Curse of Knowledge

What it is

Colin Camerer, George Loewenstein and Martin Weber introduced the term in a 1989 paper in the Journal of Political Economy. They were testing a quiet assumption in economic models of asymmetric information, that a better-informed person can reproduce the judgement of a less-informed one. If that were true, extra information would never be a handicap in predicting how other people see things. Their experiments found that it is. Better-informed people could not set their extra knowledge aside, even when doing so would have paid them to, and they concluded that more information is not always better. Comparing individuals judging alone with judgements made in market settings, they found that markets reduced the error by about half but did not remove it.

The name describes a specific failure. It is not ignorance of what others know, which is a plain gap. It is a pull in one direction: because you know the answer, other people seem to know it too, or to find it easier to reach than they do. The vault places it next to the hindsight bias, the loss of the ability to reconstruct what you believed before you knew how something turned out, and sets it against pluralistic ignorance, in which everyone underestimates what is shared. Under pluralistic ignorance people think they know less than others do. Under the curse of knowledge one person thinks others know what they know.

It also needs to be kept apart from the false consensus effect, the tendency to assume other people share your opinions and habits. A person who owns a phone and guesses that half the population does is closer to false consensus, because the error concerns a shared trait. The curse of knowledge concerns information: what a person has learned and cannot unlearn.

The effect has been studied in several settings, and each setting adds something. In education, the same error is called the expert blind spot. In writing and design, it is the reason that instructions, error messages and specifications drift out of reach of the people they are for. The common feature is a gap in position between someone who has already crossed the problem and someone who is about to.

In effect

In research: the tapping study

The most quoted demonstration is Elizabeth Newton's 1990 Stanford dissertation, The Rocky Road from Actions to Intentions. Forty tappers each tapped the rhythm of well-known songs for 40 listeners, 120 songs in all, and tappers predicted how often listeners would name them. The mean prediction was 50 per cent, with individual guesses from 10 to 95 per cent. Listeners named three songs, a success rate of 2.5 per cent. The tapper hears the melody in their head while tapping. The listener hears only knocks. The dissertation's abstract refers to 150 tunes while its table note gives 120 songs, so the arithmetic here follows the table and the 2.5 per cent figure the text reports.

In everyday life: following a manual

Pamela Hinds tested the same gap on instructions. In a 1999 paper in the Journal of Experimental Psychology: Applied, she asked people with different levels of expertise to predict how long novices would need to learn a new mobile phone task and what problems they would meet. Secondary accounts report that experts were too optimistic about both, and that the more expert the person, the further out the estimate. The abstract was not available for this page, so the direction is reported and the figures are not. The practical reading is familiar to anyone who has assembled furniture from a booklet written by the person who designed it: the steps that feel too obvious to write down are the ones the beginner needs.

In commerce and design: the maker's view

Richard Shotton applies the idea to advertising in The Choice Factory (Bias 16, pp. 111 to 116). Makers see a poster at a desk, and consumers glimpse it from a moving car. He cites an industry eye-tracking figure of under a second for the average glance at an online display ad. The vault treats that as an industry figure and not published research. His advice is to judge work as the audience meets it, which in practice means testing it on people outside the team.

Cindy Xiong, Lisanne van Weelden and Steven Franconeri found a related effect in data visualisation. In four experiments with between 17 and 29 students each, participants read a backstory that made one pattern in a graph stand out to them, then predicted what a viewer who had not read it would notice. They were told to ignore what they knew and still predicted that the naive viewer would see the same pattern. The effect was weaker when the graph had no visual annotations, and the samples were small.

In teaching: the expert blind spot

Mitchell Nathan, Kenneth Koedinger and Martha Alibali reported in 2001 that 105 teachers ranked algebra problems for beginning students in an order that, for high school teachers, bore no relation to how two samples of ninth-graders actually performed. Middle school teachers ranked them in close agreement with the students. The finding is a conference paper and is covered at Expert Blind Spot.

What it does not say

The curse of knowledge is well supported in direction and disputed in size. The most direct check is the replication by Rachel Ryskin and Sarah Brown-Schmidt, published in PLoS ONE in 2014. They re-ran a false-belief version of the task first reported by Susan Birch and Paul Bloom in 2007, with 3,074 participants across seven experiments. The direction held. The knowledge effect was 0.20 against 0.47 in the original, less than half the size. The vault's standing note is therefore to cite the existence of the bias and not large effect sizes.

The studies also measure different things. The 1989 markets, the tapping study, the phone task, the algebra rankings and the graph experiments use different tasks and different units, so their sizes cannot be compared with each other or averaged into a single figure for "how much" experts mislead novices. The term describes one specific error and does not show that experts are bad communicators in general.

Barbara Minto's The Pyramid Principle describes the gap on page 73, where she says the writer and the reader are not peers in interpreting the writer's thinking, but she cites no experiment for it, and the book is a method resting on consulting practice. It is useful as a procedure and is not evidence for the effect.

Who developed it

The term comes from Colin Camerer, George Loewenstein and Martin Weber (1989). See Colin Camerer, George Loewenstein and Martin Weber. The experimental demonstration is theirs. The idea sits next to earlier work on hindsight bias, and Elizabeth Newton's 1990 tapping study gave it its best-known picture. See Elizabeth Newton.

---