Allais paradox
A classic choice pattern that breaks expected utility theory.
What it means
The Allais paradox is a pair of choice problems in which most people's preferences reverse depending on whether a shared outcome is presented as certain or merely highly probable, in direct violation of the independence axiom of expected utility theory. The mechanism behind the reversal is the certainty effect: the move from a sure thing to a near-sure thing is overweighted relative to an equivalent probability change away from certainty, so the certain option exerts an outsized pull. Allais devised it as a deliberate challenge to the then-dominant von Neumann–Morgenstern framework, and it became early, decisive evidence that a descriptive theory of choice was needed rather than a purely normative one. A subtlety is that the 'paradox' is a violation of a normative axiom, not a logical contradiction in the chooser: the preferences are coherent once probabilities are weighted nonlinearly, which is exactly what prospect theory later formalized. It matters because it helped legitimize behavioral models of risk and remains a standard teaching case for why real preferences depart from expected-value reasoning.
How the reversal is constructed
The two problems usually shown are Kahneman and Tversky's rescaling of Allais's original pair, built from one shared block of probability. The first pair offers a certain 2,400 Israeli pounds against a gamble paying 2,500 with probability .33, 2,400 with probability .66, and nothing with probability .01. The second pair strips the common .66 chance of 2,400 from both, leaving 2,500 with probability .33 against 2,400 with probability .34. Independence says a shared component cannot decide between them, so deleting it should leave preferences untouched. It does not. While the block sits there, one option reaches certainty and the other carries a sliver of zero. Remove it and both are gambles; the sliver stops mattering.
What the evidence shows
Kahneman and Tversky put the pair to 72 respondents: 82 percent took the sure 2,400 in the first problem, 83 percent took the higher-paying gamble in the second. Its robustness is another matter. A meta-analysis of 81 experiments from 29 studies finds it fragile and design-dependent: it appears most reliably with high hypothetical payoffs, when the middle outcome sits close to the top one, and when lotteries are shown as probability distributions rather than compound gambles. Under some structures it reverses outright. Nor is it a novice's error: Millroth and colleagues found the paradoxes most prevalent among the most numerate.
Certainty effect, or aversion to zero?
The certainty effect is the standard mechanism and the one the term is usually taught with; a competing account credits the zero. In the first problem the sure option removes any chance of getting nothing; in the second, both options carry a large chance of nothing, so it no longer separates them. Incekara-Hafalir, Kim and Stecher built lotteries that pull the explanations apart and found support for the zero effect, while evidence for certainty itself was weak to nonexistent. The distinction has teeth. If what people flee is the zero rather than the uncertainty, a floor under the downside buys the switch, and certainty as such is not what you pay for.
What Allais actually argued
The paradox is now taught as an empirical refutation of expected utility, which is not what its author had in mind. Allais put his own version to economists at a 1952 Paris conference — 100 million francs certain against a .10 chance of 500 million, with a .89 block of 100 million to strip — as a normative argument: a reasonable person could hold these preferences, and the axiom, not the person, was at fault. Mongin's history traces how decision theorists from the late 1970s onward repurposed it into a behavioral finding about what people do. Read normatively, the paradox attacks independence; read descriptively, it catalogues an error. Allais made the first argument; the second attributes to the chooser an error Allais explicitly denied.
Using it in practice
On either account the lever is the last increment of risk — the guarantee and the removal of the zero coincide in most commercial cases — not risk in general. Moving someone from a high chance to a guarantee buys more than a larger improvement that still leaves a gap. That asymmetry is one reason warranties, money-back windows, guaranteed delivery and fixed-price quotes tend to cost more than the risk they absorb. Partial reassurance does not earn a partial share of the premium, so half a guarantee often is not worth its cost. And the record ties the effect to framing and hypothetical stakes, so a certainty claim that tests well on paper may not survive a live price.
Examples
People take a sure $1M over a gamble for more, yet flip to the gamble when both options become merely probable.
A game-show contestant banks the guaranteed fifty thousand rather than spin for more. Put both prizes behind one extra ball that must drop first, and she happily plays for the bigger sum.
A claimant accepts a certain settlement over a trial worth more on average; once a preliminary ruling leaves both routes uncertain anyway, he decides to fight for the larger award.
A pesticide that removes a poisoning risk entirely commands a premium far above the one that merely halves it. Introduce a background contamination that neither product can eliminate, and the premium collapses — buyers switch to the cheaper bottle even though it still halves the risk it always did.
A buyer pays over the odds for a fixed-price contract rather than cost-plus. Once a materials clause makes both contracts variable anyway, she stops paying and takes the cheaper, riskier deal.
First described in Maurice Allais (1953).
Key references
- Blavatskyy, P., Ortmann, A., & Panchenko, V. (2022). On the experimental robustness of the Allais paradox. American Economic Journal: Microeconomics, 14(1), 143-163. doi.org/10.1257/mic.20190153
- Incekara-Hafalir, E., Kim, E., & Stecher, J. D. (2021). Is the Allais paradox due to appeal of certainty or aversion to zero? Experimental Economics, 24(3), 751-771. doi.org/10.1007/s10683-020-09678-4
- Millroth, P., Nilsson, H., & Juslin, P. (2019). The decision paradoxes motivating Prospect Theory: The prevalence of the paradoxes increases with numerical ability. Judgment and Decision Making, 14(4), 513-533. doi.org/10.1017/S1930297500006161
- Mongin, P. (2019). The Allais paradox: what it became, what it really was, what it now suggests to us. Economics & Philosophy, 35(3), 423-459. doi.org/10.1017/S0266267118000469
- Kahneman, D., & Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263-291. doi.org/10.2307/1914185
- Allais, M. (1953). Le comportement de l'homme rationnel devant le risque: Critique des postulats et axiomes de l'ecole americaine. Econometrica, 21(4), 503-546. doi.org/10.2307/1907921