Preference Reversals

Definition

Procedure invariance — the assumption that "the relation of preference should not depend on... the method of elicitation" — is a load-bearing pillar of rational choice theory: if it fails, preferences can't be represented as maximizing any single utility function. A preference reversal is exactly this failure made concrete: a gamble is chosen over another in a direct choice, yet priced lower than it in a separate pricing task (or vice versa) — the two elicitation formats disagree about which option is better. Slovic frames the deeper stakes with a borrowed line: preferences may not be values that exist and are merely reported, but something closer to "they ain't nothing till I call them" — constructed, not revealed, by the act of asking.

In the Book

Chapter 27 traces the phenomenon back to Slovic and Lichtenstein's early finding that choices between paired gambles track mainly the probability of winning, while buying and selling prices for the very same gambles track mainly the dollar amounts at stake — so a low-probability, high-payoff "P-bet"-versus-"$-bet" pair can be chosen one way and priced the opposite way by the same person. Chapter 28's contingent-weighting account explains the mechanism: whichever attribute (probability or payoff) is more compatible with the response scale being used gets weighted more heavily in that response — probabilities loom larger when choosing, dollar amounts loom larger when pricing, because pricing is itself denominated in dollars. The book notes economists' initial dismissal of the finding as an experimental artifact (Grether and Plott, 1979) and its survival "despite extreme attempts by economists to make preference reversals disappear," including tests with real stakes and market incentives designed to arbitrage the inconsistency away.

Why It Matters

Preference reversals undercut the working assumption behind most measurement of value — that there is a single underlying preference order waiting to be elicited, and any reasonable elicitation method will recover it. In practice this means the format of a survey, auction, or rating scale is not a neutral window onto a person's values — it actively shapes the answer, which has direct consequences for anyone designing willingness-to-pay studies, policy referenda, juror damage awards, or product ratings: switching the question format can switch the conclusion, even though nothing about the underlying options changed.