Skip to main content
business computer-science economics engineering-and-technology political-science psychology

​
Sunk cost fallacy

​
Description

A decision-making error in which unrecoverable prior investment is allowed to influence a forward-looking choice whose expected value depends only on remaining costs and remaining benefits. The diagnostic question — “if I were arriving at this situation fresh, with no history, would I make the same choice?” — strips the prior spend from the calculation and exposes whether the continuation is genuinely justified or is being propped up by the leakage. The structural property that makes this a fallacy and not merely an error in judgment: the variable that has entered the calculation (prior spend) is known to be irrelevant to the forward decision under standard decision theory, yet the cognitive pull persists, often even when the decider has been explicitly told to disregard it. Sunk-cost-fallacy is the family of errors produced by that pull. The pull’s strength scales with the magnitude and visibility of the prior investment. Small, private, easily-forgotten investments produce weak pulls; large, public, costly-to-acknowledge investments produce strong pulls. The latter case shades into escalation-of-commitment dynamics where the decider’s own prior commitment becomes a trap.

​
Triggers

User-initiated: User describes a decision being shaped by prior investment, by reluctance to “waste” earlier spend, or by escalating commitment to a course of action whose forward case has eroded. Vocabulary cues: “sunk cost,” “we have come this far,” “cannot stop now,” “throwing good money after bad,” “already invested too much.” Agent-initiated: Agent notices an apparent justification for continuing a course of action that hinges on past expenditure rather than forward expected value. Candidate inference: “is the prior spend doing real work in this decision, or is it leaking into a forward calculation?” Situation-shape signals: Project-rescue decisions; abandon-vs-continue forks in long-running efforts; financial loss-cutting moments; long-relationship reassessments; organizational reviews of programs whose champions have publicly committed to them. The signal is strongest when the discussion centers the past instead of the future.

​
Exclusions

  • Information value of completing the work — sometimes finishing a project produces information (a publishable null result, a learning, a credential, a portfolio piece) whose forward value justifies completion. This is not sunk-cost-fallacy because the prior investment has reshaped the future option set; the spend is doing forward work, not just historical work. Diagnostic: would you pay to acquire the same information from scratch?
  • Reputation and commitment as future assets — when prior public commitment creates ongoing reputational stakes (delivering on a promise, honoring a contract, sustaining a strategic alliance), continuation is sometimes rational because the reputational asset is real and forward-relevant. The decision is genuinely about future payoffs, not the past spend.
  • Network-effect lock-in where switching cost exceeds remaining value — staying on a less-preferred platform because everyone else is on it is not sunk-cost-fallacy; it is a rational response to a coordination externality. The cost of switching is a real forward cost, not a sunk historical one.
  • Honest re-evaluation that happens to reach the same conclusion — sometimes the math really does favor continuation, and the appearance of sunk-cost-pull is just observers projecting bias onto what is actually a defensible call. The fallacy requires the prior spend to be doing the work; sometimes the forward case is simply solid.

​
Structure

Internal structure of sunk-cost-fallacy: a table of its component slots and the concepts that fill them.

​
Relationships

Relationship neighborhood of sunk-cost-fallacy: a graph of the concepts it connects to and the concepts it is a part of.
  • lifecycle-cost — sunk-cost-fallacy lets unrecoverable past spend drive forward decisions; lifecycle-cost is forward-looking total-cost discipline that ignores sunk spend — opposite temporal orientation.
  • loss-aversion — partial mechanistic explanation. Abandoning a project converts the prior spend from “investment-in-progress” to “realized loss,” and loss-aversion makes that realized loss psychologically heavier than the same magnitude of future loss from continuing. Reading them together explains why the fallacy persists when the math clearly says to stop.
  • doctrine — named explicit doctrines exist as structural counter-pressure: “kill your darlings,” “the next dollar is what matters,” “do not throw good money after bad.” The doctrine names the move that resists the pull.
  • hoist-by-own-petard — when escalation continues because the decider’s earlier public commitment now constrains them, the prior commitment is the petard. Sunk-cost composes with own-petard when the past investment is also a binding public stake.
  • mean-reversion — the trader who refuses to sell because “it will come back” is fighting mean-reversion (the position’s losses are likely permanent) on sunk-cost grounds. The two concepts together explain why mean-reversion strategies require pre-committed loss-cutting rules.
  • reframe — the corrective move is often a reframe: from “we have invested X already” to “the next dollar is the only one that matters.” The reframe is what enables the sunk-cost-corrected decision; without the reframe, the original frame keeps the prior spend salient.

​
Examples

The Concorde supersonic program · engineering-and-technology

the canonical case: the UK and French governments continued investing in the Concorde long after the project’s economic case had collapsed, in part because they had already invested so much. The episode is so emblematic the bias is sometimes called the “Concorde fallacy.”

Gambling escalation · economics

chasing losses: increasing bet size to recover prior losses, rather than evaluating the next bet on its own merits. The pattern is structurally identical to organizational escalation.
completing a degree program because of the years already invested, when the forward calculation (additional years + opportunity cost vs. value of the credential) does not justify continuation.
Arkes and Blumer’s 1985 paper is the canonical empirical demonstration of the sunk-cost fallacy as a robust cognitive phenomenon. Across a series of experiments — most famously a hypothetical ski-trip vignette in which subjects had to choose between a $100 trip and a $50 trip after having paid for the more-expensive option — they showed that prior expenditure systematically pulled forward choices toward continuation, even when the forward expected value of the alternative was higher. The effect held across stake sizes, framings, and respondent populations, and persisted when subjects were explicitly informed that the prior spend was irrelevant to the forward decision.The paper named the phenomenon as a fallacy in the technical sense: a systematic deviation from the prescription of standard expected-utility theory, attributable to a specific cognitive mechanism (the leakage of irrecoverable past costs into a forward calculation that should depend only on future costs and benefits). The empirical robustness — surviving stake-size variation, instruction to disregard, and population diversity — established sunk-cost-fallacy as a load-bearing target for behavioral-decision research and a recurring concern for organizational decision-making.Inference: When a decision-maker’s stated reasoning leans on prior expenditure (“we have already spent so much”), the diagnostic move is to reformulate the choice from the perspective of an outsider arriving fresh with no investment history. If the outsider would choose differently, the sunk-cost pull is doing structural work in the decision and the decision is at risk; if the outsider would also continue, the forward case is independently solid.
staying in a relationship “because we have been together so long” against an honest assessment that the future trajectory is negative; relationship counselors routinely surface the sunk-cost framing as the trap.
continuing to invest in a product because of historical R&D, despite forward unit economics that would never justify launching it today. Common in pharma, hardware, and entertainment.
engineering teams continue investing in a rewrite or rescue effort because of the months already spent, even when starting fresh would be cheaper from where they sit; the explicit doctrine “kill your darlings” exists precisely to counter this pull.
Barry Staw’s 1976 study is the foundational laboratory demonstration of escalation of commitment — the empirical pattern that decision-makers who have personally committed to a failing course of action invest more in continuation than decision-makers without that prior commitment, even when the forward economics are identical. Business-school subjects were given a capital-allocation scenario and asked to divide an additional R&D budget between two divisions of a hypothetical firm; some subjects had previously made the initial allocation themselves (and one division was subsequently performing badly), others had not. Those who had made the original choice put significantly more money into the losing division than those who came to the situation fresh. The title — borrowed from Pete Seeger’s protest song about a sergeant marching his platoon deeper into a swamp — captures the dynamic: personal responsibility for the prior decision becomes a force that drives further commitment to the now-failing path. Staw’s later work (and the broader escalation-of-commitment literature it spawned) showed the effect holding across labor disputes, defense procurement, and venture-capital follow-on rounds.Inference: The structural correction is to separate the decider for continuation from the decider for initiation. When the same person who launched a program is also asked whether to fund its next phase, the prior commitment leaks into the forward calculation; when a different person — or a stage-gate committee, or a written-in-advance kill criterion — makes the continuation call, the leakage is structurally blocked. This is why mature venture-capital firms have explicit “no follow-on without independent re-underwrite” doctrines, why defense-acquisition reform includes Nunn-McCurdy breach reviews by parties outside the program office, and why effective project-portfolio governance reassigns kill-decisions to a board distinct from the program’s champions. The diagnostic question is: who is making the continuation decision, and were they responsible for the original commitment? If yes, build in the structural separator before the sunk-cost pull overwhelms the forward case.
Richard Thaler’s 1980 paper situated sunk-cost effects within a broader catalogue of consumer-choice anomalies inconsistent with standard economic theory. Building on Kahneman and Tversky’s prospect theory, Thaler identified sunk-cost behavior as one of several systematic departures from the expected-utility-maximizing agent — alongside endowment effects, mental accounting, and self-control failures. The paper was load-bearing in establishing that these were not isolated curiosities but a coherent family of phenomena reflecting how real consumers actually decide.Thaler’s specific contribution to the sunk-cost concept was the integration with mental accounting: consumers track expenditures within psychological accounts and resist closing an account at a loss. The unrecovered prior spend is not merely a memory; it is an open accounting entry that pressures the decider to “make it back” by continuing the activity that generated the spend. The mental-accounting frame explains why sunk-cost effects scale with the magnitude and visibility of the prior expenditure — small or hidden expenditures generate weak account-closure pressure; large or public ones generate strong pressure.Inference: To counter sunk-cost-driven decisions, the corrective often operates at the level of how the accounts are constructed, not at the level of individual rational override. Reframing — closing the prior account explicitly (“that money is gone; this is a new decision”) — can be more effective than reasoning. The mental-accounting frame predicts which decisions will be most vulnerable to sunk-cost: those in which the prior spend sits in a clearly-bounded mental account that the decider has not yet been able to close.
The Concorde is the emblematic sunk-cost case, and the story of how it got its name shows the fallacy crossing domains. The Anglo-French supersonic program (commercial service January 1976 to October 2003) is the textbook instance: by the early 1970s the economic case had collapsed — development costs ran an order of magnitude over the 1962 estimate, fuel consumption was punishing, the cabin held only ~100 passengers, and overland supersonic bans gutted the market. A 1971 British government memo reportedly called it a “commercial disaster” that “should never have been started.” Yet both governments kept funding it. Part of the reason was forward-looking (the treaty lacked a break clause, so either side could be sued for withdrawing), but the diagnostic core was retrospective: decision-makers felt too much had already been spent to stop.The fallacy’s name, though, was coined in biology. Richard Dawkins and Tamsin Carlisle (1976, Nature) used the Concorde as an analogy to criticize an account of animal parental investment — the idea that a parent’s decision to keep investing in offspring depends on how much it has already invested. They argued this was the “Concorde fallacy”: a rational organism, like a rational government, should decide based only on future expected payoff, never on past irrecoverable spend. The label then migrated back into economics and psychology. The cross-domain coinage is itself the point: the same forward-decision structure — payoff depends only on future costs and benefits, while past expenditure exerts an illegitimate pull — recurs identically in supersonic aircraft programs and in the reproductive decisions of birds, which is exactly why a single name travels across both.Inference: that an abstract decision-error gets named after a concrete engineering megaproject and is independently diagnosed in animal behavior is strong evidence the structure is domain-general, not an artifact of human institutions. The corrective is identical in every domain it appears: re-derive the decision from the current state forward, treating the accumulated investment — money, eggs, years — as informationally irrelevant to the go/no-go call.
the political-military literature on Vietnam treats the persistence of US involvement past the point of strategic justification as a sunk-cost-fallacy at state level; “we cannot have those lives have been lost in vain” is the diagnostic phrase.