Ellsberg paradox
People shun bets with unknown odds, breaking expected utility.
What it means
The Ellsberg paradox is a thought experiment showing that people prefer betting on events with known probabilities over events with unknown ones so consistently that their choices cannot be reconciled with any single set of subjective probabilities. In the classic urn, knowing there are 30 red balls but an unknown split of 60 black and yellow, people bet on red over black yet also on 'not red' over 'not black' — a combination that no coherent probability assignment can justify. The mechanism is ambiguity aversion: vagueness about the odds is itself aversive, over and above the risk involved. The result was a foundational blow to the assumption that uncertainty can always be captured by subjective probabilities, and it spurred a family of decision theories — maxmin expected utility, Choquet expected utility, and smooth-ambiguity models — built to accommodate it. It matters because so many real decisions, from novel investments to emerging health threats, involve genuinely unknown rather than merely risky odds, and the paradox shows standard theory systematically mispredicts behavior there.
Why it happens
The pull is not a miscalculation but a feeling: not knowing the odds is itself aversive, distinct from knowing the odds are bad. People will pay to avoid the vague bet even when told the two urns are, on average, identical. Brain imaging tracks the split. When choices are ambiguous rather than merely risky, activity rises in the amygdala and orbitofrontal cortex — regions tied to threat and emotional appraisal — and falls in a striatal system that codes expected reward. Patients with orbitofrontal lesions are strikingly insensitive to the level of ambiguity, treating the vague and the known urn alike. The evidence suggests ambiguity recruits a partly different circuit than risk, which is why the two cannot be folded into a single number.
How the theories absorb it
Rather than call the pattern irrational, decision theorists rebuilt the model around it. Maxmin expected utility holds a whole set of candidate probabilities instead of one, then evaluates each act by its worst case across that set — so a vague bet, whose downside is unpinned, looks pessimistically bad. Choquet expected utility replaces additive probability with a non-additive capacity, letting the weights on outcomes fail to sum in the usual way. Smooth-ambiguity models go further, separating what you believe about the odds from how much the vagueness bothers you, so belief and attitude can be measured apart. Each abandons the single subjective prior that Ellsberg's two bets jointly destroy, and each reduces to standard expected utility when the ambiguity vanishes.
What the evidence shows
The core pattern is real and repeatable, but not universal or uniform. Reviews of the experimental record find ambiguity aversion is the modal response for bets with moderate-to-high chances of winning, yet it flips to ambiguity seeking for low-likelihood bets and often in the loss domain, where the unknown becomes appealing. A sizable minority of subjects are ambiguity neutral throughout. Halevy's controlled replication uncovered a deeper link: the people who show the Ellsberg pattern are largely the same people who fail to reduce compound objective lotteries, suggesting the paradox rides on how we handle layered uncertainty rather than on ambiguity alone. Attitudes also vary with the source of uncertainty — a familiar domain feels less ambiguous than an artificial urn.
Competing explanations and limits
Not every avoidance of the vague urn is ambiguity aversion. Some subjects suspect the experimenter stacked the unknown urn against them, so declining is a sensible hedge against a possibly rigged game, not a preference over odds. Others may simply find the ambiguous option more complex to reason about; experiments that hold complexity fixed show it accounts for part, though not all, of the effect, mostly among less sophisticated participants. And the response is malleable: when people are taught the logic of the paradox and shown why their two bets conflict, ambiguity aversion shrinks but does not disappear. That residue matters — it says the tendency is more than confusion, but the raw laboratory number overstates how much is a stable taste for the known.
Examples
Given an urn with 30 red balls and 60 unknown black/yellow, people bet on red over black and on 'not-red' over 'not-black' — a contradiction.
Savers pile into a familiar fund with a long track record over a new market nobody can price, while conceding the unknown one might well do better — the vagueness itself repels.
Insurers happily price hurricanes off a century of records, yet demand a steep premium, or refuse outright, to cover a novel cyber risk whose odds nobody can pin down.
A patient weighing a proven therapy with charted survival rates against an experimental trial whose benefit is genuinely unquantified tends to take the known odds, even when the trial's ceiling looks higher.
First described in Daniel Ellsberg (1961).
Key references
- Jia, R., Furlong, E., Gao, S., Santos, L. R., & Levy, I. (2020). Learning about the Ellsberg Paradox reduces, but does not abolish, ambiguity aversion. PLOS ONE, 15(3), e0228782. doi.org/10.1371/journal.pone.0228782
- Trautmann, S. T., & van de Kuilen, G. (2015). Ambiguity Attitudes. In G. Keren & G. Wu (Eds.), The Wiley Blackwell Handbook of Judgment and Decision Making (pp. 89-116). Wiley. doi.org/10.1002/9781118468333.ch3
- Halevy, Y. (2007). Ellsberg Revisited: An Experimental Study. Econometrica, 75(2), 503-536. doi.org/10.1111/j.1468-0262.2006.00755.x
- Hsu, M., Bhatt, M., Adolphs, R., Tranel, D., & Camerer, C. F. (2005). Neural Systems Responding to Degrees of Uncertainty in Human Decision-Making. Science, 310(5754), 1680-1683. doi.org/10.1126/science.1115327
- Gilboa, I., & Schmeidler, D. (1989). Maxmin expected utility with non-unique prior. Journal of Mathematical Economics, 18(2), 141-153. doi.org/10.1016/0304-4068(89)90018-9
- Ellsberg, D. (1961). Risk, Ambiguity, and the Savage Axioms. Quarterly Journal of Economics, 75(4), 643-669. doi.org/10.2307/1884324