Illusory Superiority: Why Most People Rate Themselves Above Average

The effect is an arithmetic impossibility held by individuals who are each being reasonable. Half of any group sits below the median by definition, but when a group is surveyed, the share placing itself above it is routinely 70 or 80 percent. No single answer in that survey is provably wrong, which is why the finding is about groups and not about the person reading this.

The engine behind it, on the account with the most support, is the room a vague quality leaves for a flattering definition. David Dunning, Judith Meyerowitz and Amy Holzberg argued in 1989 that most of the qualities people are asked about, being sensitive or a good leader or disciplined, have no fixed meaning, so each person answers using the version of the quality they happen to be good at. Being a good listener can mean staying quiet, asking the right question, or remembering what was said last month. Whichever you do well is the definition that arrives when you rate yourself, and you are not lying.

That explains the pattern the demo above is built to show. Where a quality is specific and checkable, there is less room to move the definition, and the effect shrinks: timed mental arithmetic has a right answer, and parallel parking is watched by the whole street. Zell and colleagues' 2020 meta-analysis found exactly this moderation: larger effects for broad traits than for specific abilities, and the smallest for the percentile format this demo uses.

Two other mechanisms are documented. Mark Alicke showed in 1985 that the effect is larger for traits people rate as more desirable and as more under their own control, a result that Ignazio Ziano, Cora Mok and Gilad Feldman replicated in 2021 with 1,573 participants and 149 traits. And Justin Kruger showed in 1999 that comparison judgments are made egocentrically: people estimate their own standing first and then adjust too little for the group, which produces an above-average effect on easy abilities and a below-average effect on hard ones. Ask about driving and most people are above average; ask about juggling or telling jokes and most place themselves below it.

None of this makes one person's high self-rating false. The finding is about the sum of a group's answers.

The study that moved the field from documenting the effect to explaining it is Dunning, Meyerowitz and Holzberg's "Ambiguity and self-evaluation: The role of idiosyncratic trait definitions in self-serving assessments of ability," published in the Journal of Personality and Social Psychology in 1989. Their claim was that people compare themselves with peers using their own definition of the quality, and that definitions quietly favor their owner.

Participants rated themselves relative to their peers on qualities chosen for how much interpretive room they allowed. Ambiguous ones, such as sensitive and neurotic, can be satisfied in many different ways; unambiguous ones, such as mathematical and gossipy, cannot. Self-enhancement was strong on the ambiguous qualities and much weaker on the clearly defined ones, which is the comparison the demo on this page reproduces. Dunning and Rory McElwee followed the argument through in 1995, showing that people's private definitions of a trait predict how they describe themselves and how they judge others.

Mark Alicke's 1985 paper in the same journal came at it from the other side. He had participants compare themselves with the average college student across a long list of trait adjectives, and then related the size of each comparison to independent ratings of how desirable and how controllable that trait was. The better-than-average effect grew with desirability, and desirability mattered more for traits people saw as under their own control. Ziano, Mok and Feldman repeated the study in 2021 with two large online samples and found the same two results.

The best single summary of the size of the effect is the 2020 meta-analysis by Ethan Zell, Jason Strickhouser, Constantine Sedikides and Mark Alicke. Pooling 124 published papers and more than 950,000 participants, they reported an average effect of d = 0.78. It did not differ by sex. Culture is less simple: across all samples no reliable difference appeared, but in 11 studies comparing matched groups directly it was larger for European Americans, d = 0.98, than for East Asians, d = 0.40. The size also depended heavily on how the question was asked: the direct comparison formats produced the largest effects and percentile placement, as used above, the smallest.

The effect surfaces wherever people rate themselves on a quality that has no agreed yardstick. The rows below are illustrations, not findings from the studies cited on this page.

Since the effect lives in the gap between a quality and a measure, the counters all close that gap.

• Convert the quality into a task. Not "am I a good listener" but "can I say back what the other person said, to their satisfaction, at the end of a conversation." A task has an outcome; a trait has a definition you supply.
• Write your definition down before you rate yourself, then ask whether somebody strong on a different version of the quality would accept it.
• Use counts where counts exist. Deliveries on time, tickets, call-backs, errors caught in review. The research shows the effect shrinking as the quality gets more checkable.
• Ask for the comparison, not the compliment. "Where would you put me among the people on this team" gets a different answer from "am I doing well."
• Expect the reverse on hard skills. Kruger's 1999 finding predicts that you will place yourself below average on anything difficult, which is the same egocentric comparison running the other way; a below-average self-rating is not evidence of good calibration.
• Treat the group arithmetic as the check. If every member of a team rates itself in the top quarter, the ratings are measuring something other than rank.

The better-than-average family has several members, and they are measured in different ways.

• It is not the Lake Wobegon effect, which is the popular name for the same arithmetic, taken from Garrison Keillor's fictional town and used mostly for groups and institutions that report being above average. That page covers the name, the driving studies and the school test-score case; this one covers the general finding and its mechanism.
• It is not the Dunning-Kruger effect, which is about how the gap between skill and self-assessment changes across the skill range. Illusory superiority is the population average; Dunning-Kruger describes who contributes most to it.
• It is not the self-serving bias, the habit of crediting your successes to yourself and your failures to circumstances. That is about explaining outcomes after they happen; this is a comparison judgment with no outcome attached.
• It is not the overconfidence effect, which is stating more certainty in an answer than accuracy warrants. Overconfidence can be checked against a score; a comparison with the average person usually cannot.