Allais Paradox¶

An engineered pair of lottery choices whose majority preference pattern (A over B, D over C) violates the independence axiom of expected-utility theory, isolating the certainty effect as the culprit.

Core Idea¶

The Allais paradox is an empirical finding (Maurice Allais, 1953) that most people's stated preferences over a specific pair of lottery choices violate the independence axiom of expected-utility theory. The stimulus works in two steps. In the first choice, most respondents prefer a certain gain of $1M (Option A) to a lottery offering an 89% chance of $1M, a 10% chance of $5M, and a 1% chance of nothing (Option B), indicating that the certainty of $1M is valued above the lottery's higher expected value. In the second choice, over a re-parameterised pair from which the 89% chance of $1M has been removed from both options, the same respondents reverse and prefer a 10% chance of $5M over an 11% chance of $1M (Option D over Option C). The independence axiom requires that adding a common consequence to both lotteries cannot change the preference ordering; removing the common 89% chance of $1M from both options in the first choice converts it into the second, so A preferred to B should imply C preferred to D. The observed majority pattern — A over B and D over C — violates this implication.

The structural source of the violation is the certainty effect: respondents overweight outcomes that are certain relative to those that are merely highly probable, so the certainty of $1M in Option A receives disproportionate weight that disappears when both options in the second choice are uncertain. The Allais paradox is the foundational empirical challenge to expected-utility theory as a descriptive account of choice under risk, and it directly motivated the development of non-expected-utility theories — prospect theory (Kahneman and Tversky, 1979), rank-dependent utility (Quiggin, 1982), and cumulative prospect theory — that accommodate probability distortion and the overweighting of certain outcomes.

Structural Signature¶

Sig role-phrases:

the outcome space — a small set of monetary prizes ($0, $1M, $5M) over which the lotteries are defined
the matched lottery pair — the same prizes presented in two different probability mixings, the second derived from the first by stripping a common consequence
the certainty contrast — a sure outcome present in one member of the first pair but absent once both options in the second pair are merely probable
the independence axiom — the EU requirement that adding (or removing) a common consequence to both lotteries cannot change the preference ordering
the observed reversal — the majority A-over-B-and-D-over-C pattern, which the independence axiom forbids
the certainty effect — the mechanism: a sure outcome is overweighted, the overweighting evaporating once both options are uncertain
the localised axiom failure — the inference that independence specifically fails, leaving transitivity, completeness, and continuity untouched
the repair locus — the missing structure belongs in a probability-weighting function, not in the curvature of the utility-of-money
the normative-descriptive boundary — the reversal refutes EU as a description without settling whether independence fails as a norm

What It Is Not¶

Not explainable by risk aversion. Risk aversion is fully consistent with expected-utility theory — it lives in the curvature of the utility-of-money function. The Allais reversal is a violation of the independence axiom that no degree of utility curvature can produce, which is exactly why relabelling the agent as more or less risk-averse cannot reconcile the A-over-B-and-D-over-C pattern. The repair must go on the probabilities, not the value of money.
Not loss aversion. Loss aversion concerns reference-dependent valuation of gains versus losses. The Allais lotteries are all-gain, and the reversal occurs across probability re-scalings of the same prizes — driven by the certainty effect, not by a gain/loss asymmetry around a reference point.
Not ambiguity aversion. Ambiguity aversion requires unknown probabilities. In Allais the probabilities are fully specified and given; the anomaly is about how known probabilities near certainty are weighted, not about an aversion to unspecified distributions.
Not a temporal preference reversal. The reversal is across probability re-scalings, not across time horizons. It has nothing to do with discounting or present-bias; an experimenter reproducing it manipulates the probability mixing, not the timing of payoffs.
Not a refutation of expected utility as a normative standard. A demonstrated Allais pattern refutes expected utility as a description of how people choose, while leaving open whether the independence axiom fails as a norm of rational choice. The two questions must be kept apart: the violation forces a new descriptive theory, not necessarily abandonment of the normative ideal.
Not a portable mechanism in the two-lottery stimulus itself. The specific lottery pair is a diagnostic instrument, applicable only where a chooser ranks lotteries over known probabilities with a certainty contrast available. What generalises is the certainty-effect / probability-weighting distortion (housed in expected_utility_theory and a probability_weighting construct) and, more loosely, the cognitive-bias point that intuitive preferences can violate normative axioms — not the named Allais stimulus applied past its precondition.

Scope of Application¶

The Allais paradox lives within a single home discipline — decision theory under risk — restaged across its applied subfields; its reach is bounded to settings with a chooser ranking lotteries over known probabilities with a certainty contrast available, and the cross-domain lessons that travel (the certainty-effect / probability-weighting distortion, and more loosely "intuitive preferences can violate normative axioms") are carried by expected_utility_theory / a probability_weighting construct and the cognitive-bias parent, not by the named two-lottery stimulus.

Decision theory and behavioural economics — the home turf: the foundational refutation of expected utility as a descriptive model, motivating prospect theory, rank-dependent utility, and cumulative prospect theory.
Experimental economics and survey methodology — the workhorse stimulus in risk-attitude elicitation, the fixed target any candidate descriptive theory must reproduce.
Descriptive asset pricing — investor-behaviour models permitting independence violations (rank-dependent and cumulative prospect models) that explain asset-pricing puzzles inconsistent with expected utility.
Policy and regulation (descriptive) — evidence that policy-relevant choices over uncertain outcomes will not match the expected-utility-maximising prescription, and health-decision research extending the same finding.

Clarity¶

Naming the Allais paradox converts a vague unease — that real choices under risk seem to disobey the textbook theory — into a sharp, localized diagnosis: it is the independence axiom that fails, and it fails systematically rather than as noise. That precision matters because expected-utility theory has several axioms, and a practitioner who knows the violation is confined to independence (not to transitivity, completeness, or the curvature of the utility function) knows exactly which part of the apparatus to repair. The paradox thereby separates two things often blurred: a normative model of how a rational agent should rank lotteries, and a descriptive account of how people actually rank them. A demonstrated A-over-B-and-D-over-C reversal cannot be reconciled by relabeling the agent as more or less risk-averse, because risk aversion lives in utility curvature and is fully consistent with expected utility — so the anomaly forces the modeler off the utility function and onto the probabilities.

In doing so it makes a productive question askable: what is the mechanism by which certainty is overweighted? By pointing at the certainty effect — the disproportionate weight attached to a sure outcome that evaporates once both options are merely probable — it tells theorists where the missing structure must go, namely in a probability-weighting function rather than in the value of money. This is what licenses and directs the non-expected-utility program (prospect theory, rank-dependent utility), and it keeps the phenomenon distinct from its neighbors that informal usage conflates with it: it is not loss aversion (reference-dependent gains-versus-losses), not ambiguity aversion (which requires unknown probabilities, whereas Allais probabilities are given), and not a preference reversal across time horizons. The reversal here is across probability re-scalings of the same prizes, and saying so tells an experimenter precisely what to manipulate to reproduce or test it.

Manages Complexity¶

Choice under risk presents a decision theorist with an unbounded space of possible anomalies: any of countless lottery pairs might generate a preference a model fails to capture, and confronted with a mismatch between theory and observed choice, the analyst could in principle suspect any part of the expected-utility apparatus — the shape of the utility function, transitivity, completeness, continuity, independence. The Allais paradox compresses that diagnostic sprawl to a single point of failure. It is a fixed two-choice stimulus that isolates exactly one axiom: a respondent who picks A over B and D over C has violated independence and nothing else, because the second pair is the first with a common 89%-chance-of-$1M consequence stripped from both options, so the axiom's own structure pins the contradiction. The theorist no longer surveys the whole apparatus for the leak; the paradox names the failing component and the parameter that drives it.

That localization carries a compressing branch structure. Because the violation lives in independence rather than in utility curvature, the entire family of "the agent is just more risk-averse" explanations is foreclosed in one stroke — risk aversion is consistent with expected utility, so it cannot produce this reversal — and the modeler is routed off the value-of-money axis and onto the probabilities. From there the certainty effect supplies the single mechanism (sure outcomes overweighted, the overweighting vanishing once both options are merely probable), which tells the theorist precisely where the repair must go: into a probability-weighting function. So a vast literature on departures from expected utility organizes around one reproducible stimulus and one diagnostic question — does certainty get overweighted here? — and the whole non-expected-utility program (prospect theory, rank-dependent utility) inherits a fixed target to fit rather than an open-ended catalog of misbehaving choices to chase. The paradox also fixes the boundaries that keep it from being re-derived under the wrong heading: it is reversal across probability re-scalings of the same prizes, not across gains-versus-losses (loss aversion), not under unknown probabilities (ambiguity aversion), and not across time — so an experimenter knows exactly which single variable to manipulate to reproduce or test it.

Abstract Reasoning¶

The Allais paradox licenses inferences that are precise because the stimulus is engineered to fail in exactly one place — it is a diagnostic instrument as much as an observation.

Diagnostic — localize the failure to a single axiom. The signature move is to read an observed A-over-B-and-D-over-C preference pattern as a violation of the independence axiom specifically, and of nothing else. The inference is licensed by the stimulus's own algebra: the second choice pair is the first with a common 89%-chance-of-$1M consequence stripped from both options, so independence demands A≻B imply C≻D, and the observed reversal contradicts that implication while leaving transitivity, completeness, and continuity untouched. The analyst therefore reasons from the surface choice pattern to a verdict naming the failing component of the expected-utility apparatus, and — running it forward — uses the mechanism (the certainty effect, overweighting a sure outcome that evaporates once both options are merely probable) to predict the reversal in advance for any lottery pair where one member offers certainty and its matched pair does not.

Boundary-drawing — foreclose the explanations that cannot apply. A central and powerful move is elimination. Because the reversal lives in independence rather than in the curvature of the utility function, the entire family of "the agent is simply more (or less) risk-averse" explanations is ruled out in one stroke — risk aversion is fully consistent with expected utility and lives in utility curvature, so it cannot generate this pattern. The analyst thus reasons from "this is an independence violation" to "the repair cannot be on the value-of-money axis," forcing the modeler off the utility function and onto the probabilities. The same boundary work keeps the phenomenon from being misfiled under its neighbors: the reversal is across probability re-scalings of the same prizes, not across gains-versus-losses (loss aversion), not under unknown probabilities (ambiguity aversion), and not across time horizons (temporal preference reversal) — so an experimenter knows exactly which single variable to manipulate to reproduce it, and a theorist knows which existing device does and does not apply.

Interventionist / constructive — direct the repair to where the structure must go. Having localized the failure and foreclosed the curvature fix, the analyst infers where the missing structure belongs: in a probability-weighting function rather than in the value of money. The certainty effect tells the theorist the shape that function must have (overweighting near-certainty, with the overweighting collapsing once both options are uncertain), which is precisely the prediction that the non-expected-utility program is built to satisfy. So the paradox functions as a fixed target: a candidate descriptive theory of choice under risk is tested by whether it reproduces the A≻B, D≻C pattern, and the analyst reasons from "does this model overweight certainty in the right way?" to "does it accommodate Allais?"

Normative-versus-descriptive separation. A further inferential move is to keep two questions apart that the bare anomaly tends to fuse: whether expected utility is a good normative model of how a rational agent should rank lotteries, and whether it is a good descriptive account of how people do. The reversal speaks only to the descriptive claim; the analyst infers that a demonstrated Allais pattern refutes expected utility as a description without thereby settling whether the respondents are irrational or the axiom is wrong as a norm — a distinction that tells the modeler the violation forces a new descriptive theory, not necessarily an abandonment of the normative standard.

Knowledge Transfer¶

Within decision theory under risk the Allais paradox transfers as mechanism, and it does so in two registers. As a diagnostic instrument the two-choice stimulus is the field's reusable test: the same engineered pair (a certain prize against a higher-expected-value gamble, then the matched pair with a common consequence stripped from both) and the same algebra (an A-over-B-and-D-over-C pattern localises the failure to the independence axiom and nothing else) carry intact wherever a candidate descriptive theory of choice under risk must be screened. As a located mechanism — the certainty effect, the overweighting of a sure outcome that evaporates once both options are merely probable — it organises the home domain's subfields: it is the foundational motivation for prospect theory, rank-dependent utility, and cumulative prospect theory; the workhorse stimulus of experimental risk-elicitation; the device behind descriptive asset-pricing models that permit independence violations; and the evidence that policy-relevant choices over uncertain outcomes will not match the expected-utility prescription. These applications all sit inside the decision-under-risk family — the same instrument and the same cognitive mechanism applied to finance, public policy, and health-decision research — so this is reach within one substrate, not transfer across substrates.

Beyond choice under risk the honest characterisation has two parts, and neither is "metaphor for a mechanism that travels." First, the stimulus does not transfer as a causal mechanism — it is an instrument, applicable only where its precondition holds: a chooser ranking lotteries over known probabilities, with a certainty contrast available between matched pairs. Where there is no probability-weighting agent there is no Allais pattern to find; the boundary to mark is instrument-reach, not analogy. Second, the content the paradox demonstrates is a property of human cognition, and what generalises is not the named paradox but the primitives behind it. The narrow, portable mechanism is the certainty effect / probability-weighting distortion, which recurs across the cognitive landscape of risky choice and is properly housed in expected_utility_theory and a probability_weighting construct, of which Allais is one worked demonstration. The broader, looser lesson — intuitive preferences can systematically violate normative axioms — is real and travels widely, but it travels as the parent, the cognitive-bias family, not as the Allais paradox; an organisation or an algorithm whose choices "violate independence" is exhibiting that general bias pattern, demonstrated here in one specific human stimulus, not a portable Allais mechanism. The home-bound cargo that stays behind is the named demonstration itself: the specific lottery pair, the independence-axiom violation, the $1M-versus-$5M prizes, and the experimental apparatus by which the certainty effect was first isolated — and, importantly, the careful normative-versus-descriptive separation the paradox forces (it refutes expected utility as a description without settling whether the axiom fails as a norm). So when the cross-domain lesson is needed, the honest move is to carry the probability-weighting mechanism (for risky-choice settings) or the cognitive-bias parent (for the general "axiom violation" point), and to treat any invocation of "the Allais paradox" outside lottery choice with known probabilities as an instrument applied past its precondition rather than a mechanism that has genuinely travelled (see Structural Core vs. Domain Accent).

Examples¶

Canonical¶

The paradox is its own precise construction. Choice 1: Option A is $1M for certain; Option B is an 89% chance of $1M, a 10% chance of $5M, and a 1% chance of nothing. Most people pick A. Choice 2: Option C is an 11% chance of $1M (else nothing); Option D is a 10% chance of $5M (else nothing). Most of the same people pick D. Now the algebra: both options in Choice 1 share a common 89%-chance-of-$1M component; strip that common consequence from both and Choice 1 becomes Choice 2 exactly (A's remaining 11%-of-$1M is C; B's remaining 10%-of-$5M-plus-1%-of-nothing is D). The independence axiom says a common consequence cannot flip the ordering, so A≻B forces C≻D. The observed A≻B, D≻C pattern violates it — and only it.

Mapped back: The $0/$1M/$5M prizes are the outcome space; Choices 1 and 2 are the matched lottery pair, the second built from the first by stripping the common 89%-of-$1M. The certain $1M in Option A is the certainty contrast, absent once both of Choice 2's options are merely probable. The reversal is the observed reversal the independence axiom forbids, and pinning the fault to independence alone (transitivity and completeness intact) is the localised axiom failure, driven by the certainty effect.

Applied / In Practice¶

Prospect theory (Kahneman and Tversky, 1979), and its cumulative successor, took Allais as a fixed target and built the repair exactly where the paradox says it belongs — in a probability-weighting function rather than in the utility of money. By letting decision weights be a nonlinear transform of probabilities that overweights certainty and small probabilities, the theory reproduces the A≻B, D≻C pattern and, deployed on real choices, explains why the same person will simultaneously buy insurance (overweighting a small chance of catastrophic loss) and a lottery ticket (overweighting a small chance of a large gain) — behavior expected-utility theory cannot house. This probability-weighting machinery is now standard in behavioral finance and insurance modeling.

Mapped back: Prospect theory accepts the localised axiom failure — independence, not utility curvature — and places the fix at the repair locus, a probability-weighting function. Overweighting near-certainty is precisely the certainty effect given functional form, and reproducing the Allais observed reversal is the screening test any descriptive theory must pass. The insurance-and-lottery coexistence shows the same probability-weighting distortion doing real predictive work beyond the original lottery stimulus.

Structural Tensions¶

T1: The engineered stimulus versus the cognitive mechanism (a single demonstration doing the work of a general phenomenon). Allais operates in two registers at once: a fixed two-choice stimulus with specific prizes and probabilities, and the certainty effect it isolates. The stimulus is one contrived point — a matched lottery pair built to strip a common 89%-of-$1M consequence — while the mechanism, overweighting of certain outcomes, is a general feature of risky cognition. Conflating the two invites over-reading: treating the particular $1M/$5M numbers as essential, or invoking "the Allais paradox" wherever certainty seems overweighted, mistakes the demonstration for the thing demonstrated. Conversely, dismissing the mechanism because the stimulus is artificial throws out the general phenomenon with the specific vehicle. The tension is that the concept's name and fame attach to a single engineered instrument, while its portable substance is the cognitive mechanism that instrument happens to expose. Diagnostic: Is the claim about the specific lottery stimulus (an instrument with a precondition), or about the certainty effect (the general mechanism the stimulus demonstrates)?

T2: Descriptive refutation versus normative standing (which of expected utility's two roles actually failed). The reversal refutes expected utility as a description of how people choose, but the paradox deliberately leaves open whether the independence axiom fails as a norm of rational choice. That gap is not a technicality: the same data support two opposite responses. Read as "EU is a poor description of rational agents," the reversal licenses building non-expected-utility theories that take the observed preferences at face value. Read as "people are systematically irrational and EU remains the standard to aspire to," it licenses debiasing the chooser and preserving the axiom as a norm. Allais itself cannot adjudicate — respondents frequently reverse their own choices once the common-consequence algebra is shown to them, which cuts toward the "error" reading. The tension is that the anomaly underdetermines whether the descriptive model or the human reasoner is at fault. Diagnostic: Is this Allais pattern being treated as evidence that the normative axiom is wrong, or that the agent erred against a norm still worth holding — and does the intended response (new theory vs debiasing) follow the right reading?

T3: Surgical localization versus the artificial edge case that achieves it (generality bought by contrivance). The paradox's celebrated virtue is that it pins the failure to exactly one axiom, leaving transitivity, completeness, and continuity untouched — a precision achieved only by careful engineering: matched pairs, a stripped common consequence, and probabilities pressed against certainty (89/10/1 versus 11/10). But that same engineering is a limitation. The violation is demonstrated at a knife-edge where certainty is contrasted with near-certainty, and its strength depends on those extreme probabilities; presenting the choices transparently as compound-versus-reduced lotteries, or nudging the stimulus off the certainty boundary, measurably shrinks the reversal. So the clean single-axiom verdict rests on a designed configuration that real choices rarely present, and the leap from "independence fails here" to "EU is descriptively false in general" generalizes from a contrived corner. The tension is that the diagnostic precision and the artificiality are the same feature. Diagnostic: Is the independence violation robust away from the certainty knife-edge and under transparent presentation, or is its sharpness an artifact of the engineered edge case that isolates it?

T4: Names where the repair goes versus underdetermining which repair (a fixed target many models hit). Having foreclosed the utility-curvature fix, the paradox routes the missing structure "onto the probabilities" and is used as a fixed target: a descriptive theory is screened by whether it reproduces A≻B, D≻C. But a single data pattern is a weak discriminant. Prospect theory, rank-dependent utility, and cumulative prospect theory all accommodate it — and regret theory reproduces the very same reversal through anticipated regret, with no probability-weighting function at all. So "the repair must live in a probability-weighting function" is itself an interpretive commitment the data underdetermine, and "accommodates Allais" is necessary but far from sufficient support for any one model. The tension is that the paradox is celebrated as a sharp target while, as a lone stimulus, it cannot distinguish the rival mechanisms that pass it. Diagnostic: Is a model being credited merely for reproducing the Allais reversal — which regret theory and several weighting schemes all do — or discriminated by evidence that goes beyond this single, jointly-satisfiable pattern?

T5: Autonomy versus reduction (a named lottery demonstration or the probability-weighting mechanism it exhibits). Within decision theory under risk the Allais paradox transfers as mechanism in two registers — the reusable diagnostic stimulus and the located certainty effect — across behavioral economics, experimental elicitation, and descriptive asset pricing. But beyond lottery choice it does not travel as a mechanism at all: the stimulus is an instrument, applicable only where a chooser ranks lotteries over known probabilities with a certainty contrast available, and where there is no probability-weighting agent there is no Allais pattern to find. What generalizes is not the named paradox but the primitives behind it — the certainty-effect / probability-weighting distortion, housed in expected_utility_theory and a probability_weighting construct, and, more loosely, the cognitive-bias parent that intuitive preferences can systematically violate normative axioms. The specific lottery pair, the independence violation, and the $1M/$5M prizes are home-bound cargo. The tension is between a famous, self-contained demonstration and the recognition that its cross-domain lesson belongs to the mechanism and the bias family it illustrates. Diagnostic: Resolve toward the probability-weighting mechanism (for risky-choice settings) or the cognitive-bias parent (for the general "axiom violation" point) when carrying the lesson outward; toward the Allais paradox when screening a descriptive theory of lottery choice against a known-probability stimulus.

Structural–Framed Character¶

The Allais paradox is framed-leaning on the structural–framed spectrum — clearly off the structural side, though not at the pure framed pole, because embedded in it is a genuine cognitive regularity (the certainty effect) that keeps it from being a bare theory-internal verdict. The criteria pull in a consistent direction. Evaluative_weight leans framed: the very identity of the entry is stated as a violation, an anomaly, a paradox — all defined relative to a normative standard (the independence axiom of expected-utility theory), so the named object carries a comparison-to-an-ideal valence built in; the entry's own insistence that the reversal refutes EU only as a description and not necessarily as a norm is exactly the work of holding that evaluative charge at bay, which shows how present it is. Human_practice_bound is mixed but tilts framed: the underlying certainty effect is a fact of human cognition that operates in choosers whether or not any theorist is watching, but the paradox as a named construct is an engineered instrument of a research practice — a designed two-choice stimulus that exists to be run on subjects — and has no meaning where there is no probability-weighting agent to exhibit it. Institutional_origin is strongly framed and decisive: there is no "paradox" at all except relative to expected-utility theory's axiom system — strip away the independence axiom and the A-over-B/D-over-C pattern is just a preference ordering, not a contradiction; the entry is an artifact of a specific theoretical tradition (Allais's 1953 construction against EU, the later prospect-theory and rank-dependent repairs), the way a survey convention is an artifact of a discipline rather than a fact nature marks. Vocab_travels points framed: the named stimulus — the matched lottery pair, the independence-axiom algebra, the $1M/$5M prizes — does not float free of decision-theoretic substrate; what travels is not this vocabulary but the certainty-effect mechanism carried in other primes. Import_vs_recognize is bimodal in the entry's own terms: within decision theory under risk the stimulus transfers as a reusable diagnostic instrument (recognition/reuse across behavioral economics, experimental elicitation, descriptive asset pricing), but beyond it any invocation of "the Allais paradox" is either an instrument applied past its precondition or a loose metaphor — the genuine cross-domain travel runs through the mechanism and the bias family, not the named paradox.

The portable structural skeleton is the certainty-effect / probability-weighting distortion — a systematic overweighting of outcomes as they approach certainty, collapsing once all options are merely probable — and, more loosely, the cognitive-bias pattern that intuitive preferences can systematically violate a normative axiom. That skeleton is genuinely portable and does real cross-domain work, which is what tempts a more structural reading. But it does not pull the paradox off its framed-leaning position, because that skeleton is exactly what the Allais paradox instantiates from its parents — expected_utility_theory, a probability_weighting construct, and the cognitive-bias family — not what makes the named demonstration travel: the substrate-spanning reach belongs to those parents, while the entry's distinctive content (the engineered lottery pair, the independence-axiom framing, the specific prizes, and the careful normative-versus-descriptive separation the construction forces) is the home-bound cargo that stays in decision theory under risk. Its character: a theory-relative, normatively-framed diagnostic construct — an anomaly that exists only against expected-utility's axiom system — structural only in the certainty-effect regularity it exhibits and hands off to its parent primes, while the named two-lottery instrument itself never leaves its home domain.

Structural Core vs. Domain Accent¶

This section decides why the Allais paradox is a domain-specific abstraction and not a prime, and it carries the case for its domain-specificity.

What is skeletal (could lift toward a cross-domain prime). Two thin structures survive stripping the lottery apparatus. The narrower is the certainty-effect / probability-weighting distortion: outcomes are overweighted as they approach certainty, and the overweighting collapses once every option is merely probable. The broader and looser is the cognitive-bias pattern that intuitive preferences can systematically — not randomly — violate a normative axiom. Both are genuinely portable: the probability-weighting distortion recurs across the cognitive landscape of risky choice (it is what makes the same person buy both insurance and a lottery ticket), and the axiom-violation pattern travels wherever a chooser's revealed preferences depart from a coherence requirement. That recurrence is mechanism, not metaphor — which is exactly why it lifts to the parents the paradox instantiates rather than remaining proprietary to it. But it is the core the paradox shares, not what makes the named demonstration distinctive.

What is domain-bound. Everything that makes this the Allais paradox in particular is decision-theoretic furniture and does not survive extraction. The engineered two-choice stimulus (the $0/$1M/$5M prizes, the 89/10/1-versus-11/10 probability mixings), the matched-pair algebra by which stripping a common 89%-chance-of-$1M consequence converts Choice 1 into Choice 2, the independence axiom whose violation the pattern demonstrates, and the surgical localization that leaves transitivity, completeness, and continuity untouched are all internal to expected-utility theory's axiom system. The decisive test: strip away the independence axiom and the A-over-B / D-over-C pattern is no longer a paradox at all — just a preference ordering, with nothing to contradict. The anomaly is constituted by the very theory it is defined against, and there is no Allais pattern to find where there is no probability-weighting agent ranking known-probability lotteries. Equally home-bound is the careful normative-versus-descriptive separation the construction forces — it refutes EU as a description without settling whether independence fails as a norm — a distinction that only arises inside the theory.

Why this does not clear the prime bar. A prime's vocabulary travels and its transfer is recognition of the same mechanism, not analogy. The paradox's transfer is bimodal, and in an unusual way, because the named object is not even a mechanism but a diagnostic instrument. Within decision theory under risk it transfers intact — the same stimulus screens any candidate descriptive theory, and the located certainty effect organizes behavioral economics, experimental elicitation, and descriptive asset pricing — reach within one substrate, not across. Beyond lottery choice with known probabilities it does not travel as mechanism at all: the stimulus is an instrument applied past its precondition, and any invocation of "the Allais paradox" elsewhere is loose metaphor. When the cross-domain lesson is wanted, it is already carried, in more general form, by the parents the paradox instantiates — expected_utility_theory and a probability_weighting construct for risky-choice settings, and the cognitive-bias parent for the general "axiom violation" point. The substrate-spanning reach belongs to those parents; the named two-lottery demonstration carries decision-theoretic baggage that stays home.

Relationships to Other Abstractions¶

Current abstraction Allais Paradox Domain-specific

Parents (2) — more general patterns this builds on

Allais Paradox is part of Certainty Effect Domain-specific

The Allais Paradox contains the certainty-boundary overweighting isolated by its paired lottery choices.
Allais Paradox presupposes Expected Utility Prime

The Allais paradox is defined by a preference reversal that can be diagnosed only against expected utility's independence axiom.

Hierarchy paths (13) — routes to 6 parentless roots

Allais Paradox → Expected Utility → Expected Value → Aggregation → Micro Macro Linkage

Show alternative paths (12)

Not to Be Confused With¶

Ellsberg paradox. The sibling demonstration, and the one most often swapped for this entry: a two-urn choice in which people prefer betting on known over unknown probabilities, revealing ambiguity aversion. It too is a designed anomaly against a coherence axiom (the sure-thing principle / a probabilistic-sophistication requirement), but its trigger is missing probability information, whereas every probability in the Allais lotteries is fully specified. The two mark different failures of expected utility — Ellsberg indicts how unknown probabilities are handled, Allais how known probabilities near certainty are weighted. Tell: are the probabilities given as exact numbers with a certainty contrast (Allais) or deliberately left unspecified so the chooser faces ambiguity (Ellsberg)?
Common ratio effect. The near-twin EU violation, easy to merge with Allais because both are independence-axiom failures driven by certainty. But Allais is the common-consequence effect — the reversal appears when a shared consequence is added to or stripped from both lotteries — while the common-ratio effect scales all the probabilities in a pair by a common factor (e.g. a sure $3000 vs an 80%-chance-$4000, then both odds cut to a quarter). Same certainty effect, different algebraic operation on the lottery pair. Tell: does the manipulation strip a shared outcome from both options (Allais/common-consequence) or multiply every probability by a common ratio (common-ratio effect)?
St. Petersburg paradox. Another famous decision-theory puzzle, but a contrast case: a gamble with infinite expected value that people will pay only a little to play. Its classical resolution is exactly the move Allais forecloses — diminishing marginal utility, curvature of the utility-of-money function (Bernoulli). Allais is defined by being unfixable that way, since its violation lives on the probabilities, not the value axis. Tell: can the anomaly be dissolved by a concave utility function (St. Petersburg) or does no utility curvature reconcile it (Allais)?
Lichtenstein–Slovic preference reversal. The classic "preference reversal" of experimental psychology, where an agent chooses one bet but prices another higher — a reversal across elicitation procedures. Sharing the word "reversal" invites confusion, but Allais reverses across probability re-scalings of the same prizes within one choice mode, implicating independence, not the choice-vs-pricing mismatch or the response-mode compatibility that drives the Lichtenstein–Slovic effect. Tell: does the flip come from changing how the preference is elicited (choose vs price) — procedural reversal — or from restructuring the lottery probabilities while the elicitation stays fixed (Allais)?
The certainty effect (the mechanism). The overweighting of sure outcomes relative to merely-probable ones. This is not a rival to sort out but the underlying mechanism the Allais stimulus was engineered to isolate — the entry's own T1 warns against collapsing the two. The Allais paradox is one specific demonstration; the certainty effect is the general cognitive regularity it exposes, and it drives other phenomena (including the common-ratio effect above) with no Allais lottery in sight. Tell: is the claim about the specific two-lottery instrument (Allais) or about the general overweighting-of-certainty it demonstrates (certainty effect)?
Prospect theory / the probability-weighting parent. The repair, not the anomaly. Prospect theory (and rank-dependent and cumulative variants) is a descriptive model built to reproduce the Allais pattern by putting a nonlinear weighting function on probabilities; the paradox is the fixed target such theories must hit, and — per the entry's T4 — regret theory hits the same target by a different route. Confusing the demonstration with the theory that explains it mistakes the disease for one of its cures. The genuinely portable content (the probability-weighting distortion, and the looser "intuitive preferences can violate a normative axiom") belongs to these parents, treated more fully in Knowledge Transfer and Structural Core vs. Domain Accent. Tell: is this a finding that people violate independence (Allais) or a model engineered to accommodate that finding (prospect theory)?

Neighborhood in Abstraction Space¶

Allais Paradox sits in a sparse region of the domain-specific corpus (82^nd percentile for distinctiveness): few abstractions share its structure, so a faithful description tends to retrieve it precisely.

Family — Choice Paradoxes & Collective Decision-Making (14 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-07-12