurement (Kirkpatrick Level 3) as the meaningful output metric. Second, it pointed to immersive role-play as a design approach that surfaces the gap between self-perceived and actual performance, because the professional-actor response in the rehearsal reveals immediately whether the participant's approach was effective in practice.
In a randomised 2020 UCL study by co-founder Ben Laumann (n = 57), rehearsal with a role-play coach produced significantly better feedback delivery than an instructional slideshow or video, rated blind by two independent assessors.
The Dunning-Kruger Effect and the Case for Kirkpatrick Level 3
Sidestream measures observed behaviour at Level 3, not the smile sheet. Level 3 as the minimum, Level 4 where the brief allows.
Source: Kirkpatrick, J. and Kirkpatrick, W. K. (2016), Kirkpatrick's Four Levels of Training Evaluation. Measurement commitment: Sidestream's own method description, not a research finding.
The Dunning-Kruger effect provides one of the strongest arguments for Kirkpatrick Level 3 (observed behaviour) as the minimum acceptable training measurement standard.
Level 1 measurement (satisfaction and confidence surveys) is vulnerable to Dunning-Kruger: the training produces awareness, awareness produces inflated confidence, inflated confidence produces high scores on the satisfaction survey, and the training is evaluated as successful without any actual behaviour change having occurred.
Level 3 measurement bypasses this vulnerability because it does not ask participants how confident they feel. It asks observers what the participant actually did. The observed behaviour either changed or it did not. No amount of Dunning-Kruger-driven confidence can produce a Level 3 outcome without actual behaviour change.
For organisations procuring training with behaviour-change objectives, the specification should include Level 3 measurement as a minimum. This is the methodological commitment that separates training designed for genuine behaviour change from training designed to produce positive satisfaction scores.
Counter-Critiques and Research Updates on the Dunning-Kruger Effect
The original Dunning-Kruger finding has been extensively replicated, but some subsequent research has raised questions about the mechanism. Nuhfer et al. (2016) and Gignac and Zajenkowski (2020) among others argued that some of the effect may be produced by regression to the mean (a statistical artefact) rather than pure metacognitive failure. The debate is ongoing in the academic literature.
The practical implications for training design remain intact regardless of the precise mechanism: self-reported competence is an unreliable measure of actual competence, particularly shortly after training; observed behaviour is a more reliable measure; and training measurement systems should prioritise behavioural observation alongside (not instead of) satisfaction measurement.
Practical Implications for L&D Buyers
The Dunning-Kruger effect has three direct practical implications for L&D procurement decisions.
Implication 1: specify Level 3 measurement in training contracts. If your training contract specifies Level 1 satisfaction as the success measure, you have no protection against Dunning-Kruger-inflated scores that mask absent behaviour change. Specify Kirkpatrick Level 3 as the minimum measurement standard, with the measurement methodology described in the contract.
Implication 2: treat post-training confidence ratings with scepticism. When a participant says "I now feel very confident in X" immediately after training, this is a prediction. The Dunning-Kruger research establishes that this prediction is systematically over-optimistic for participants who have not yet developed genuine capability. High post-training confidence does not indicate behaviour change; it indicates exposure to content.
Implication 3: design for behavioural exposure rather than content exposure. The most direct way to counter the Dunning-Kruger pattern is to design training that exposes the gap between perceived and actual capability rather than widening it. Immersive simulation achieves this: the professional actor's response to the participant's behaviour reveals immediately whether the participant's approach was effective. This feedback is not available from passive content consumption.
Implication 4: building in embedding and follow-up. The Dunning-Kruger confidence curve typically dips as genuine competence develops, because the learner acquires enough knowledge to recognise what they do not yet know. Training that includes structured follow-up and practice after the initial event supports learners through this dip rather than leaving them to navigate it alone.
The Dunning-Kruger Effect and Communication Skills Training
Communication skills training is one of the domains where the Dunning-Kruger pattern is most consistently observed in L&D contexts. Participants who attend communication skills workshops reliably report high post-training confidence in their communication capability. When their actual communication behaviour is observed in real working contexts, the performance often does not match the self-assessment.
This is the specific pattern the 2020 UCL study behind our method ran into. Self-rated confidence rose after every format, including the two whose delivery was rated significantly weaker, and in no condition did self-reported confidence predict assessed performance. Confidence and competence moved independently of each other. That is why we measure observed behaviour rather than how capable people feel.
The Dunning-Kruger finding is one of the reasons Sidestream's design uses professional-actor feedback and Kirkpatrick Level 3 observation as measurement standards rather than post-training self-assessment. The actor's response and the observer's behavioural rating are external calibration sources that the participant's internal self-assessment cannot provide.
What is the Dunning-Kruger effect? In summary
The Dunning-Kruger effect, from Kruger and Dunning (1999), is the finding that people with low competence in a domain tend to overestimate their ability because they lack the metacognition to see what they do not know. For workplace training this matters directly: post-training confidence surveys are unreliable, because exposure to a topic inflates perceived competence well before genuine capability has formed. The methodological response is to measure observed behaviour rather than self-reported confidence.
- Low competence produces overconfidence; high competence often produces a temporary confidence dip as genuine skill develops.
- Kirkpatrick Level 3 observation bypasses the bias because it records what people do, not how confident they feel.
- L&D buyers should specify Level 3 measurement in contracts and treat post-training confidence ratings with scepticism.
- Immersive rehearsal surfaces the gap between perceived and actual performance that passive modalities leave intact.
Frequently Asked Questions
What is the Dunning-Kruger effect?
The Dunning-Kruger effect is the finding by Kruger and Dunning (1999, Journal of Personality and Social Psychology) that people with low competence in a domain tend to overestimate their own ability in that domain. They lack the metacognitive capability to recognise their own incompetence. Conversely, highly competent people tend to underestimate their relative ability because they assume tasks that are easy for them are easy for others.
What does the Dunning-Kruger effect mean for workplace training?
The Dunning-Kruger effect means that post-training self-confidence surveys are an unreliable measure of behavioural competence. Participants who have just attended training about a topic (harassment prevention, leadership behaviour, coaching skills) typically report high confidence. This confidence does not indicate actual capability, it may reflect the Dunning-Kruger pattern where exposure to a topic produces inflated perceived competence before genuine capability is developed. This is why Kirkpatrick Level 3 (observed behaviour) is a more reliable measure than satisfaction surveys.
What is an example of Dunning-Kruger in the workplace?
A manager attends a one-day coaching skills workshop. Post-workshop, they rate themselves as highly confident in coaching conversations. When observed in actual one-to-one meetings, the manager continues to default to directive advice-giving rather than coaching questions. The confidence is real; the capability is not yet there. The Dunning-Kruger pattern explains why the self-assessment score is high while the observed behaviour has not changed.
How does Sidestream address the Dunning-Kruger effect in training design?
By replacing self-report with observed-behaviour measurement. In Sidestream's randomised 2020 UCL study, self-rated confidence rose after every format with no significant difference between them, while blind-rated delivery differed significantly, and no significant relationship was found between final confidence and assessed performance. Sidestream's design substitutes Kirkpatrick Level 3 behavioural observation for Level 1 satisfaction surveys, which removes the Dunning-Kruger-produced false positive from the measurement.
Is the Dunning-Kruger effect real? Has the research been replicated?
The original Kruger and Dunning (1999) finding has been replicated extensively across domains, including logical reasoning, grammar, emotional intelligence, and financial literacy. Some subsequent research has raised questions about the statistical methodology and suggested that some of the effect may be produced by regression to the mean (a statistical artefact). The core finding, that people with limited skill tend to overestimate their skill, is well-supported. The precise mechanism and the effect size in specific contexts remain active areas of research.
What is the opposite of the Dunning-Kruger effect?
Impostor syndrome describes the phenomenon where highly competent people systematically underestimate their own capability and attribute their success to luck or circumstance rather than skill. This is in some ways the mirror of the Dunning-Kruger pattern. Both describe a miscalibration between perceived and actual capability, in different directions.
Sources
- Kruger and Dunning (1999)
- Nuhfer et al. (2016)
- Gignac and Zajenkowski (2020)
- Laumann, B. (2020)
- Kirkpatrick, J. and Kirkpatrick, W. K. (2016)