The sixty-seventh article was about combining what you know. This one is about a stranger problem: that knowing something makes you systematically wrong about what other people know, and that you cannot undo it by trying.
Key Takeaway
The paper that named it reports: "Better-informed agents are unable to ignore private information even when it is in their interest to do so; more information is not always better." And: "market forces reduce the curse by approximately 50 percent but do not eliminate it."[1] On our own arithmetic, halving a twentyfold overestimate leaves one that is still ten and a half times too high.
The Verdict, Stated First
Five claims, in descending order of confidence.
One. The economic finding is well sourced and specific. A 1989 paper in a top journal, whose abstract we obtained verbatim from a copy hosted by a co-author's own university, reporting a quantified partial correction by market forces.
Two. The famous demonstration is not in the peer-reviewed record. The tapping study everyone cites is an unpublished 1990 dissertation, and every source we could find for its numbers is a blog or a popular book.
Three. Those popular retellings disagree with each other on the success rate, the songs used, and the scale of the study, which is what usually happens to a result that travels without a paper behind it.
Four. The partial correction is much weaker than it sounds. On our own arithmetic a fifty percent reduction applied to a twentyfold error leaves a 950 percent overestimate, because proportional reductions bite feebly on multiplicative errors.
Five. The commercial consequence is a planning error, not a communication one. On our own illustration, someone who believes each explanation works half the time plans four attempts where the reported figures imply ninety-one.
Our Grades For These Claims
Applying the scheme from the first article in this series.
Grade A for the 1989 findings, from an abstract obtained verbatim from the paper itself, hosted by a co-author's institution, and corroborated by an economics database.
Grade D for the tapping study's numbers, which is the lowest grade we have given a widely cited result. Every source is non-academic, the underlying work is unpublished, and the accounts conflict.
Grade C for the debiasing literature, which we can name from a title and did not obtain.
Grade A for our own arithmetic, which is elementary.
Our position: the economics is solid and modest; the famous illustration is unverifiable and enormous, and the gap between those two sentences is the reason this article exists.
A Note On Method
Everything here is verified to August 2026.
We obtained the 1989 paper's abstract verbatim from the paper itself, in a copy hosted by the second author's own university department, together with its opening paragraph and acknowledgements[1]. That abstract is corroborated word for word by an economics database[2]. We did not obtain the paper's experiments or results tables.
For the 1990 tapping study we obtained nothing academic at all. Its figures reach us through two personal newsletters and two popular books quoted on a reader-notes site[3][4], all flagged at every use.
The account of the paper's motivation and its relationship to earlier work comes from an encyclopedia entry[5], flagged at every use.
We did not obtain the debiasing study, the expertise study, or the 1975 hindsight paper, and report all from citation records and titles.
All arithmetic is ours, and one of the three tables rests on an independence assumption we believe is false and explain in the body.
This article discusses research on communication and judgment. It is not marketing, pricing or project management advice.
The 1989 Paper
The source, and the provenance is unusually good.
Camerer, C., Loewenstein, G., and Weber, M. (1989), The Curse of Knowledge in Economic Settings: An Experimental Analysis, Journal of Political Economy, 97(5), 1232–1254, October, DOI 10.1086/261651[2].
Its opening states the commercial case directly: "In many economic transactions, some agents know more than others. For example, sellers are better informed about the true value of their products than buyers; workers know more about their ability and motivation than prospective employers."[1]
Three observations, ours.
The examples chosen are selling and hiring, which are the two decisions this series has covered most, and they are the paper's own framing rather than our transfer.
The author list overlaps with the sixty-second and sixty-seventh articles. Loewenstein co-wrote the discounting review; Camerer co-wrote the cab driver study. Three appearances in this series in three different literatures.
And we obtained the paper's front matter from a copy hosted by the second author's own university department, which is the best provenance short of the publisher and better than most sources in this series.
The Assumption Being Violated
What the paper is arguing against, in its own words.
"In economic analyses of asymmetric information, better-informed agents are assumed capable of reproducing the judgments of less-informed agents. We discuss a systematic violation of this assumption that we call the 'curse of knowledge.'"[1]
An encyclopedia entry describes the aim as countering "conventional assumptions in such (economic) analyses of asymmetric information in that better-informed agents can accurately anticipate the judgement of less-informed agents"[5]. This is an encyclopedia, not an academic source, flagged here and at every use.
Four observations, ours.
The target is an assumption inside a body of theory, not a piece of folk wisdom. Models of information asymmetry required informed parties to be able to simulate uninformed ones, and the paper says they cannot.
Note the direction that assumption points. Standard theory treated knowing more as an unambiguous advantage, and the whole literature on asymmetric information is about how the less-informed party catches up.
The encyclopedia records that framing: "Most theoretical analyses of situations where one party knew less than the other focused on how the lesser-informed party attempted to learn more information to minimize information asymmetry."[5]
And the paper's contribution is to ask the reverse question. Whether the better-informed party is disadvantaged by knowing, which the encyclopedia describes as an idea "seen as something important to bring to the sphere of economic theory."[5]
More Information Is Not Always Better
The central claim.
"Better-informed agents are unable to ignore private information even when it is in their interest to do so; more information is not always better."[1]
Four observations, ours.
The clause "even when it is in their interest to do so" is what makes this a finding rather than an observation. These are people with an incentive to set the knowledge aside, failing to.
That distinguishes it from ordinary carelessness. Nobody is being lazy; they are being asked to do something and cannot.
"More information is not always better" is a strong claim in an economics journal, where the default is that information has non-negative value.
And the practical form is worth stating now, since the rest of the article follows from it. Learning something can make you worse at a specific task: predicting what someone who has not learned it will think.
Markets Halve It
The quantified result, which almost never survives into popular accounts.
"Comparing judgments made in individual-level and market experiments, we find that market forces reduce the curse by approximately 50 percent but do not eliminate it."[1]
The paper's own list of applications follows: "Implications for bargaining, strategic behavior by firms, principal-agent problems, and choice under uncertainty are discussed."[1]
Four observations, ours.
This is the sentence to keep and it is two findings, not one. Markets help substantially, and markets do not solve it.
The 50 percent is a specific figure attached to a specific comparison, between individual judgments and the same judgments made in a market setting where prices aggregate them.
It is also a partial answer to the obvious objection to any laboratory bias. The standard reply is that real markets discipline error, and this paper tested that and reports the discipline working halfway.
And we did not obtain the experiments, so we cannot tell you what was traded, by whom, or how the fifty percent was computed.
One further thing the fifty percent figure implies, which we would put to anyone who dismisses laboratory findings as artificial. The market condition is the realistic one and it still showed the effect. A bias that survives a market at half strength is not an artefact of asking undergraduates odd questions.
And the direction of the residual matters commercially. Markets discipline error by making it expensive, and the half that survived did so despite being expensive, which tells you the effect is not a matter of insufficient motivation. That is the same conclusion the 1975 precursor reached by a different route, as the section below sets out.
What Halving Actually Leaves
Why fifty percent is less encouraging than it sounds. Our own arithmetic.
A proportional reduction applied to a multiplicative error bites feebly. Halving the excess in an overestimate leaves:
A 20x overestimate becomes 10.50x, still 950 percent too high.
A 10x becomes 5.50x, still 450 percent too high.
A 5x becomes 3.00x, still 200 percent.
A 3x becomes 2.00x, still 100 percent.
A 2x becomes 1.50x, still 50 percent.
Three observations.
Halving a twentyfold error leaves a ten-and-a-half-fold one. The phrase "reduced by half" reads like a solution and describes a residual error of nearly a thousand percent.
The general point is worth carrying beyond this article. Bias reductions are usually reported proportionally and experienced multiplicatively, and the two are not the same news.
And a caution on our own table. The 50 percent figure applies to the 1989 experiments, not to the tapping study below, and combining them is our illustration of how partial correction behaves rather than a finding about anything.
The Tapping Study
The demonstration everybody knows, which we are about to grade very poorly.
The account, as popular sources give it: participants sat back to back. One was given well-known songs and tapped the rhythm on a table. The other tried to identify them. Tappers predicted listeners would identify the song about half the time; listeners identified 2.5 percent[3].
The explanation offered: "While the tappers heard the entire song in their minds, complete with lyrics, melody, and even instrumental accompaniments, the listeners were only exposed to the simple rhythmic pattern."[3]
The attributed source is Newton, E. L. (1990), The Rocky Road From Actions to Intentions, a doctoral thesis at Stanford University[3].
Where Those Numbers Reach Us From
The sourcing, stated plainly because it is the point of this section. Ours.
Four observations.
We found the figures in two personal newsletters[3][4] and in two popular books quoted on a reader-notes site[6]. Not one academic source.
An encyclopedia entry confirms the study's existence and describes it as "A 1990 experiment by a Stanford University graduate student, Elizabeth Newton"[5], and is itself not an academic source.
We did not obtain the thesis, and a doctoral thesis is not a peer-reviewed publication.
And this is a widely taught result. It appears in business books, education writing, and management training, and on the sourcing available to us it rests on an unpublished student project from thirty-six years ago that we could not read.
Two further points on why we are labouring this. A reader who repeats the tapping figures in a presentation is making a factual claim about an experiment, and on the sourcing available to us that claim cannot be checked by anyone in the room, including them.
And the alternative is easy rather than costly. The published paper says the same thing, with a smaller number and a peer review behind it, and a presenter who quotes it loses a good anecdote and gains a defensible position.
An Unpublished Dissertation
What that status means, since we are not saying the study is wrong. Ours.
Four observations.
A dissertation is examined but not peer-reviewed in the journal sense, and it is not indexed, replicated or corrected in the way a published paper is.
The practical consequence is that nobody can check it. There is no abstract to obtain, no method section circulating, and no record of whether anyone has attempted a replication.
We are not claiming the effect is absent. The 1989 paper is peer-reviewed, published in a top journal, and reports the underlying phenomenon with a quantified partial correction.
We are claiming that the number everyone quotes is not the number anyone verified, and that the peer-reviewed finding is much more modest than the illustration attached to it. That gap is the single most common structure this series has recorded.
The Retellings Disagree
The evidence that the number has drifted, which is what happens without a paper to anchor it. Ours.
Three discrepancies, from four popular sources.
On the success rate: three sources say 2.5 percent[3][4], and one popular book says the listeners "correctly guessed about 3 percent of the songs."[6]
On the songs: one source names "Twinkle Twinkle, Little Star"[3], another names "Happy Birthday" and "Jingle Bells."[4]
On the scale: two sources refer to 120 melodies[4], while one asserts the experiment "was done with thousands of people across the world and with almost 120 songs"[4], which describes a very different study.
Four observations.
The 2.5 against 3 percent difference is small and it is the kind of rounding that accumulates. A reader encountering only the second would quote a figure nobody reported.
The song variation is harmless and revealing. It suggests retellers are reconstructing a plausible version rather than reporting one.
The scale claim is the serious one. Thousands of people across the world is not a doctoral thesis, and one source has upgraded a student project into an international programme.
And that is the mechanism this series watches for. Where there is no paper to check against, the account grows in the direction that makes it more impressive, and every reteller is honest.
The Size Of The Gap
Taking the reported figures at face value for a moment. Our own arithmetic, on numbers we have graded D.
Predicted 50 percent against observed 2.5 percent:
An overestimate factor of 20 times. An absolute gap of 47.5 percentage points. Predicted odds of one in 2 against observed odds of one in 40.
Three observations.
A twentyfold overestimate is not a calibration error. Someone mis-set by ten or twenty percent has the right picture badly measured; someone off by a factor of twenty has a different picture.
The odds framing is the one we would use in a meeting. One in two against one in forty is easier to feel than fifty percent against two and a half.
And we would not build anything on these figures. They are the least verifiable numbers in this series, and we compute them only to show what would follow if they were right.
What It Does To Getting A Message Across
The consequence that matters commercially. Our own arithmetic, and see the caveat immediately following.
If each attempt at explaining something succeeds with probability p, the number of attempts needed to reach 90 percent cumulative comprehension:
At 50 percent per attempt: 4 attempts.
At 25 percent: 9. At 10 percent: 22. At 5 percent: 45. At 2.5 percent: 91.
Four observations.
Someone who believes they are at fifty percent plans four touches. If the reported figures were right they would need ninety-one, a plan wrong by a factor of 23.
The error is therefore not in the explanation. It is in the plan around the explanation: how many times, through how many channels, with how much checking.
That reframing is the useful one. The curse of knowledge produces a scheduling failure rather than a wording failure, which is why better writing does not fix it and repetition does.
And the shape holds even at far gentler assumptions. Even at ten percent per attempt the answer is 22 rather than 4, which is a fivefold planning error from a much less dramatic starting point.
The Assumption In Our Own Table
Which we think is false, and say so before anyone uses the numbers. Ours.
Four observations.
The table assumes each attempt succeeds independently. That is almost certainly wrong, and wrong in a helpful direction.
A second explanation is usually better than the first, because you have seen where it failed. Attempts are not independent draws; they are a learning process.
So the real number of touches needed is lower than our table says, possibly much lower, and we would not quote ninety-one to anyone as a target.
What survives is the ordering, which is what we are actually claiming. The gap between believed and actual comprehension translates into a planning error of the same order as the gap itself, and the direction of that error is always to under-plan.
The Precursor
Where the idea came from, which connects this to an article already in this series.
An encyclopedia records that the work "drew from Baruch Fischhoff's work in 1975 surrounding hindsight bias", and that his research "revealed that participants did not know that their outcome knowledge affected their responses, and, if they did know, they could still not ignore or defeat the effects of the bias."[5]
It adds: "Study participants could not accurately reconstruct their previous, less knowledgeable states of mind."[5] We did not obtain the 1975 paper.
Four observations, ours.
The two findings are the same mechanism pointed at different targets. Hindsight bias is failing to reconstruct your own earlier state; the curse of knowledge is failing to reconstruct someone else's current one.
The clause about knowing and still not defeating it is the important one, and it is the reason the remedies at the end of this article are structural rather than motivational.
The encyclopedia also quotes Fischhoff as noting that how well people reconstruct the perceptions of lesser informed others is "a crucial question for historians and 'all human understanding'"[5], which is a larger claim than the economics.
And this series has already covered hindsight bias, so a reader wanting the reconstruction-of-your-own-past half should read that article alongside this one.
Can It Be Lifted?
The debiasing question, from a title we can name and a paper we did not obtain.
Reference lists identify Damen, D., van der Wijst, P., van Amelsvoort, M., and Krahmer, E. (2020), Can the curse of knowing be lifted? The influence of explicit perspective-focus instructions on readers' perspective taking, Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(8), 1407–1423, DOI 10.1037/xlm0000830[7].
We did not obtain it and report the title and citation only.
Three observations, ours.
The title poses the exact question a business reader has, and we cannot tell you the answer, which is the most frustrating gap in this article.
The intervention tested is "explicit perspective-focus instructions," which is the obvious remedy: telling someone to consider the other person's viewpoint.
And the Fischhoff finding above gives a reason for pessimism we would hold lightly. Participants who knew about the bias still could not defeat it, which is a different task from the one this 2020 paper tested, but points the same way.
Thirty Percent Of What They Know
A striking figure we can report only as a quotation, and we flag it heavily.
A public broadcaster's science article quotes a professor of human-computer interaction and psychology "as saying that experts can articulate only about 30 percent of what they know," noting the problem this creates in designing courses because "the experts creating them often can't adequately explain what they know to the novice learner."[8]
Four observations, ours.
This is a quotation in a media article, not a study. We could not trace it to a paper, an experiment, or a method, and we report it as a claim attributed to a named academic rather than as a finding.
If it were right it would describe a different mechanism from the one this article covers. Not misjudging what others know, but being unable to retrieve what you know yourself.
That distinction matters for remedies. A perspective-taking exercise addresses the first and does nothing for the second, and the second would be addressed by observation rather than by explanation.
And we include it mainly to mark the boundary of what we verified. A memorable number in a reputable outlet is not evidence, and the sourcing standard we applied to the tapping study applies here too.
Why The Weak Version Travelled
An observation about the two findings side by side, which we think is the general lesson. Ours.
The published paper and the unpublished thesis are both about the same phenomenon. One is peer-reviewed in a top journal with a quantified partial correction. The other is a student project nobody can read. It is the second that everybody knows.
Four observations.
The reason is not mysterious. The thesis has a story, a number and a demonstration you can run at a dinner table. The paper has an abstract about asymmetric information and market experiments.
The gap in the numbers points the same way. A twentyfold overestimate is a better anecdote than a fifty percent partial correction, and the more dramatic figure is the less verifiable one.
That correlation is not a coincidence and it is worth naming as a rule. The most quotable version of a finding is systematically the one with the weakest paper behind it, because quotability and methodological caution pull against each other at every step of retelling.
And the fiftieth article in this series found the same structure in eleven separate literatures. By this point we would treat it as the default expectation for any behavioural claim that has reached general circulation, and the test is one search: find the paper, and see whether the famous number is in it.
The Half Nobody Applies
An implication of the finding that runs opposite to how it is normally used. Ours.
Four observations.
The curse is always presented as an expert's problem: the specialist who cannot explain, the founder who assumes too much.
But the mechanism as stated is symmetric. Better-informed agents cannot ignore private information, and the general form is that anyone's estimate of another's knowledge is pulled toward their own state.
Which means it should also operate on the less-informed party, and in a commercially important direction. A client who does not understand what your work involves will underestimate what you know, not merely fail to appreciate it.
We found no study testing that direction and flag it as our own inference. If it holds, the gap between what a professional does and what a client believes they do is produced from both ends at once, which would make it larger than either party's error alone.
What Actually Survives
Our reading, stated directly.
Five statements.
The phenomenon is real and published. A 1989 paper in the Journal of Political Economy, whose abstract we obtained from the paper itself, reporting that better-informed agents cannot ignore private information even when it pays to.
Markets correct about half of it. That is a specific quantified finding and it is the part popular accounts always drop.
Half is not much, on multiplicative errors. Our own arithmetic: a twentyfold overestimate reduced by half is still 950 percent too high.
The famous demonstration is unverifiable and its retellings conflict, on the success rate, the songs and the scale.
And the useful consequence is about planning rather than wording. The error shows up in how many attempts you budget, not in how well you phrase the first one.
What We Would Have Liked To Report
Naming the gaps explicitly, because this article has more of them than most and a reader should know where they are. Ours.
Four things we went looking for and did not find.
The thesis itself. A method section, a sample size, and a description of how songs were chosen would settle every discrepancy between the retellings. We could not obtain any of it.
Any replication of the tapping paradigm. A result this famous, thirty-six years old, ought to have been repeated many times, and we found no record of one. That absence may reflect our search rather than the literature, and we report it as an absence rather than a finding.
The 1989 paper's experiments. We have the conclusion that market forces halve the curse and none of the design that produced it, so we cannot tell you whether the market in question resembles anything a business operates in.
An answer on debiasing. A 2020 paper asks in its title whether the curse can be lifted by explicit perspective-taking instructions. That is the single most useful question for a practitioner and we can report only that somebody asked it.
Two observations.
The pattern in that list is worth noticing. What we obtained is the part that establishes the phenomenon exists; what we could not obtain is every part that would tell you what to do about it.
And that is a fair description of most of this series. Existence claims are published and travel; magnitude and remedy sit behind paywalls and in unpublished theses, which is why so many of these articles end with a procedure rather than a prescription.
Explaining Your Price
The first application. Ours, untested, and not pricing or marketing advice.
Four points.
You know why your price is what it is. You know the hours, the risk, the things that go wrong, and the work nobody sees, and the finding says you cannot reconstruct a mind that does not.
The 1989 paper's own example is exactly this: "sellers are better informed about the true value of their products than buyers"[1]. The curse is that you also cannot tell how the price looks to someone who is not.
The predictable failure is a quote that states the number and omits the reasoning you consider obvious, because obvious is precisely the property the curse assigns wrongly.
And the practical test costs nothing. Ask someone outside the trade to read the quote back to you and say what they think they are buying. Their answer is data about the quote, not about them.
Estimates And Instructions
The second application, and the more expensive one. Ours.
Four points.
Whoever writes an instruction is, by construction, the person who least needs it. That is not a coincidence; it is the selection rule.
The same holds for time estimates. The estimate is usually made by whoever knows the work best, and the finding says that person cannot simulate someone who does not know it.
The direction of the error is predictable. Toward underestimating the time, the questions, and the number of things that need saying, because all of them are invisible from inside the knowledge.
And this compounds with the planning fallacy, which this publication has covered separately. Two independent reasons to underestimate, both operating on the same person at the same moment, which is our own inference rather than anything either literature claims.
Your Website And Your Documentation
The third application, and the one with the widest reach. Ours.
Four points.
Every page a firm writes about itself is written by someone who already knows what the firm does. The curse predicts systematic overestimation of how much a reader will pick up.
The failure mode is not bad writing. It is unexplained assumptions that feel like context rather than jargon, and they are invisible to the author by definition.
The cheapest diagnostic we know is a comprehension test rather than a readability one. Give the page to someone outside the trade and ask three questions about what it said, and count wrong answers rather than collecting opinions.
And the reason to count rather than ask is the finding itself. Asking whether it was clear invites a judgment you cannot calibrate, while a wrong answer to a factual question is unambiguous.
One caution on running that test, which we would apply to ourselves. A single reader who misunderstands one page is four observations at best, and the sixty-sixth article set out what a rate built on four observations is worth. Two readers is not a study; it is a smoke alarm.
So the right use of it is directional and repeated rather than statistical. Run it on every substantial page over a year, keep a tally of which questions get missed, and treat a question that repeatedly fails as the finding rather than treating any single reader as one.
The One Fix We Would Trust
Because everything above says trying harder will not work. Ours.
Four observations.
The Fischhoff finding, as reported, is that people who knew about the bias could still not ignore or defeat it. If that holds here, awareness is not a remedy.
So the remedy has to bypass the reconstruction rather than improve it. Do not try to imagine what the other person knows. Find out.
In practice that is one habit: test comprehension on a real person outside the knowledge, before the thing goes out, and treat the result as a measurement rather than as feedback.
And that is the same instrument this series has now recommended in eight articles for eight different failures. Replace an introspective estimate with an external observation, which works precisely because it does not require the introspection to improve.
And The Cost Of Overcorrecting
The section this article needs, because it would otherwise recommend explaining everything to everyone. Ours.
Four observations.
Over-explaining has real costs. It is slow, it reads as condescending, and it buries the point a knowledgeable reader came for, which is a familiar failure in professional writing.
The finding does not say to assume ignorance. It says your estimate of what others know is biased toward your own state, which for an expert means correcting downward and, in principle, for a novice means correcting upward.
That second half is rarely mentioned and follows from the same mechanism. Someone who does not know something should also overestimate how many others share their state, which is the false consensus pattern and which we did not find tested here.
And the honest resolution is not a rule about how much to explain. It is that the right amount is a fact about your audience which you should measure rather than infer, in either direction.
What To Do
Treat your estimate of what others know as biased, not merely uncertain. The 1989 finding is that better-informed agents cannot ignore private information even when it is in their interest to do so.
Do not take comfort from partial corrections. On our own arithmetic, halving a twentyfold error leaves one that is still 950 percent too high.
Fix the plan, not the wording. The error shows up in how many attempts and channels you budget, and the direction is always to under-plan.
Count wrong answers rather than collecting opinions. Ask three factual questions about what a page or a quote said, because asking whether it was clear invites a judgment you cannot calibrate.
Notice who writes your instructions. By construction it is the person who least needs them, which is a selection rule rather than an accident.
Do not over-explain by default. The finding says your estimate is biased toward your own state, not that your audience knows nothing, and over-explanation has its own costs.
Check the sourcing of the tapping study before quoting it. It is an unpublished 1990 dissertation, every source we found is a blog or a popular book, and the retellings disagree on the success rate, the songs and the scale.
Quote the published finding instead. That better-informed agents cannot set aside what they know, and that market forces halve the effect without removing it, are both in a peer-reviewed paper.
The Limits Of This Analysis
Several caveats matter, and one of them concerns the most famous thing in this article. This article discusses research on communication and judgment and is not marketing, pricing or project management advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We obtained the 1989 paper's abstract, opening paragraph and acknowledgements from a copy hosted by the second author's own university department, corroborated word for word by an economics database, and we did not obtain its experiments or results, so we cannot tell you what was traded in the market experiments or how the fifty percent figure was computed. For the 1990 tapping study we obtained nothing academic whatever. Its figures reach this article through two personal newsletters and two popular books quoted on a reader-notes site, all flagged at every use; the underlying work is an unpublished doctoral thesis we did not read, and the four accounts disagree on the success rate, the songs used and the scale of the study. We grade those figures D and would not build anything on them, and we compute with them only to show what would follow if they were right. The account of the paper's motivation and its relationship to earlier work comes from an encyclopedia, not an academic source, flagged at every use. We did not obtain the 1975 hindsight paper and report it through that encyclopedia's description. We did not obtain the 2020 debiasing study, which asks the exact question a reader would want answered, and report its title alone. The thirty percent figure is a quotation attributed to a named academic in a broadcaster's article, not a study, and we could not trace it to any method. All arithmetic is ours. The overestimate factors and the halving table are exact. The table of attempts needed for comprehension assumes each attempt succeeds independently, which we believe is false, since a second explanation is usually better than the first; the real figures are lower, possibly much lower, and we would not quote ninety-one as a target to anyone. And the connection we draw between this and the planning fallacy is our own inference rather than anything either literature claims.
Frequently Asked Questions
What is the curse of knowledge?
Do markets fix it?
Is the tapping study reliable?
What do the retellings disagree about?
Can awareness fix it?
What should I actually change?
Should I explain more?
References
- Camerer, C., Loewenstein, G., & Weber, M. (1989). The Curse of Knowledge in Economic Settings: An Experimental Analysis. Journal of Political Economy, 97(5), 1232–1254. Copy of the paper hosted by the second author's own university department, reproducing its abstract in full: on better-informed agents being assumed, in economic analyses of asymmetric information, capable of reproducing the judgments of less-informed agents; on the authors discussing a systematic violation of that assumption which they call the curse of knowledge; on better-informed agents being unable to ignore private information even when it is in their interest to do so, and more information not always being better; on the authors comparing judgments made in individual-level and market experiments and finding that market forces reduce the curse by approximately 50 percent but do not eliminate it; and on implications for bargaining, strategic behavior by firms, principal-agent problems, and choice under uncertainty being discussed. The same copy reproduces the paper's opening, that in many economic transactions some agents know more than others, with sellers better informed about the true value of their products than buyers and workers knowing more about their ability and motivation than prospective employers; together with its author affiliations and funding acknowledgements. Note: a copy of the paper hosted by a co-author's own university department, which is the best provenance short of the publisher. We obtained the abstract, opening and front matter, and not the experiments or results. cmu.edu
- Economics database record for the same paper, confirming the citation as Camerer, Colin and Loewenstein, George and Weber, Martin, 1989, The Curse of Knowledge in Economic Settings: An Experimental Analysis, Journal of Political Economy, University of Chicago Press, volume 97(5), pages 1232–1254, October, DOI 10.1086/261651, and reproducing the abstract word for word as it appears in the paper itself. Note: a bibliographic database record, used to confirm the citation, DOI and month independently and to corroborate the abstract text against the hosted copy. ideas.repec.org
- Personal newsletter article on novices and experts, describing a study conducted by Elizabeth Newton for her 1990 Stanford University thesis, The Rocky Road From Actions to Intentions: recording that participants were seated back to back, that one was given a list of well-known songs including Twinkle Twinkle Little Star and tasked with tapping out the rhythms on the desk using an index finger while the other identified the songs from the tapping; that the tappers expected the listeners to identify the song correctly about half the time while the listeners managed 2.5 percent; and that while the tappers heard the entire song in their minds, complete with lyrics, melody and instrumental accompaniments, the listeners were exposed only to the rhythmic pattern. Note: a personal newsletter, not an academic source, flagged at every use. One of four non-academic sources for a study we could not obtain; the underlying work is an unpublished doctoral thesis. jonathanboymal.substack.com
- Two further personal newsletter articles describing the same study, one recording that Newton assigned people the roles of tapper or listener, that tappers picked easy songs such as Happy Birthday or Jingle Bells, that Newton asked tappers to predict the chance of listeners guessing correctly and all predicted 50 percent, and asserting that the experiment was done with thousands of people across the world and with almost 120 songs, of which only 2.5 percent were correctly guessed; and the other recording that in a sample of 120 melodies listeners were correct only 2.5 percent of the time while tappers had predicted a 50 percent success rate. Note: personal newsletters, not academic sources, flagged at every use. Recorded specifically to document the disagreement between accounts, in the songs named and in the assertion that the study involved thousands of people worldwide, which describes something other than a doctoral thesis. 1minutestories.substack.com
- Encyclopedia entry on the curse of knowledge, recording that the term was coined in a 1989 Journal of Political Economy article by economists Colin Camerer, George Loewenstein and Martin Weber; that the aim of their research was to counter conventional assumptions in economic analyses of asymmetric information that better-informed agents can accurately anticipate the judgement of less-informed agents; that such research drew from Baruch Fischhoff's 1975 work on hindsight bias; that Fischhoff's research revealed participants did not know their outcome knowledge affected their responses and, if they did know, could still not ignore or defeat the effects of the bias; that participants could not accurately reconstruct their previous, less knowledgeable states of mind; that Fischhoff questioned the failure to empathize with ourselves in less knowledgeable states and noted that how well people reconstruct the perceptions of lesser informed others is a crucial question for historians and all human understanding; that most theoretical analyses of asymmetric information focused instead on how the lesser-informed party attempted to learn more; that the idea that better-informed parties may suffer losses in a deal was seen as important to bring to economic theory; and that a 1990 experiment by a Stanford University graduate student, Elizabeth Newton, illustrated the concept. Note: an encyclopedia, not an academic source, flagged at every use. Our source for the paper's stated motivation and for its relationship to the 1975 hindsight work, neither of which we obtained directly. en.wikipedia.org
- Reader-notes pages on a book-cataloguing site, reproducing passages from two popular books: one recording that in Newton's study the tappers predicted the listeners would guess the tune half of the time but the listeners correctly guessed about 3 percent of the songs; and the other recording that tappers have been given knowledge, the song title, that makes it impossible for them to imagine what it is like to lack that knowledge, that once we know something we find it hard to imagine what it was like not to know it, and that it becomes difficult to share knowledge because we cannot readily re-create our listeners' state of mind. Note: popular books quoted on a reader-notes site, not academic sources, flagged at every use. Recorded specifically to document the discrepancy between the 3 percent figure given here and the 2.5 percent given in three other sources. goodreads.com
- Reference list carried on an academic paper about predicting others' knowledge, confirming Camerer, C. F., Loewenstein, G., and Weber, M. (1989), Journal of Political Economy, 97, 1232–1254, DOI 10.1086/261651; Damen, D., van der Wijst, P., van Amelsvoort, M., and Krahmer, E. (2020), Can the curse of knowing be lifted? The influence of explicit perspective-focus instructions on readers' perspective taking, Journal of Experimental Psychology: Learning, Memory, and Cognition, 46(8), 1407–1423, DOI 10.1037/xlm0000830; and Dawes, R. M. (1989), Statistical criteria for a truly false consensus effect, Journal of Experimental Social Psychology, 25. Note: a reference list carried on an academic paper; citations only. Our source for the 2020 debiasing study, which we did not obtain and whose title poses the exact question this article cannot answer. ouci.dntb.gov.ua
- Public broadcaster science article on lifting the curse of expertise, quoting a professor of human-computer interaction and psychology at Carnegie Mellon University as saying that experts can articulate only about 30 percent of what they know, and noting this is a problem when designing courses because the experts creating them often cannot adequately explain what they know to the novice learner. Note: a broadcaster's article quoting a named academic. This is a quotation, not a study; we could not trace it to any paper, experiment or method, and report it as an attributed claim rather than a finding. pbs.org
This article discusses research on communication and judgment and is not marketing, pricing or project management advice. The 1989 paper was obtained as an abstract, opening and front matter from a copy hosted by a co-author's university department, corroborated by an economics database; its experiments and results were not obtained. The 1990 tapping study was not obtained in any form, its figures reach this article only through non-academic sources, the underlying work is an unpublished doctoral thesis, and the accounts conflict on the success rate, the songs and the scale. The thirty percent figure is a quotation in a broadcaster's article, not a study. All arithmetic is the authors' own; the table of attempts needed for comprehension assumes independence between attempts, which the authors believe is false and explain in the body.