Sunk cost fallacy

The paid-for ticket

Someone paid $80 for a concert ticket. On the night, they have a bad cold and a storm is coming in. “I’d skip it if the ticket had been free, but I paid eighty dollars, so I’m going.”

The $80 is gone whether they go or not. The real choice is between an evening at the concert with a cold in a storm and an evening at home. By their own account, they’d pick home if not for the payment, so the payment is the only thing tipping the decision, and it can’t change which evening is better.

Hours instead of dollars

“I’ve practiced the clarinet for three years. I don’t enjoy it anymore and I’d rather take up climbing, but quitting now would mean all that practice was for nothing.”

The practice has already happened, and whatever it gave (skill, discipline, some good memories) is kept either way. What quitting changes is the future: fewer hours of an activity they no longer enjoy, more of one they want to do. Sunk time works the same way as sunk money, although in studies its effect is less consistent.

“We can’t waste what we’ve spent”

A company has spent $400,000 building its own booking system, which is still months from working. An off-the-shelf product would do the job for less than the cost of finishing. In the meeting, a manager says: “If we switch now, we’ve thrown away four hundred thousand dollars.”

The $400,000 is lost under both options. The comparison that matters is the remaining cost of finishing (and the value of the result) against the cost of the product (and its value). This version sounds responsible, because “don’t waste money” is usually good advice. Here it is applied to money that can no longer be saved.

Economists and argumentation theorists have pointed out that responding to past investment is often reasonable, because the investment is linked to something that does bear on the future:

• When the investment carries information. Having spent a lot may tell you something: that the remaining cost is small, that you’ve learned how to do it, or that you judged it valuable when you had good information.
• When quitting has its own future costs. Abandoning a project can damage your reputation or credibility, break a promise, or trigger penalties. Those are costs of stopping, not sunk costs.
• When you committed on purpose. Paying up front to bind yourself (a nonrefundable class, a training plan announced to friends) is a deliberate strategy, and following through can be the point.
• When the spending has changed what you can afford. If the money or time is gone, the alternatives you could once have chosen may no longer be open.
• When you’re nearly finished. If little remains to be spent and the payoff is real, finishing is justified by the future, even though it looks like sticking with a big investment.

The test: if the past spending had cost you nothing, but you knew everything you know now, would you still choose to continue?

One semester left

A graduate student has spent four years on a degree and has lost some enthusiasm for the field. One semester remains, and the degree qualifies them for the jobs they want. “I’ve come this far. I’m finishing.”

“I’ve come this far” sounds like a sunk cost argument, but the case for finishing is about the future: one semester of cost for a qualification with real value. Someone who’d spent nothing so far and was offered the same deal (one semester for this degree) would reasonably take it. That’s the nearly finished condition above.

Keeping a promise

A small bakery agreed months ago to cater a friend’s wedding at a fixed price. Ingredient costs have since gone up, and the order will now barely break even. The owner says: “We said we’d do it, and we’ve already bought half the supplies. We’re doing it.”

The supplies already bought are a sunk cost, and if they were the only reason, it would be the fallacy. But backing out would damage the bakery’s reputation with the friend and with everyone at the wedding, some of them future customers. That’s a future cost of quitting, and it makes following through a reasonable choice.

Arkes and Blumer, who ran the classic early studies, proposed that it comes from a desire not to appear wasteful. “Don’t waste” is a good rule in most of life, and applying it to money that’s already lost feels like applying it correctly. A related account is mental accounting: people treat spending on a project as an “open” account that is settled only when the thing is used or finished, so stopping early feels like recording a loss. On this view, framing matters: described as “writing off” what you’ve spent, stopping feels like accepting a certain loss (see Framing effect).

A second explanation is self-justification: continuing protects the belief that the original decision was a good one, especially when you made it yourself.

Escalation of commitment (sometimes loosely called “commitment bias”) is a closely related but distinct idea. It is a behavioral pattern: continuing to invest in a course of action after receiving evidence that it’s failing. Sunk costs are one driver of it, but research since Barry Staw’s 1976 study has found several others, including personal responsibility for the original decision, threats to the decision maker’s self-image, how close the project is to completion, and social and organizational pressures. The sunk cost fallacy, in turn, can happen without any failing project: the concert ticket above involves no negative feedback at all. Projects that overrun their original plans (see Planning fallacy) are a common place where the two meet.

Several remedies have been tested, with mixed results.

• Knowing the principle isn’t always enough, but training on it helped. In Arkes and Blumer’s studies, students who had taken economics courses showed the effect as much as those who hadn’t. Larrick, Morgan and Nisbett (1990) instead gave people about half an hour of training on the rule that only future costs and benefits count, using everyday examples. Trained participants chose the normative answer far more often on hypothetical problems, including problems unlike the training examples. Differences in their reported behavior a month later were in the same direction but weaker.
• Deciding in advance when to stop helped. In a business-case experiment with 193 students, Simonson and Staw (1992) found three procedures reduced commitment to a losing course: setting a minimum target in advance that would trigger a change of course if missed, evaluating people on the quality of their decision process rather than the outcome, and making a bad outcome less threatening to the decision maker.

In practice, the tested advice amounts to this: ask what the options cost and deliver from here, and set stopping points before you’re invested enough to argue with them.

Status: contested. The effect in hypothetical choices meets this site’s bar for “robust”: it has survived a preregistered, multi-lab replication. But the entry’s claim is about how people decide, and evidence from decisions with real money at stake is mixed, including one incentivized study that found the opposite effect. Where the evidence falls between two labels, the standard is to choose the less confident one.

The foundational work. Staw (1976) had 240 business students role-play a business investment decision. They committed the most additional money to a course of action when they had personally chosen it and it had then turned out badly. Arkes and Blumer (1985) presented a series of experiments. In the best known, a university theater randomly gave unannounced discounts to some season-ticket buyers; those who had paid full price attended more plays over the first half of the season, though the difference didn’t persist through the second half. Questionnaire studies, such as choosing between a cheaper ski trip you’d enjoy more and a pricier one you’d already paid more for, showed the same pull of past spending.

Replications in hypothetical choices.

• Many Labs (Klein and colleagues, 2014) preregistered a replication across 36 samples and about 6,300 participants. People imagined having a ticket to an important football game on a freezing day; those told they’d paid for the ticket were more likely to say they’d go than those told it was free. The effect replicated at about the original size, which is small (d ≈ 0.27): only half the individual samples found it on their own.
• Petrov and colleagues (2023), in a Registered Report with 821 participants, replicated sunk cost effects for money. The earlier finding that sunk time has a weaker effect held in one study but not in another.
• Roth, Robbert and Straus (2015) meta-analyzed 98 effect sizes and found clear evidence of the effect, with a size that depended on the type of decision. It weakened with time in decisions about using something already paid for, and older adults showed it less.

Reasons for doubt.

• Behavior rather than scenarios. Friedman and colleagues (2007) built a computer game in which players dug for treasure and could pay a cost to move to a new island, and tested whether they stayed longer on islands that had cost more to reach. They found a “surprisingly small” sunk cost effect that didn’t respond to the psychological factors thought to drive it. Negrini, Riedl and Wibral (2022) found a reverse effect in an incentivized investment task (larger past investments made people less likely to continue), even though the same participants showed the standard effect on hypothetical scenarios.
• Confounds. Sleesman and colleagues (2012), meta-analyzing escalation studies, found sunk costs associated with escalation overall. But many scenarios bundled high spending with a project being nearly finished, and in studies that didn’t, the sunk cost effect’s confidence interval included zero. The authors concluded it “may not be as robust as the literature would suggest”, while noting only a handful of studies allowed the comparison.
• Measurement. Białek and Biesiada (2025) found that classic sunk cost scenarios, including several from Arkes and Blumer, barely correlate with one another, so a person’s answer to one says little about their answer to the next.
• Rational explanations. McAfee, Mialon and Mialon (2010) argued that in many real situations, responding to sunk costs is rational because of the information they carry, reputational concerns, or financial and time constraints. Walton (2002) analyzed the “argument from waste” and concluded it is not always fallacious and can be a rational precommitment strategy.

Field evidence exists but is modest. Ho, Png and Reza (2018) used changes in the up-front cost of owning a car in Singapore and estimated that higher sunk costs led to more driving, an effect that faded over time. The effect was small relative to the money involved; the authors linked it to about 5.6% more driving under the policies of 2009–2013.

What remains uncertain. Whether sunk costs, separated from information, reputation, project completion and constraints, move real decisions by a meaningful amount, and in which settings. That people say past spending matters in hypothetical choices is well established.