When people in a group are asked to rate their own driving ability, most of them rate themselves as better than average. That can’t be true for everyone, since some people must be below average, yet very few people place themselves there. The least skilled drivers are often among the most confident.

This pattern isn’t limited to driving. It appears in self-assessment surveys of many kinds, including skills inventories, performance reviews, workshop evaluations, and quizzes where people estimate their score before seeing the answers. People who know the least about a subject often believe they know a good deal about it, while people who perform well often underrate themselves. This is more than a minor flaw in self-reported data. It raises a basic question about how much weight to give people’s judgments of their own competence.

What the Dunning-Kruger Effect Actually Is

The Dunning-Kruger effect is named after psychologists David Dunning and Justin Kruger, who published a study in 1999 comparing how well people actually performed on tests of logical reasoning, grammar, and humor with how well they thought they had performed.

They found that low performers consistently overestimated their performance by a wide margin. Participants who scored near the bottom tended to place themselves above average. Dunning and Kruger argued that people with low skill in an area face two problems: they make mistakes, and they lack the knowledge needed to recognize those mistakes. Judging whether an answer, joke, or argument is good requires some competence in that subject, so people without that competence also lack the means to evaluate their own work accurately.

The other side of the finding gets less attention. In the same studies, top performers tended to underestimate how they compared with others. They found the tasks easy and assumed other people would find them easy too. The result is a bias at both ends of the skill range: low performers rate themselves too high, and high performers rate themselves somewhat too low.

Researchers continue to debate how much of the effect reflects a real gap in self-awareness and how much reflects statistical factors such as regression to the mean. The general pattern of poorly calibrated self-ratings, however, shows up across many studies and settings.

Why This Happens: The Skill Required to Judge Skill

Low skill at a task and low skill at judging one’s own performance on that task often come from the same source. Someone who lacks one usually lacks the other, and typically has no way of noticing the gap from the inside.

Take a survey that asks how knowledgeable you are about a piece of software. Someone who has never used its advanced functions doesn’t just lack those skills. They may not know what the advanced functions are, so their understanding of what the software can do is incomplete in a way they can’t detect. From their point of view, what they know looks like the full picture, because it is everything they know about it.

This is not the same as arrogance. It is an information gap. Respondents with lower skill may believe they are answering honestly, which is why rewording a question often does little to correct their answers.

How This Shows Up in Real Self-Assessment Surveys

The pattern appears in many applied settings, not just laboratory studies:

• Workplace skills inventories: Employees may be asked to rate their proficiency in technical skills such as data analysis, programming, or software use. Their self-ratings often correlate only weakly with their scores on objective skills tests in the same areas.
• Training needs assessments: Employees with the largest skill gaps may also be the least aware that they need training. Self-assessment can show what people believe they need, but relying on it alone may miss real gaps.
• Manager and leadership self-evaluations: Leaders’ ratings of their own communication and leadership skills often differ from the ratings they receive from direct reports and peers in 360-degree feedback. Comparing these perspectives can reveal areas for development.
• Financial literacy surveys: Research has found that people who score lower on objective tests of basic financial concepts, such as compound interest and diversification, often rate their own financial knowledge highly.
• Post-training surveys: Participants may report high confidence right after a training session without having mastered the material. Clear, well-organized instruction can make a topic feel easier than it is, and participants may take that ease as a sign they have learned it. Measuring confidence alongside knowledge retention or task performance gives a fuller picture of whether the training worked.

In each case, the survey asks respondents to evaluate their performance on something they may not fully understand. The survey may end up measuring their perception of their competence rather than their competence.

The Consequences for Survey Data and Decision-Making

When the Dunning-Kruger effect isn’t accounted for, it does more than add noise to self-assessment data. The error may not be random. Confidence can be especially poorly calibrated among people with lower knowledge or skill, so the people a survey is trying to identify may be the least likely to recognize their own gaps.

This has practical consequences:

• Training programs: If employees with the largest skill gaps rate themselves as highly capable, a training program based only on self-ratings may under-support the people who need the most help. The survey may suggest a group needs little training when an objective assessment would show otherwise.
• Hiring and promotion: If self-reported confidence is treated as a strong sign of competence, confident candidates may be rated more favorably than capable candidates who are more cautious about rating themselves. Confidence tells you how someone sees their abilities, but it isn’t evidence that those abilities are present.
• Product onboarding: Product teams sometimes ask, “How much experience do you have with this tool?” to decide whether a user needs beginner or advanced onboarding. If users overestimate their familiarity, some will be sent to advanced material before they have learned the basics, which leads to a worse experience and misleading conclusions about what users know.
• Trend analysis: Suppose a company tracks employees’ self-rated technical skills every year. Employees’ actual skills may improve over time while they also become more aware of how complex the skill is and how much they still don’t know. Their self-ratings could stay flat or even drop despite real improvement.

A change in self-perception is not necessarily a change in capability, so researchers should be careful about treating changes in self-assessment scores as evidence of changes in actual skill.

Strategies to Reduce Dunning-Kruger Bias in Survey Design

Self-assessment doesn’t need to be removed from a survey to make the data more useful. A better approach is to design the survey so that perceived ability can be compared with stronger evidence wherever possible.

(1) Pair Self-Assessment With Objective Measures

Combine self-reported confidence with an objective measure of performance. For example, instead of only asking employees to rate their Excel proficiency from 1 to 10, include a short task covering the specific skills the organization needs.

The two measures answer different questions:

• Self-assessment: How capable does the respondent believe they are?
• Objective assessment: How well can the respondent actually perform the task?

The gap between the two answers is useful information in itself. Even a small set of well-designed tasks can provide evidence that confidence ratings alone can’t.

(2) Ask About Specific Behaviors

General questions lead to general judgments. “Rate your data analysis skills from 1 to 10” asks respondents to compress a complex set of abilities into one number.

More specific questions might ask:

• “Can you create a pivot table from a raw dataset without assistance?”
• “How often have you independently used SQL to extract and analyze data in the past three months?”

Behavior-based questions give respondents something concrete to evaluate and reduce the distance between what they think they can do and what they have actually done.

(3) Give Respondents a Clear Reference Point

A rating scale means little if respondents interpret each point differently. Instead of asking people to rate themselves as “beginner,” “intermediate,” or “advanced,” describe what each level means.

For example, an advanced user might be defined as someone who can independently complete a specific set of tasks, troubleshoot common problems, and explain the process to someone else. Descriptions like this give respondents a clearer standard to judge themselves against.

(4) Compare Self-Ratings With External Assessments

Self-assessment is more informative when it is reviewed alongside assessments from people who have observed the respondent’s work, such as manager evaluations, peer feedback, or 360-degree reviews.

The external rating shouldn’t automatically be treated as correct and the self-rating as wrong, since external assessments have their own limitations and biases. Instead, look at the gap between the two. A large difference between how someone rates themselves and how others rate them is worth investigating rather than averaging away.

(5) Ask About Recent, Specific Experiences

People often find it hard to judge a broad trait such as “leadership ability” or “technical expertise.” Questions tied to recent experiences are easier to answer accurately.

Instead of asking:

“How good are you at managing difficult conversations?”

A survey might ask:

“Think about the most recent difficult conversation you had with a colleague. How confident were you that you handled it effectively?”

The second question gives the respondent a specific event to recall and evaluate, which tends to produce a more grounded answer than a question about their overall ability.

What This Means for Anyone Designing or Interpreting Self-Assessment Data

The Dunning-Kruger effect is not a reason to abandon self-assessment surveys. Self-reported data captures things objective tests can’t, including confidence, perceived barriers, attitudes, motivation, and how prepared people feel.

Problems arise when those measures are treated as substitutes for demonstrated ability. A person can be highly confident without being highly capable, and a capable person may rate themselves conservatively. Neither response should be taken as the full picture.

For researchers, this means measuring confidence and capability separately:

• Where actual performance matters, add an objective measure.
• Where behavior matters, ask about specific behaviors.
• Where perception matters, measure perception directly and report it as perception.

Pay attention to the gap between what respondents believe they can do and what they can demonstrate. That gap shows how accurately people understand their own abilities, which is useful information in its own right.

Conclusion

Self-assessment surveys carry a built-in distortion, because judging your own competence requires some of the same competence being measured. This doesn’t make self-reported data useless, but it does mean it shouldn’t stand alone when accuracy matters.

If the purpose of a survey is to understand how confident people feel, self-report may be exactly what you need. If the purpose is to measure competence, confidence alone is not enough. Stronger survey designs combine self-ratings with objective tasks, ask about specific behaviors instead of general labels, give clear definitions for each rating level, and compare self-assessments with external measures when a decision depends on the results.