Lake Wobegon Effect: When Almost Everyone Is Above Average

The name comes from a joke and stuck because the joke kept coming true. Lake Wobegon is the fictional Minnesota town of Garrison Keillor's radio monologues, "where all the women are strong, all the men are good looking, and all the children are above average." The phrase entered research in 1988, when John Jacob Cannell published a survey of elementary achievement testing in Educational Measurement: Issues and Practice under a title that said it all: how all 50 states are above the national average.

Nothing mysterious was going on in the states' arithmetic. Each state was comparing its pupils against norms established years earlier by a test publisher, on tests that schools administered year after year and taught toward, so scores drifted up against a yardstick that stood still. Cannell also reported test-security failures and some outright cheating; the measurement explanation is the part that has held up. The individual version has a different cause and the same shape: ask people to place themselves in a distribution and the answers will not fit inside it.

Two mechanisms are documented for the individual case. Justin Kruger showed in 1999 that comparison judgments are made egocentrically: people assess their own ability first and then adjust too little for everybody else, which produces above-average ratings on anything that feels easy and below-average ratings on anything that feels hard. And David Dunning, Judith Meyerowitz and Amy Holzberg showed in 1989 that loosely defined qualities leave each person free to use their own definition, which tends to be the version they satisfy.

That second mechanism predicts which of the three sliders above behaves. Driving is the classic case: skill, safety and courtesy mean different things to a fast driver and a cautious one, and both can rate themselves highly. Honesty is similar. Humor is the interesting one, because it is the kind of ability people suspect they lack; joke-telling is one of the domains where Kruger found a below-average effect rather than an above-average one.

The word effect is doing a lot of work in this name, and it is worth being clear about what it does and does not cover. The impossibility is a property of the tally, not of any single answer: in a group where 90 percent claim the better half, at least 40 percent are wrong, and the data cannot say which. For the aggregate version, as Cannell's case showed, the explanation may be entirely about measurement rather than about anyone's self-image.

The study behind the driving version is Ola Svenson's "Are we all less risky and more skillful than our fellow drivers?", published in Acta Psychologica in 1981. It is two pages of data and it has been cited ever since.

Svenson asked 161 licensed drivers to place themselves on a percentile scale, marking one of ten intervals. Eighty-one were American students recruited through the University of Oregon student paper, with a median age of 22; 80 were psychology students at the University of Stockholm, with a median age of 33. Within each group, some answered about driving skill and the rest about how safely they drove. The comparison group was deliberately narrow: not drivers in general, but the other people in the room that day, which Svenson chose so that stereotypes about other regions' drivers could not explain the result.

In the American sample, 93 percent of the 41 who rated their skill put themselves above the median of that room, and 88 percent of the 40 who rated their safety did. The medians of the distributions sat in the 61 to 70 percent interval for skill and the 81 to 90 percent interval for safety, so half of the American sample believed they were among the safest fifth of the people around them. Nearly half, 46.3 percent, placed their skill in the top 20 percent.

The Swedish sample showed the same direction and less of it: 69 percent above the median for skill and 77 percent for safety, with 15.5 percent claiming the top fifth for skill. Svenson noted that earlier and less well documented surveys had put the share placing themselves in the safer half at around 70 to 80 percent, so the American figures were high but the pattern was not new. What his design added was a comparison group the participants could see.

The tally shows up wherever self-ratings or self-reports are collected and then added up. The rows below are illustrations, not findings from the studies cited on this page.

The effect is a problem for groups and measurements more than for individuals, so the useful counters are procedural.

• Ask where the distribution comes from. A self-rating against "the average person" has no distribution behind it; a rating against the ten people in the room, as in Svenson's study, at least has one.
• Replace the comparison with a count. Miles per claim, deliveries on time, tickets closed, errors caught. Counts can be ranked; impressions cannot.
• Add the tally up. If 80 percent of a team rates itself in the top quarter, the form is broken, no matter how sincere each answer is. That check takes one minute and it is the whole diagnosis.
• For any above-average claim by an institution, check the yardstick and its date. Cannell's 50 states were all above a national average set years earlier on a test the schools knew well.
• Expect the reverse on hard things. Kruger's egocentric account predicts below-average self-ratings for difficult abilities, so a modest answer about your sense of humor is the same mechanism, not better calibration.
• Keep in mind what the demo above cannot do. Your three sliders are not wrong, and the tally of everyone's results is a count of one response pattern, not the arithmetic check: that needs every rating on one trait set against one known distribution, which the page does not collect.

The name is used loosely, and it is worth separating from the research it sits on top of.

• It is not a separate finding from illusory superiority, which is the academic name for the same comparison pattern and the term used in the literature. That page covers the mechanisms and the meta-analytic size; this one covers the nickname, the drivers and the test scores.
• It is not the Dunning-Kruger effect, which is about how the gap between skill and self-assessment changes with skill. The Lake Wobegon tally says most people claim the better half; Dunning-Kruger says who claims it most.
• It is not optimism bias, the expectation that good things are likelier to happen to you than the odds allow. Optimism bias is about future events; this is about present standing.
• It is not the overconfidence effect, stating more certainty than accuracy warrants. Overconfidence can be scored against an answer key; a claim about being an above-average driver usually cannot be.