The Urn You Cannot See Into

Almost no commercial decision comes with a stated probability. You do not know the chance a new market works, a new hire succeeds, or a supplier fails. This article is about what people do when the odds are unknowable rather than merely unfavourable, and about how much less settled that literature is than its textbook status implies.

The Verdict, Stated First

Five claims, in descending order of confidence.

One. The demonstration is airtight as arithmetic. The preference pattern Ellsberg elicited cannot be held by anyone with a definite belief about the ambiguous urn, and we set out both versions of the proof below.

Two. What produces that pattern is much less settled than the pattern itself. Three separate papers offer three different accounts, and the most recent of them puts ambiguity aversion in second place.

Three. The strongest reinterpretation is contextual. A 1995 paper reports the aversion arises only when an unambiguous alternative is available for comparison, which would make it a property of how choices are presented rather than of anything about uncertainty.

Four. That reinterpretation was partly walked back. A 2001 replication found the aversion persists in both conditions, though larger when compared, so the direction survives and the strong version does not.

Five. The commercially useful part is a cost, not a bias. On our own arithmetic, requiring an unfamiliar opportunity to beat a familiar one by four percentage points removes half of the opportunities that would in fact have beaten it.

Our Grades For These Claims

Applying the scheme from the first article in this series.

Grade A for the paradox itself, which is a logical demonstration rather than an empirical finding and which anyone can verify with the arithmetic below.

Grade B for the empirical prevalence of the pattern, because we did not obtain the 1961 paper and no source we found gives us the proportion of people showing it.

Grade B for comparative ignorance, from one paper's abstract as described by a citing source, partially confirmed and partially contradicted by a replication whose abstract we obtained.

Grade A for the 2012 challenge as stated, from an abstract obtained verbatim from two independent sources, though it is one experiment.

Grade A for our own arithmetic, which is elementary probability and reproducible.

Our position: the paradox is real and the explanation for it is genuinely open, which is an unusual combination and one the textbook treatment does not convey.

A Note On Method

Everything here is verified to August 2026.

We did not obtain the 1961 paper. Every statement about it comes from a 2024 scholarly review of it and from citing descriptions[4].

We did not obtain the 1995 paper. Our account of comparative ignorance comes from one working paper's characterisation of it[1] and from the 2001 replication's abstract[2].

We obtained the 2001 replication's abstract verbatim from the publisher, and the 2012 challenge's abstract verbatim from two independent sources[2][3]. We obtained neither paper.

We did not obtain the 1991 competence paper, the 2007 Econometrica study, the 2022 Econometrica paper, or any of the theoretical literature, and report all from citation records and titles.

All arithmetic is ours. The paradox derivations are elementary and follow from the setups as described; the cost calculation uses invented parameters throughout.

This article discusses research on decision-making under uncertainty. It is not investment, strategy or risk management advice.

Who Ellsberg Was

Worth a paragraph, because the biography is not incidental and most treatments omit it.

A 2024 scholarly review records: "Daniel Ellsberg (1931–2023) was an American decision theorist, Marine, military analyst, and, after disclosing the Pentagon Papers in 1971, a political activist." It adds that "he became a consultant for the US Defense Department, the State Department, and the White House, advising on nuclear deterrence and crisis decision-making."[4]

Three observations, ours.

The same person wrote the founding paper on choice under unmeasurable uncertainty and advised on nuclear deterrence, which is the largest real instance of the problem the paper describes.

The review also records where the paper was first presented: the Econometric Society meeting in St. Louis in December 1960[4], so the work predates the publication by a year.

And the sequence is worth noticing. The paper came first and the Pentagon Papers a decade later, so a reader encountering the name in a decision theory context has usually met the second story first.

The 1961 Paper

The source.

Ellsberg, D. (1961), Risk, Ambiguity, and the Savage Axioms, The Quarterly Journal of Economics, 75(4), 643–669, DOI 10.2307/1884324[4][5].

The 2024 review describes what it did: "In that article, Ellsberg envisaged a choice situation, later referred to as the Ellsberg paradox, in which a decision maker has to express her preferences between gambles with uncertain outcomes. The gambles yielded either $0 or $100 depending on the color of a ball drawn from an urn containing balls of different colors. Ellsberg observed that several reasonable decision makers expressed deliberate preferences that violated the axioms of the then-dominant theory of decision-making under uncertainty, the version of expected utility theory advanced by Leonard J. Savage."[4]

Four observations, ours.

The phrase "several reasonable decision makers" is doing a lot of work and is worth pausing on. This was not a survey. Ellsberg canvassed a small number of sophisticated people, and the strength of the demonstration is logical rather than statistical.

The word "deliberate" matters more. The claim is not that people were confused; it is that they held the preferences on reflection, which makes the violation harder to dismiss as an error.

The target is Savage's axioms, specifically the requirement that a decision maker act as though holding some definite probability over unknown states. The paradox says people knowingly decline to.

And we did not obtain the paper, so we cannot tell you how many people, who they were, or how the preferences were elicited. For a demonstration this famous that is a real gap and it recurs below.

The Two-Urn Version

The simplest form, and the one to carry. Our own setting out, following the standard description.

Urn A contains 50 red and 50 black balls. You are told this.

Urn B contains 100 balls, red and black, in a proportion nobody tells you.

You name a colour, draw one ball, and win $100 if it matches.

Two questions. To bet on red, which urn? Most people say A. To bet on black, which urn? Most people say A again.

Three observations.

Both answers are individually reasonable. You know exactly what you are getting from urn A, and from urn B you know nothing.

The two answers together are not reasonable, and the next section shows why in one line.

And note the structure. Nothing about urn B is worse than urn A. It is not stated to have fewer of your colour. It is only unstated.

The Contradiction

Our own arithmetic, and it takes three lines.

Preferring urn A for a bet on red means you judge P(red in B) is less than 0.50.

Preferring urn A for a bet on black means you judge P(black in B) is less than 0.50.

But urn B contains only red and black, so P(red in B) plus P(black in B) equals 1.00, and two numbers each below one half cannot sum to one.

Four observations.

The preferences are not merely cautious. They are arithmetically impossible to hold together by anyone with any definite belief about urn B, which is exactly what Savage's framework requires.

That is what makes the paradox stronger than most findings in this series. It does not depend on an effect size, a sample or a replication. Either you hold both preferences or you do not, and if you do, no probability assignment rationalises you.

The escape available is to say you have no definite belief about urn B at all, which is precisely Ellsberg's point and the reason the paper is about Savage's axioms rather than about psychology.

And this is why the literature that follows is about what to do instead, producing a large theoretical apparatus this article does not attempt to cover.

The Three-Colour Version

The sharper form, which is the one the 2012 experiment used and which closes the obvious loophole. Our own setting out.

The 2012 paper describes its apparatus: "an urn contains ten red balls and another twenty balls of which it is only known that they are either black or white."[6] We use the standard proportions of thirty and sixty.

One urn. 30 red. 60 more that are black or yellow, in an unknown split. So P(red) is exactly one third, and P(black) plus P(yellow) is two thirds.

First pair. Bet I wins on red. Bet II wins on black. Most people take I.

Second pair. Bet III wins on red or yellow. Bet IV wins on black or yellow. Most people take IV.

Three observations.

Taking I over II means you judge P(black) is less than one third.

Taking IV over III means P(black) plus P(yellow) exceeds one third plus P(yellow), which reduces to P(black) is greater than one third.

And those are the same quantity, so the two choices together assert that one number is both above and below one third.

Why The Yellow Balls Matter

The structural point, which is the reason this version is used in preference to the two-urn one. Ours.

Four observations.

In the second pair, yellow appears on both sides. Bet III is red-or-yellow, bet IV is black-or-yellow, so the yellow balls are common to both and cancel from the comparison.

Once yellow cancels, the second comparison is red against black, which is exactly the first comparison. The same choice, offered twice, gets opposite answers.

That is a direct violation of the sure-thing principle, which says a component common to two options should not affect the choice between them. The 2012 paper names that principle among those it tested[3].

And the design forecloses the escape available in the two-urn version. Everything is drawn from one urn, so no story about urns being differently constituted can rescue the pattern.

This Is Not Caution

Distinguishing the finding from the thing it is usually confused with. Ours.

Four observations.

Risk aversion is not at issue. A risk-averse person can hold consistent beliefs; they simply dislike variance. Nothing above involves variance, since both bets pay $100 or nothing.

Nor is it pessimism. A pessimist who believes the ambiguous urn is stacked against them holds a definite belief, and would therefore take the ambiguous urn when betting on the colour they think is over-represented. The observed pattern refuses both colours.

What is at issue is whether a person has any probability distribution at all over the unknown composition. The pattern says they behave as though they do not.

And that distinction is the commercially useful one. A business owner who says a new market is risky is making a claim about variance. One who says nobody can know is making Ellsberg's claim, and the two call for different responses.

The Comment In The Same Volume

An objection published almost immediately, which we report because it is rarely mentioned.

Reference lists identify Raiffa, H. (1961), Risk, ambiguity, and the Savage axioms: comment, Quarterly Journal of Economics, 75, 680–694[7].

We did not obtain it and report the citation only.

Three observations, ours.

The comment appears in the same volume, eleven pages after the paper ends, which means the objection was contemporaneous rather than a later reassessment.

We would very much like to report what it argues and cannot, which is a conspicuous gap given that a named critique by a decision theorist of that standing is likely to be the strongest one available.

And the existence of an immediate published comment is itself informative. This was contested from the day it appeared, not treated as settled and later questioned, which is the more common pattern in this series.

Comparative Ignorance

The reinterpretation that reframed the whole finding.

Fox, C. R., and Tversky, A. (1995), Ambiguity Aversion and Comparative Ignorance, The Quarterly Journal of Economics, 110(3), 585–603, DOI 10.2307/2946693[1][7].

A working paper describes its result: "Fox and Tversky (1995) make an important contribution to understanding behavioral responses to ambiguity. In an individual choice setting they demonstrate that an aversion to ambiguous lotteries arises only when a comparison to unambiguous lotteries is available."[1]

We did not obtain the paper and report this characterisation.

Four observations, ours.

The claim is that the aversion is comparative rather than absolute. Presented with an ambiguous bet alone, people price it much as they price a clear one. Presented alongside a clear one, they discount it.

If correct, this relocates the phenomenon entirely. It becomes a fact about how the choice was displayed rather than about uncertainty, which is the structural reinterpretation pattern the fiftieth article found in eleven separate literatures.

It also connects to a wider literature the same reference lists carry, on joint versus separate evaluation, including Hsee's evaluability hypothesis and a 1999 review of preference reversals between the two modes[2]. We obtained none of it.

And the practical implication is immediate and testable. Whether you compare an unfamiliar option against a familiar one, or assess it on its own, would change the decision, which is a fact about your process rather than about the opportunity.

What That Reinterpretation Does

Why it matters more than a boundary condition normally would. Ours.

Three observations.

Under the original reading, ambiguity aversion is a preference, something a person carries into any decision. Under comparative ignorance it is a context effect, something a decision generates depending on what sits next to it.

Those two readings give opposite advice. A preference is something to correct in the person; a context effect is something to correct in the process, and this series has consistently found the second kind of remedy more reliable.

And it would explain a puzzle the original framing leaves open. Most real business decisions have no unambiguous comparator, since almost nothing comes with stated odds, so under this reading the aversion should be much rarer in the field than in the laboratory.

A Partial Replication

Which is where the strong version runs into trouble.

Chow, C., and Sarin, R. (2001), Comparative Ignorance and Ambiguity Aversion, Journal of Risk and Uncertainty, 22, 129–139[2].

Its abstract: "We investigate the evaluation of known (where probability is known) and unknown (where probability is unknown) bets in comparative and non-comparative contexts. A series of experiments support the finding that ambiguity avoidance persists in both comparative and non-comparative conditions. The price difference between known and unknown bets is, however, larger in a comparative evaluation than in separate evaluation. Our results are consistent with Fox and Tversky's (1995) Comparative Ignorance Hypothesis, but we find that the strong result obtained by Fox and Tversky is more fragile and the complete disappearance of ambiguity aversion in non-comparative condition may not be as robust as Fox and Tversky had supposed."[2]

Four observations, ours.

This is a partial confirmation stated precisely, and the precision is the valuable part. Comparison amplifies the effect, which supports the hypothesis. The effect does not vanish without comparison, which contradicts the strong version.

The phrase "may not be as robust as Fox and Tversky had supposed" is measured language for a direct contradiction of a headline result in a top journal.

So the state of play after 2001 is: ambiguity aversion is real without comparison and larger with it. Both halves matter and summaries usually keep one.

And we obtained no effect sizes from either paper, so we cannot tell you how much larger.

The 2012 Challenge

The paper that puts the whole phenomenon in second place.

Binmore, K., Stewart, L., and Voorhoeve, A. (2012), How much ambiguity aversion? Finding indifferences between Ellsberg's risky and ambiguous bets, Journal of Risk and Uncertainty, 45(3), 215–238, December[3][8].

Its abstract: "Experimental results on the Ellsberg paradox typically reveal behavior that is commonly interpreted as ambiguity aversion. The experiments reported in the current paper find the objective probabilities for drawing a red ball that make subjects indifferent between various risky and uncertain Ellsberg bets. They allow us to examine the predictive power of alternative principles of choice under uncertainty, including the objective maximin and Hurwicz criteria, the sure-thing principle, and the principle of insufficient reason."[3]

We did not obtain the paper.

Four observations, ours.

The method is the contribution. Rather than asking which bet you prefer, it finds the objective probability at which you are indifferent, which converts a binary choice into a measured quantity.

That is a considerable improvement on the original design. A preference tells you a direction; an indifference point tells you a magnitude, and the title asks exactly that question.

The paper then treats the competing accounts as rival predictive theories and compares them, which nobody can do with binary preference data alone.

And the first sentence is careful in a way worth noticing. Behavior "commonly interpreted as" ambiguity aversion flags the interpretation as an assumption before testing it.

Contrary To Our Expectations

The result, and it is the most striking sentence in this article.

"Contrary to our expectations, the principle of insufficient reason performed substantially better than rival theories in our experiment, with ambiguity aversion appearing only as a secondary phenomenon."[3]

Four observations, ours.

"Contrary to our expectations" is the authors reporting that they were wrong about what they would find, in the abstract, unprompted. This series has praised that behaviour repeatedly and it remains rare.

"Substantially better than rival theories" is not a marginal result. The winner was not one of the ambiguity models but the simplest available alternative.

"Only as a secondary phenomenon" concedes the effect exists and demotes it. That is the same shape as the sixty-third article's one-eighth finding: not a refutation but a bounding.

And this is one experiment, which we grade accordingly. A single well-designed study reporting a surprise is a reason to hold a finding loosely, not to discard it.

The Principle Of Insufficient Reason

What actually won, explained, because the name is opaque and the idea is not. Ours.

The principle says that when you have no reason to think one possibility more likely than another, treat them as equally likely. Applied to the ambiguous urn, it says: assume fifty-fifty and get on with it.

Four observations.

It is the oldest and simplest answer to the problem, considerably older than any of the theories it beat, and it is what a working manager would do without being told to.

Under it, the two urns are equivalent and the paradox does not arise. So a population behaving this way would show no ambiguity aversion at all, which is roughly what the paper reports.

It has real philosophical difficulties this article will not litigate, chiefly that the answer depends on how you carve up the possibilities. We flag that rather than pretend the principle is unproblematic.

And its practical form is the useful part. Faced with genuinely unknown odds, put your best flat estimate on the table and proceed, which is advice that requires no decision theory and which one experiment says describes what people mostly do anyway.

And A 2022 Problem Of Observability

A recent methodological development we can report only as a title.

Reference lists identify Baillon, A., Halevy, Y., and Li, C. (2022), Randomize at your own risk: on the observability of ambiguity aversion, Econometrica, 90(3), 1085–1107[9].

We did not obtain it and report the title and citation only.

Three observations, ours.

The subtitle raises the question of whether ambiguity aversion is observable at all given standard experimental procedures, which if answered negatively would bear on every study in this article.

The same authors appear elsewhere in the reference lists on experimental elicitation of ambiguity attitude using the random incentive system[9], so this is a sustained methodological programme rather than an isolated objection.

And we will not speculate about what it concludes. A title is not a finding, and we name it so a reader knows the question is live in 2022 rather than settled in 1961.

What Actually Survives

Our reading, stated directly.

Five statements.

The paradox is logically airtight. The preference pattern cannot be rationalised by any definite probability assignment, and that does not depend on any empirical result.

How common the pattern is, we cannot tell you. The original canvassed "several reasonable decision makers" and we obtained no prevalence figure from any source.

Comparison amplifies the effect and is not necessary for it. The 1995 strong claim was partly walked back in 2001, with the direction surviving.

One good experiment puts it second. The 2012 paper found the principle of insufficient reason predicted substantially better, with ambiguity aversion secondary.

And the methodology is under active challenge, with a 2022 Econometrica paper questioning observability.

What The Three Explanations Share

A pattern across the disagreement, which we think is the most useful thing to take from a literature this unsettled. Ours.

Three accounts have been offered. Comparative ignorance says the effect appears when an unambiguous alternative sits beside the ambiguous one. Competence says it depends on how much the chooser knows relative to what is knowable. Insufficient reason says the effect is mostly absent and people assume an even split.

Four observations.

None of the three treats ambiguity aversion as a stable trait of the chooser. Two make it a property of the situation and the third largely denies it. That is a striking amount of agreement inside a disagreement.

All three therefore point at the same class of remedy. If the effect is situational, change the situation, which means the presentation of the choice rather than the psychology of the person making it.

And that convergence is the thing to act on, because it is robust to which account turns out to be right. Evaluating an unfamiliar option on its own terms helps under comparative ignorance, is neutral under the competence account, and costs nothing under insufficient reason.

The fifty-second and fifty-seventh articles reached the same structural conclusion by different routes. Where a literature disputes the mechanism but agrees the phenomenon is contextual, the process fix survives the dispute and the psychological fix does not.

The Ambiguity Premium

Translating the finding into the commercial quantity that matters. Ours.

Whatever produces it, the behaviour is a premium: an unfamiliar option must beat a familiar one by some margin before it is chosen. Three observations.

That premium is invisible in most decision documents, because it operates as a decision rather than a line item. Nobody writes down "we required an extra four points because we did not know."

It is also not always wrong, and a section below sets out when it is correct. Unknown odds sometimes genuinely accompany worse odds.

But it is measurable in principle. The premium is the gap between what you would accept from a known opportunity and what you require from an unknown one, and that gap is a number you can name.

What It Costs You

Our own arithmetic, invented parameters throughout, demonstrating a shape rather than measuring anything.

Suppose your familiar option returns 10 percent, and unfamiliar opportunities have true expected returns spread evenly between 6 and 18 percent. You require a premium before accepting an unfamiliar one.

With no premium: you accept 66.7 percent of opportunities, and those you accept average 14.0 percent.

With a 2 point premium: you accept 50.0 percent, averaging 15.0 percent, and you decline 16.7 percent of the whole set that would have beaten your familiar option.

With 4 points: you accept 33.3 percent, averaging 16.0 percent, and forgo 33.3 percent of the set that would have beaten it.

With 6 points: you accept 16.7 percent and forgo 50.0 percent of the set that would have beaten it.

Four observations.

Since 66.7 percent of the whole set beats your familiar option, a four point premium discards exactly half of the opportunities that would in fact have been better.

The quality of what you do accept rises, from 14.0 to 16.0 percent, which is why the premium feels like it is working. Every accepted opportunity looks good and the rejected ones are never observed.

That is the fifty-eighth article's survivorship problem arriving inside a decision rule. A filter that removes good and bad together will always look successful from the inside.

And at an 8 point premium you accept nothing at all, which is the reductio: a large enough ambiguity premium is indistinguishable from a policy of never trying anything new.

New Markets

The first application. Ours, untested, and not strategy advice.

Four points.

A new market is the canonical ambiguous urn. You cannot state the probability of success and nobody can supply one, which is the exact structure of the finding.

The home bias article in this publication already identifies ambiguity aversion as one mechanism behind Canadian owners over-concentrating domestically, and this literature is where that mechanism comes from.

The useful discipline is to name the premium. If you would fund a domestic expansion at a projected 12 percent, ask what number a foreign one would need, and write both down. The gap is your premium and it is now a decision rather than a reflex.

And the 2012 result suggests a default. Put a flat estimate on the unknown and proceed, which is the principle of insufficient reason and which that experiment found describes most behaviour better than ambiguity aversion does.

New Suppliers And New Technology

The second application, and the more frequent one. Ours.

Four points.

An incumbent supplier has a known failure rate, being whatever you have observed. A new one has an unknown one, and the finding says you will overweight that difference.

This is the mechanism behind the fifty-fifth article's status quo finding, arriving from a different direction. That article framed the cost as an unwritten switching cost; this one supplies a reason the cost gets inflated.

The check is arithmetic and cheap. Your incumbent's failure rate is an estimate from a small sample too, and if you have used them for eleven months your confidence interval on their reliability is wider than it feels.

And that is the general form of the correction. The familiar option's odds are usually less known than they appear, so part of the perceived gap between known and unknown is itself an illusion.

Do Not Put Them Side By Side

The process implication, if comparative ignorance holds even partially. Ours.

Four points.

The 1995 finding says the aversion arises when an unambiguous option is available for comparison; the 2001 replication says the difference is larger in comparative than in separate evaluation. Both agree comparison amplifies.

So the format of your decision affects its outcome. Assessing a new market beside a well-modelled existing one is the comparative condition, and it is how almost every board paper is written.

The alternative is separate evaluation: assess the unfamiliar option against a threshold on its own terms, before it meets the familiar one on a page.

And we would flag the limits of this advice honestly. It rests on a 1995 finding we did not obtain, partly contradicted by a 2001 replication we did, and the underlying phenomenon was demoted to secondary by a 2012 experiment. Treat it as a cheap thing to try rather than an established remedy.

The Competence Angle

A related account we can name but not evaluate.

Reference lists identify Heath, C., and Tversky, A. (1991), Preference and Belief: Ambiguity and Competence in Choice under Uncertainty, Journal of Risk and Uncertainty, 4(1), 5–28, DOI 10.1007/BF00057884[7].

We did not obtain it and report the citation and title.

Three observations, ours.

The title pairs ambiguity with competence, which suggests the relevant variable is not how much is knowable but how much the chooser feels they know relative to it.

If so, the prediction differs from plain ambiguity aversion in a testable way. An expert should accept ambiguity in their own field and avoid it elsewhere, which is a different pattern from avoiding it everywhere.

And that would connect directly to the sixtieth article's four cues, one of which is familiarity. We flag the connection and do not assert it, since we obtained neither paper.

When The Aversion Is Correct

The section this article needs, because most of it reads as a case against caution. Ours.

Four observations.

Unknown odds often correlate with worse odds, and for a specific reason: if the odds were good and knowable, someone would usually have measured and advertised them. Absence of information is sometimes information.

Ambiguity also raises the chance of ruin rather than merely loss, and a business that cannot survive the bad tail should decline it regardless of expected value. That is not a bias; it is solvency.

The urn experiments deliberately strip both of those out. Nobody is hiding anything about urn B and nobody goes bankrupt, which is what makes them a clean test and also what limits their transfer.

And the honest statement is therefore conditional. An ambiguity premium is an error when the unknown odds are genuinely symmetric and the downside is survivable, and a sensible policy otherwise. Those two situations look identical from inside, which is the actual difficulty.

What To Do

Learn the contradiction, not the label. Preferring the known urn for red and for black implies two probabilities each below one half summing to one, which is impossible. That is the whole finding in a line.

Name your ambiguity premium in numbers. Ask what return an unfamiliar option would need against a familiar one you would fund, and write the gap down. It exists whether or not you state it.

Notice that the premium filters good and bad together. On our own arithmetic a four point premium discards half the opportunities that would have beaten your familiar option, while raising the average quality of what you accept, so it always looks like it is working.

Try evaluating the unfamiliar option on its own first. Comparison amplifies the effect on both the 1995 and 2001 accounts, and board papers are written in the comparative format by default.

Check how known the known option really is. An incumbent supplier judged on eleven months of data is not the fifty-fifty urn, and part of the perceived gap is an illusion.

Consider the flat estimate. The 2012 experiment found the principle of insufficient reason predicted substantially better than the ambiguity models, and it amounts to assuming an even split and proceeding.

Distinguish ambiguity from ruin. Declining an unknown bet you could not survive is solvency, not bias, and the urn experiments deliberately exclude that case.

Hold the explanation loosely. Three papers offer three accounts, the most recent puts ambiguity aversion in second place, and a 2022 Econometrica paper questions whether it is observable at all.

The Limits Of This Analysis

Several caveats matter, and the sourcing here is thin relative to the topic's stature. This article discusses research on decision-making under uncertainty and is not investment, strategy or risk management advice; the applications are our own reasoning and untested. Everything is verified to August 2026. We did not obtain the 1961 paper, and every statement about it comes from a 2024 scholarly review and from citing descriptions; in particular we obtained no prevalence figure, so we cannot tell you what proportion of people show the pattern, and the original is described as canvassing "several reasonable decision makers" rather than running a survey. We did not obtain the 1995 comparative ignorance paper, which is the most consequential reinterpretation here, and report it through one working paper's characterisation plus the 2001 replication's description of it. We obtained the 2001 replication and the 2012 challenge as abstracts only, and no effect sizes from any paper in this article. We did not obtain the 1961 published comment by a decision theorist of considerable standing, which is likely the strongest available objection and which we can name but not report. We did not obtain the 1991 competence paper, the 2007 or 2022 Econometrica papers, or any of the joint-versus-separate evaluation literature, and report all from citations and titles; a title is not a finding and we have flagged each as such. The 2012 result is one experiment, and we would not treat a single study as settling a sixty-five-year-old question in either direction. All arithmetic is ours. The paradox derivations are elementary and follow from the setups as standardly described, though we did not verify the exact proportions against the 1961 paper and one source describes a variant with ten red and twenty unknown rather than thirty and sixty. The cost calculation uses entirely invented parameters, including a uniform distribution of returns between 6 and 18 percent, and demonstrates a shape rather than measuring anything. And the urn experiments deliberately exclude two features of real commercial ambiguity, being information asymmetry and the possibility of ruin, which limits how far any of this transfers to a business decision.

Related Reading

No. 66 The Cell You Remember Two groups, identical ratios, zero correlation. People see a relationship anyway. And in your own data the most memorable cell is the least reliable one. Behavioural Finance The Cost Nobody Writes Down The same premium, arriving as an unwritten switching cost. Behavioural Finance Home Bias And Concentration Where this mechanism shows up in Canadian owners' actual portfolios.

Frequently Asked Questions

What is the Ellsberg paradox?

Given an urn with 50 red and 50 black balls and another with 100 balls in an unknown mix, most people prefer the known urn whether betting on red or on black. That implies both probabilities in the unknown urn are below one half, and they must sum to one. The preferences cannot both be held by anyone with a definite belief.

How is that different from risk aversion?

Risk aversion is a dislike of variance, and a risk-averse person can hold perfectly consistent beliefs. Both bets here pay $100 or nothing, so variance is identical. What is at issue is whether the chooser has any probability distribution over the unknown urn at all.

Is the effect well established?

The logical demonstration is airtight. The explanation is genuinely open. A 1995 paper said the aversion arises only under comparison; a 2001 replication found it persists without comparison but is larger with it; and a 2012 experiment found the principle of insufficient reason predicted substantially better, with ambiguity aversion appearing only as a secondary phenomenon.

What is the principle of insufficient reason?

When you have no reason to think one possibility likelier than another, treat them as equally likely. Applied to the unknown urn it says assume fifty-fifty and proceed. It is the oldest and simplest answer available, and one experiment found it beat every rival theory it was tested against.

What does an ambiguity premium cost?

On our own arithmetic, with a familiar option at 10 percent and unfamiliar ones spread between 6 and 18, requiring a four point premium discards exactly half the opportunities that would in fact have beaten the familiar one. The average quality of what you accept rises, which is why the filter always looks like it is working.

Is avoiding ambiguity ever right?

Often. Unknown odds sometimes correlate with worse odds, because good knowable odds usually get measured and advertised. And ambiguity raises the chance of ruin rather than mere loss, which a business that cannot survive the bad tail should decline regardless. The urn experiments strip both features out by design.

What is the practical takeaway?

Name the premium in numbers. Ask what return an unfamiliar option would need against a familiar one you would fund, and write the gap down. And check how known the known option really is, since an incumbent judged on eleven months of data is not the fifty-fifty urn.

References

1. Economics working paper on ambiguity in individual choice and market environments, recording that Fox and Tversky (1995) make an important contribution to understanding behavioral responses to ambiguity, and that in an individual choice setting they demonstrate that an aversion to ambiguous lotteries arises only when a comparison to unambiguous lotteries is available; carried on a page whose reference list confirms Craig R. Fox and Amos Tversky, 1995, Ambiguity Aversion and Comparative Ignorance, The Quarterly Journal of Economics, volume 110(3), pages 585–603; Daniel Ellsberg, 1961, Risk, Ambiguity, and the Savage Axioms, The Quarterly Journal of Economics; Larry G. Epstein, 1999, A Definition of Uncertainty Aversion, The Review of Economic Studies, 66(3), 579–608; and Fox, C. R. and Weber, M., 2002, Ambiguity Aversion, Comparative Ignorance, and Decision Context, Organizational Behavior and Human Decision Processes, 88(1), 476–498. Note: an economics working paper. Our only characterisation of the 1995 comparative ignorance result, which is the most consequential reinterpretation in this article and which we did not obtain. ideas.repec.org
2. Chow, C., & Sarin, R. (2001). Comparative Ignorance and Ambiguity Aversion. Journal of Risk and Uncertainty, 22, 129–139. Publisher record reproducing the abstract in full: on the authors investigating the evaluation of known and unknown bets in comparative and non-comparative contexts; on a series of experiments supporting the finding that ambiguity avoidance persists in both comparative and non-comparative conditions; on the price difference between known and unknown bets being larger in a comparative evaluation than in separate evaluation; and on the results being consistent with Fox and Tversky's Comparative Ignorance Hypothesis while finding the strong result obtained by Fox and Tversky to be more fragile, with the complete disappearance of ambiguity aversion in the non-comparative condition possibly not being as robust as Fox and Tversky had supposed. The same record carries a reference list confirming Ellsberg (1961), Quarterly Journal of Economics, 75, 643–669; Heath and Tversky (1991), Journal of Risk and Uncertainty, 4, 5–28; Hogarth and Kunreuther (1988), Risk, Ambiguity, and Insurance, Journal of Risk and Uncertainty, 2, 5–35; Hsee (1996) on the evaluability hypothesis, Organizational Behavior and Human Decision Processes, 67, 247–257; and Hsee, Loewenstein, Blount and Bazerman (1999) on preference reversals between joint and separate evaluations, Psychological Bulletin, 125, 576–590. Note: the publisher's record. We obtained the abstract only and no effect sizes; we obtained none of the works in its reference list. link.springer.com
3. Binmore, K., Stewart, L., & Voorhoeve, A. (2012). How much ambiguity aversion? Finding indifferences between Ellsberg's risky and ambiguous bets. Journal of Risk and Uncertainty, 45(3), 215–238, December. Publisher record reproducing the abstract in full: on experimental results on the Ellsberg paradox typically revealing behavior commonly interpreted as ambiguity aversion; on the experiments reported finding the objective probabilities for drawing a red ball that make subjects indifferent between various risky and uncertain Ellsberg bets; on those experiments allowing examination of the predictive power of alternative principles of choice under uncertainty, including the objective maximin and Hurwicz criteria, the sure-thing principle, and the principle of insufficient reason; and on the authors finding, contrary to their expectations, that the principle of insufficient reason performed substantially better than rival theories in their experiment, with ambiguity aversion appearing only as a secondary phenomenon. Note: the publisher's record. We obtained the abstract only and not the paper; this is one experiment and we grade it accordingly. link.springer.com
4. Scholarly review article (2024) analysing Ellsberg's 1961 paper, recording that in 1961 Daniel Ellsberg published an article titled Risk, Ambiguity, and the Savage Axioms in the Quarterly Journal of Economics which became a seminal contribution to the theory of decision-making under uncertainty; that the paper analyses that classic and situates it within decision-making theory of the 1950s and early 1960s; that Daniel Ellsberg (1931–2023) was an American decision theorist, Marine, military analyst, and, after disclosing the Pentagon Papers in 1971, a political activist; that he became a consultant for the US Defense Department, the State Department, and the White House, advising on nuclear deterrence and crisis decision-making; that in the article Ellsberg envisaged a choice situation, later referred to as the Ellsberg paradox, in which a decision maker expresses preferences between gambles yielding either zero or one hundred dollars depending on the color of a ball drawn from an urn containing balls of different colors; that Ellsberg observed several reasonable decision makers expressing deliberate preferences violating the axioms of the then-dominant expected utility theory advanced by Leonard J. Savage; and that he presented the paper at the Econometric Society meeting held in St. Louis in December 1960. Note: a scholarly review of the 1961 paper. Our principal source for the paper's content and for the biographical detail; we did not obtain the 1961 paper itself. link.springer.com
5. Academic thesis reference list confirming D. Ellsberg, Risk, Ambiguity, and the Savage Axioms, The Quarterly Journal of Economics, volume 75, issue 4, pages 643–669, 1961, DOI 10.2307/1884324; and C. R. Fox and A. Tversky, Ambiguity Aversion and Comparative Ignorance, The Quarterly Journal of Economics, volume 110, issue 3, pages 585–603, 1995, DOI 10.2307/2946693; together with Y. Halevy, Ellsberg Revisited: An Experimental Study, Econometrica, volume 75, issue 2, pages 503–536, 2007, DOI 10.1111/j.1468-0262.2006.00755.x. Note: a thesis reference list, used to confirm citations and DOIs independently. We obtained none of the works named. theses.hal.science
6. Repository copy of the 2012 paper carrying its abstract and a figure caption describing its apparatus, recording that in the version illustrated an urn contains ten red balls and another twenty balls of which it is only known that they are either black or white, and that a ball is chosen at random from the urn, the color of which determines the award of a prize, here taken to be one dollar. Note: a repository copy. Our source for the experimental apparatus; note that the proportions described here differ from the thirty and sixty commonly used in textbook statements of the three-colour version, and our derivation uses the latter. academia.edu
7. Reference list carried on a hosted copy of the 2012 paper, confirming Ellsberg, D. (1961), Risk, ambiguity and the Savage axioms, Quarterly Journal of Economics, 75, 643–669; Raiffa, H. (1961), Risk, ambiguity, and the Savage axioms: comment, Quarterly Journal of Economics, 75, 680–694; Fox, C. and Tversky, A. (1995), Quarterly Journal of Economics, 110, 585–603; Chow, C. and Sarin, R. (2001), Journal of Risk and Uncertainty, 22, 129–139; Ellsberg, D. (2001), Risk, ambiguity, and decision, New York and London, Garland Publishing; Savage, L. (1954), The foundations of statistics, New York, Wiley; and Curley, S. and Yates, J. (1989), An empirical evaluation of descriptive models of ambiguity reactions in choice situations, Journal of Mathematical Psychology, 33, 397–427. Together with a separate publisher reference list confirming Heath, C., and Tversky, A. (1991), Preference and belief: ambiguity and competence in choice under uncertainty, Journal of Risk and Uncertainty, 4, 5–28, DOI 10.1007/BF00057884. Note: reference lists; citations only. Our source for the 1961 published comment and the 1991 competence paper, neither of which we obtained. personal.lse.ac.uk
8. Economics database citation page for the 2007 Econometrica study, listing works citing it and confirming Ken Binmore, Lisa Stewart and Alex Voorhoeve, 2012, How much ambiguity aversion?, Journal of Risk and Uncertainty, Springer, volume 45(3), pages 215–238, December; together with Mohammed Abdellaoui, Peter Klibanoff and Laetitia Placido, 2015, Experiments on Compound Risk in Relation to Simple Risk and to Ambiguity, Management Science, 61(6), 1306–1322. Note: a database citation page, used to confirm the 2012 citation independently including issue and month. ideas.repec.org
9. Academic preprint reference list confirming A. Baillon, Y. Halevy and C. Li, Randomize at your own risk: on the observability of ambiguity aversion, Econometrica, 90(3), 1085–1107, 2022; the same authors on experimental elicitation of ambiguity attitude using the random incentive system, Experimental Economics, 25, 1–22, 2022; K. Binmore, L. Stewart and A. Voorhoeve, How much ambiguity aversion?, Journal of Risk and Uncertainty, 45(3), 215–238, 2012; and G. Charness, E. Karni and D. Levin, Ambiguity attitudes and social interactions: An experimental investigation, Journal of Risk and Uncertainty, 46(1), 1–25, 2013. Note: a preprint reference list; citations and titles only. We did not obtain the 2022 Econometrica paper and report its title, which raises a question this article cannot answer. arxiv.org

This article discusses research on decision-making under uncertainty and is not investment, strategy or risk management advice. The 1961 paper was not obtained and is reported through a 2024 scholarly review; no prevalence figure for the preference pattern was obtained. The 1995 comparative ignorance paper was not obtained. The 2001 replication and 2012 challenge were obtained as abstracts only, and no effect sizes from any paper are reported. The 1961 published comment, the 1991 competence paper and the 2022 Econometrica paper are reported from citations and titles only. All arithmetic is the authors' own; the cost calculation uses entirely invented parameters and demonstrates a shape rather than measuring anything.