Mental accounting

Evidence

Thaler (1985) developed a model of consumer behavior that links mental accounting with the notion of transaction utility: customers derive utility not only from the product but also from the quality of the deal. His best-known example is the beer on the beach. The same bottle of the same brand, drunk in the same place — but people whose friend fetches it from a fancy resort name a median maximum price of $2.65, while those whose friend fetches it from a run-down grocery store name $1.50. The consumption is identical; only the expected reference price differs.

The paper’s evidence, however, is thin and entirely hypothetical. Thaler tested the four rules for combining gains and losses with 87 students in a statistics class, who judged which of two invented people was happier — 56 of 87, for instance, on whether two separate lottery wins of $50 and $25 feel better than a single one of $75. The willingness-to-pay question went to participants in an executive development programme, the ticket questions to 26 to 31 MBA students per version. In no case did anyone buy anything.

Thaler (1999) summarizes two decades of research. The paper contains no experiments of its own; it organizes three components: how outcomes are perceived and evaluated, how money is assigned to budgets and accounts, and how often the books are balanced.

Three points of that review bear directly on pricing pages. Thaler reports them from other people’s work: the preference for flat-rate plans even where paying by use would cost less; the effect of “pennies-a-day” framing, in which a $100 annual membership appears as “a mere 27 cents a day” and so falls below the threshold at which expenses are booked at all; and the observation that gym attendance is highest in the month the dues are paid, declines afterwards, and rises again with the next bill. On the middle point there are data now (see below); the other two remain reading leads without evidence.

Since 2025 these problems have been re-run — all seventeen at once. Li and Feldman (2025) repeated Thaler’s review as a pre-registered Registered Report: 1,007 respondents on a US online panel, around 500 per problem, the procedure reviewed and accepted before any data were collected. This is the test Thaler himself asked for in 1985 — though not as a field experiment.

The beer on the beach holds — just. The authors replaced the beer with soda, because not every respondent drinks alcohol. The difference survives: $7.09 against $4.17 in willingness to pay (t(506) = 2.30, p = .011 — a difference chance alone would produce in about one case in a hundred). At d = .20 the effect is small, though, and the spread in the hotel condition is enormous (standard deviation 19.85 against 3.98) — a few very high figures carry the mean.

The four combination rules also hold, but more weakly — and with a shift worth noting. Where 64 percent of Thaler’s respondents thought two separate wins more pleasing, 35 percent did so in the repetition, while 52 percent ticked “no difference”. For losses the same picture: 76 percent in the original against 42 percent, with 44 percent undecided. The direction holds; the conviction does not.

That bundled losses weigh more heavily is confirmed. Sixty-one percent found two payments on the same day more burdensome than two weeks apart, only 19 percent the reverse (57 against 34 in the original). For gains, by contrast, the result came out the other way round this time.

One failure bears directly on pricing pages. The most famous example of topical accounts — the same $5 saving feels different on a calculator than on a jacket — could not be repeated. In the original 68 against 29 percent would make the trip to the other shop; in the repetition 11 against 7 percent (p = .066). The source, Tversky and Kahneman (1981), is among the most cited papers in the field at over 28,000 citations. The authors offer a plain explanation: driving twenty minutes now costs more than $5.

The “house money effect” failed too, and in the opposite direction. After an unexpected win people are supposed to become more willing to gamble; in Thaler 70 percent took the bet, in the repetition 28 percent. The authors point to a longer debate: after gains, either direction can be argued theoretically.

And one of the three reading leads above has now been tested. Li and Feldman tested the “pennies-a-day” framing directly for the first time — Thaler had merely asserted it. “A mere 27 cents a day” was rated 44.5 out of 100, “$100 a year” 26.0; t(331) = 5.57, p < .001, d = .61. That is the strongest single finding in the whole paper — and precisely the one for which there had been no data at all. Shown side by side, the gap shrinks to d₂ = .29.

With sunk costs it depends on the case. That a paid ticket draws people out into a snowstorm more than a gifted one was confirmed (32 against 14 percent). That a paid club membership keeps people playing through pain was not: 76 percent stopped. The authors’ conclusion: sunk costs lose against anticipated physical pain.

Limits

Thaler did not find his best-known rule confirmed. The value function implies that losses should be combined — the basis of the marketing advice to bundle costs. The 1985 survey still fitted: 66 of 87 judged the person with two separate tax demands of $100 and $50 more upset than the one with a single demand of $150. A follow-up using a different question format reversed it. Asked whether two events are better on the same day or two weeks apart, respondents wanted losses separated. Thaler (1999) records that people follow the rules “with the exception of multiple losses”. His reading: one loss makes you more sensitive to the next rather than blunting it.

Thaler calls the evidence thin himself. On the budgeting part of the model he writes in 1985 that both theory and evidence are less well developed than the rest; the data come from a small sample of households that will not support statistical tests. He explicitly asks for field experiments in a real marketing environment. The 1999 review rests largely on other people’s work, including unpublished working papers.

Not automatically a mistake. Thaler explicitly declines to judge mental accounting irrational: the procedures evolved to economize on thinking time and to deal with self-control problems. Teach people to ignore sunk costs, he asks, and do they stop abiding by “waste not, want not”?

The concept is broadly framed, and individual predictions are supported to varying degrees. The claim that B2B buyers can justify a euro discount internally more easily is a hypothesis without a verified source.

The repetition has limits of its own, and the authors name them. The sample comes entirely from the United States and from an online panel, collected during the Covid pandemic — the authors consider altered risk behaviour in that period possible. Because many of the original papers reported only percentages without tests, the analyses had to be reconstructed; the authors explicitly warn against over-extending the comparisons between original and repetition. Each problem was also answered by only about half the respondents.

A second, larger test already exists. Li and Feldman point to an international collaboration by Priolo and colleagues, released while they were revising, reporting “very similar results” — here, then, at second hand.

The evidence is better than the thin original data suggest. Eleven of the seventeen problems hold, three come out mixed, three fail — under a pre-registered procedure and with roughly three times the sample per problem. Among the failures, however, is a central piece, the calculator example, and the confirmed effects are consistently smaller than in the original.