Microeconomics
Revealed Preference (Samuelson)
If you bought it when something cheaper was on the shelf, you preferred it
Revealed preference is Paul Samuelson's 1938 reframing of consumer theory. Instead of starting from invisible "utility" and deducing demand, it starts from what people actually buy and works backwards. Out of that move came two testable axioms — WARP and SARP — and a whole research program asking when choice data is consistent with any rational preference at all.
- Proposed byPaul Samuelson, 1938
- Core testWARP / SARP / GARP
- Equivalence theoremAfriat (1967)
- Uses utility?No (recovered, not assumed)
- Empirical violation rate5–25% of households
Interactive visualization
Press play, or step through manually. The visualization is yours to drive — try it before reading on.
Watch the 60-second explainer
A condensed visual walkthrough — narrated, captioned, under a minute.
The core idea
Classical consumer theory says: people maximize a utility function subject to a budget constraint. That works mathematically, but utility is invisible — you can't open someone's head and read it. Samuelson thought this was a weakness. He asked: can we do all of consumer theory using only what we observe — namely, what people buy at what prices?
The answer turned out to be yes, with one ingredient: a consistency axiom. Suppose at prices p1 a consumer chooses bundle x1, and at prices p2 they choose x2. If x2 was affordable at prices p1 (the consumer could have bought it then but didn't), we say x1 is directly revealed preferred to x2. Notation: x1 RD x2.
From that single observable definition, the rest of the theory follows.
WARP — the weak axiom
The Weak Axiom of Revealed Preference says: if x1 RD x2, then x2 is not directly revealed preferred to x1. Stated less abstractly: you can't both have "chosen A when B was on the table" and "chosen B when A was on the table" — that's a flat contradiction.
A worked WARP test
Suppose a consumer faces two periods, with prices and chosen bundles for two goods (apples, bread):
| Period | Prices (pA, pB) | Chosen bundle (xA, xB) | Spent |
|---|---|---|---|
| 1 | ($1, $1) | (4, 6) | $10 |
| 2 | ($2, $1) | (2, 8) | $12 |
Was bundle 2 affordable in period 1? Cost at p1: 1·2 + 1·8 = $10. Yes — the consumer could have bought it but didn't. So x1 RD x2.
Was bundle 1 affordable in period 2? Cost at p2: 2·4 + 1·6 = $14. The consumer only spent $12. No — bundle 1 was unaffordable. So x2 is not revealed preferred to x1. WARP holds. Good.
If instead bundle 1 had cost only $11 at p2 (affordable on a $12 budget), and the consumer had still picked bundle 2, that would be a direct WARP violation — they'd be choosing inconsistently across the two periods.
SARP and GARP — closing the loop
WARP rules out two-step contradictions but allows three-step cycles. The Strong Axiom (SARP) extends to chains: if x1 is revealed preferred to x2, x2 to x3, ..., xk-1 to xk, then xk is not revealed preferred to x1. Houthakker (1950) proved SARP is equivalent to the existence of a strictly increasing utility function generating the data.
The Generalized Axiom (GARP), introduced by Hal Varian (1982), allows for ties. GARP is the practical workhorse: given a finite choice dataset, you can run a polynomial-time algorithm to check whether GARP holds. Afriat's theorem (1967) tightens the result — GARP-consistent data can be rationalized by a continuous, monotone, concave utility function.
Revealed preference vs utility maximization
| Utility maximization | Revealed preference | |
|---|---|---|
| Starting point | Postulated utility function U(x) | Observed choices at varying prices |
| Primitive object | Preferences (introspective) | Choice data (observable) |
| Empirical content | Demand curves, Slutsky equation | Same demand curves — same content |
| Testability | Indirect (test demand restrictions) | Direct (test WARP/GARP on data) |
| Welfare claims | Cardinal interpretations possible | Ordinal only — no interpersonal comparisons |
| Recovers utility? | Assumes it | Reconstructs it (when GARP holds) |
| Foundational paper | Marshall (1890), Hicks-Allen (1934) | Samuelson (1938), Houthakker (1950) |
The deep result of the Samuelson-Houthakker-Afriat program is that the two approaches are equivalent when SARP/GARP holds — the empirical content is identical. The choice between them is methodological, not substantive.
Counterarguments and limitations
Behavioral violations. Lab experiments and scanner-panel data routinely turn up GARP violations. Choi, Fisman, Gale, and Kariv (2007) found about 12% of subjects' choice sequences violated GARP — too many for the violations to be pure noise. Cohen (1995) emphasized that intransitive cycles do appear in stated preferences in surveys.
Status-quo bias and framing. If the same bundle is chosen because it was the default, not because it was preferred, revealed preference attributes a preference where none exists. Behavioral economists argue this severs the welfare interpretation.
Hidden quality variation. If the "same" good differs between periods (a higher-quality apple this week), the choice data is misclassified.
Welfare ambiguity. When GARP fails, there's no single utility function rationalizing the data — yet welfare comparisons must still be made. Bernheim and Rangel (2009) propose "behavioral revealed preference" as a refinement.
Variants and extensions
- Stochastic revealed preference. McFadden's choice models (1974) allow random shocks; the consumer's deterministic preferences are recovered up to a distribution.
- Preferences over lotteries. Expected-utility revealed preference adds the independence axiom on top of WARP.
- Production sets. Varian (1984) extended GARP to firm input-output data — testing whether observed production decisions are consistent with profit maximization.
- Behavioral RP. Bernheim-Rangel and Manzini-Mariotti models permit limited inconsistency while preserving welfare statements.
Empirical impact
Revealed preference axioms have a long empirical track record. Hal Varian's 1982 algorithm runs in O(n3) on n observations and made GARP testing routine. Studies have used scanner data, household panels, lab experiments, and even animal choice data (rats, monkeys) to ask whether subjects rationalize. Choi, Kariv, Müller, and Silverman (2014) used a budget-allocation experiment with 1,182 Dutch participants and found rational consistency scores correlated with wealth and cognitive ability — the violators tended to be poorer and lower-scoring.
Common pitfalls
- Confusing WARP with rationality. WARP is necessary, not sufficient. SARP/GARP is the right test for full utility maximization.
- Forgetting the "could have bought" check. Bundle 2 being cheaper at prices p1 doesn't reveal anything until you confirm it was within the period-1 budget.
- Reading welfare into preference. Revealed preference says nothing about whether the consumer is happy — only about consistency of choice.
- Ignoring time variation. If preferences themselves changed between periods (new tastes, new information), there is no single underlying preference to reveal.
- Confusing it with stated preference. Asking people what they prefer is stated preference and is methodologically distinct.
Frequently asked questions
What is revealed preference?
Revealed preference is the idea that a consumer's preferences can be inferred from the bundles they actually choose at various prices and incomes. Samuelson (1938) wanted to remove psychological constructs like "utility" from demand theory and rebuild it on observable purchases alone.
What is WARP?
The Weak Axiom of Revealed Preference says: if bundle A is chosen when bundle B is affordable, then in any situation where A is affordable but B is chosen, A must not have been affordable. WARP rules out direct contradictions in choice but is weaker than full utility-maximization.
What is SARP?
The Strong Axiom of Revealed Preference rules out cycles of indirect preference. If A is revealed preferred to B, B to C, and C to A, that's a SARP violation. SARP is equivalent to the existence of a well-behaved utility function rationalizing the data (Houthakker 1950).
Why did Samuelson invent it?
In the 1930s, economists were uneasy about utility — it couldn't be measured and seemed metaphysical. Samuelson proposed that demand theory's empirical content could be expressed entirely in terms of choices. If we never need to peek inside the head, we don't need utility.
What's GARP?
The Generalized Axiom of Revealed Preference (Varian 1982) extends SARP to handle indifference. GARP is now the standard test in empirical work — Afriat's theorem says GARP-consistent data can be rationalized by a continuous, monotone, concave utility function.
Do real consumers satisfy WARP?
Mostly, but not always. Studies of household scanner data find WARP and GARP violations in 5–25% of consumers depending on the dataset. Choi, Kariv, Müller, and Silverman (2014) found GARP-rational behavior correlates with income, education, and cognitive ability.