Sunk Cost Fallacy

Descriptive

Full Practice · Foundation · Watching Your Own Reasoning

Mechanism

The sunk cost fallacy is the systematic tendency to let unrecoverable past investments — money already spent, time already given, effort already exhausted — influence decisions about whether to continue a course of action, when only the expected future returns from this point forward should govern the decision. The mistake is not in caring about cost; the mistake is in caring about a cost that no choice can recover. The dollar already spent is gone whether you continue or quit. The five years already given to the failing project are gone whether you finish it or stop. Future-oriented reasoning treats sunk costs as zero. The mind, by default, does not.

The benchmark for the bias is the rational-actor decision model: at each decision point, look at the costs and benefits of continuing from here, look at the costs and benefits of stopping from here, choose the higher-expected-value option. Past investment does not enter the calculation, because past investment is not a future cost or benefit. The model is straightforward to state and notoriously hard to follow. People in carefully designed laboratory studies, given explicit framings that make the sunk cost irrelevance obvious, still let the past investment tilt their choice. People in real life, where the framings are not explicit and the affective stakes are higher, do this dramatically more.

The bias has multiple drivers operating together. The most studied is loss aversion: losses loom larger than equivalent gains, and abandoning a course of action in which one has already invested feels like accepting a loss, even though the loss has already occurred and continuing does not recover it. The mind frames "stopping now" as "losing the investment" rather than as "preventing the additional loss of continuing." The framing is partly arbitrary — both are accurate descriptions — but the loss-aversion framing dominates intuition.

A second driver is the consistency motive. The mind has a deep preference for behavior that aligns with prior decisions, partly because consistency is a cooperative signal to others and partly because reversing course is a public admission that the prior decision was wrong. Continuing a failing course preserves the public posture that the prior decision was correct. Stopping concedes the prior was wrong, even when the concession is the better available move from here. The consistency motive makes the sunk cost fallacy especially powerful for decisions that were public, identity-bound, or made over the objections of others, because reversing them carries social cost in addition to the decisional cost.

A third driver is identity. A person who has spent ten years pursuing a career, a relationship, a research program, or a creative project has, by that point, a sense of self that is partly built around the pursuit. Quitting is not just abandoning the project; it is dismantling part of who they have become. The cognitive operation that should ask "what is the best move from here?" gets routed through "who am I if I stop?", and the latter question has very different answers than the former. The fusion of identity with course-of-action is what makes the sunk cost fallacy hardest to break in exactly the cases — long-term careers, long-term relationships, identity-bound pursuits — where breaking it would prevent the largest accumulated loss.

A fourth driver is the asymmetric availability of the past. Past investments are vivid: you remember writing the check, putting in the hours, defending the choice. The future expected value is abstract — it is a projection, with uncertainty. The mind weighs the vivid past more heavily than the abstract future even when the math runs the other way, because the past has concrete representations that the future does not. The availability heuristic is operating quietly inside the sunk cost decision: the past costs are easier to retrieve than the future costs, and the retrieval ease shapes the felt weight.

The empirical finding worth holding is that even being explicitly told about the sunk cost fallacy does not eliminate it. Subjects in laboratory studies who have been taught the fallacy, asked to identify it in scenarios, and shown that they have identified it correctly, still exhibit the fallacy in subsequent decisions with structurally identical features. Understanding the bias does not produce immunity to it. The understanding is necessary; it is not sufficient. The corrective is the practice — applied repeatedly, at each decision point where past investment is present and creating the pull toward continuation.

For the Meridian Range, the sunk cost fallacy is a quiet driver of Control — of the inability to update when updating is what the evidence requires. A leader who has committed publicly to a strategy will continue it past the point where new evidence has made it clearly the wrong strategy, because abandoning it would concede the prior was wrong. A relationship maintained because of years already invested, even though both parties have known for some time it is not working, is the same bias. A project finished because too much has already been spent to walk away, even though finishing will cost more than the finished product will be worth, is the same bias. The mind that holds the Meridian Range has to be able to walk away from past investments when the future does not support them, and the walking away is psychologically expensive in ways that have to be noticed and overridden.

Practice

The core diagnostic question is this: "If I were arriving at this situation fresh today, would I make this commitment?"

The question forces the reasoner to set the past investment to zero and evaluate the situation as a new decision. If the answer is "no, I would not start this from where things now stand," then continuing is the sunk cost fallacy operating, and the past investment is the thing telling you to do what you otherwise would not. If the answer is "yes, I would still make this commitment from scratch given everything I now know," then continuing is not the fallacy; it is a forward-justified choice that happens to coincide with the past investment. The question separates the two cases. Only the first case requires correction.

A caution about the corrective. The watching can drift into a kind of compulsive abandonment, where the practitioner treats any past investment as a reason for suspicion and reaches for stopping or pivoting too readily. Persistence has value. Many successful projects, relationships, and pursuits required pushing through periods where the immediate evidence did not justify continuing, and where a strict forward-expected-value analysis would have produced a wrong call. The discipline is not to ignore the past investment; it is to know whether the past investment is informing the forward analysis (because it has built capabilities, learning, infrastructure, or position that change the forward returns) or whether it is operating as a separate input that the forward analysis should not be weighted by. The first is legitimate; the second is the bias.

In the Wild

A founder had raised significant funding and spent three years building a product that was not finding traction in the market. Every pivot suggested by advisors and investors had been rejected because each pivot would have meant acknowledging that the original vision was not working. He had defended the vision publicly many times. The team had reorganized around it. He had described it in pitches to a hundred investors. When the company finally ran out of money, he reflected that he had recognized eighteen months earlier that the original direction was not viable. He had spent the eighteen months trying to make a non-viable direction work, because admitting it was non-viable would have meant admitting the previous eighteen months of work had been wrong. By the time the bias was no longer survivable, the runway was gone.

A reader had a partner she had been with for nine years. The relationship had been good for the first four and had been steadily declining for the last five. Friends had begun to notice; her own self-assessment, when she let it run, was that the relationship had become a source of consistent low-grade unhappiness with no plausible path back to its earlier state. She did not leave. Every time she tried to evaluate the decision, the nine years entered the calculation: she had spent so much already, leaving would mean conceding the nine years had ended badly. When she eventually did leave, she described the experience to a therapist as having been held in place by years rather than by feelings. The years were the sunk cost. The feelings had been telling her something different for a long time.

A government had been pursuing a long-running infrastructure project that had become a public commitment of administration after administration. Three independent reviews had concluded the project's expected value, from current cost to completion, was substantially negative. The project continued. Each administration that received the reviews concluded, partly correctly, that cancellation would carry political costs that the prior administrations had not had to bear. The political cost of cancellation was real. It was also a sunk-cost-adjacent dynamic: previous administrations' commitments were creating decisional inertia that locked in continued losses. The project was eventually completed, at roughly three times its original budget. The completed asset's operating returns confirmed the reviews. The cancellation that should have happened a decade earlier did not, because the political mechanism of acknowledging a prior administration's error was harder than the financial mechanism of continuing to lose money.

Take a course of action you are continuing — a project, a relationship, a position, a pursuit. Ask yourself whether you would begin it today, from scratch, knowing what you now know.

Lineage

Hal Arkes and Catherine Blumer's "The Psychology of Sunk Cost," published in Organizational Behavior and Human Decision Processes in 1985, is the canonical experimental paper. Arkes and Blumer demonstrated the effect across a series of clean experimental designs, including the famous theater-ticket scenarios that have anchored teaching of the bias ever since. Their paper also offered an early account of the underlying mechanisms, including loss aversion and the desire to avoid appearing wasteful.

Richard Thaler's work on mental accounting provided the broader theoretical context. The Winner's Curse (1992) and the underlying papers established mental accounting as a systematic departure from rational decision theory: people organize transactions and investments into mental categories, and sunk costs entered into a category are not erased by the recognition that they are sunk. The mental account remains open, and the open account exerts pull on subsequent decisions related to it. Thaler's framing helps explain why the bias persists even after the math has been done: the math operates in a different mental register than the account it should override.

Barry Staw's research on escalation of commitment, beginning in the 1970s, extended the bias from individual decisions into organizational and institutional contexts. "Knee-Deep in the Big Muddy" (1976) demonstrated experimentally that decision-makers who had personally authorized an initial investment escalated their commitment to that investment when it failed, while decision-makers who had inherited the same investment did not. The personal authorship of the prior decision was a major driver of the bias. The implication is that the watching is most needed when the practitioner is evaluating their own past decisions, and most easily applied when evaluating someone else's.

Daniel Kahneman and Amos Tversky's prospect theory (1979) supplied the loss-aversion architecture. The asymmetric weight of losses relative to gains, plus the reference-point dependence of loss perception, explains why sunk costs feel different from forward costs: a sunk investment is anchoring a reference point, and any move that closes the position at less than the reference is experienced as a loss to be avoided. The math is doing what the math does; the mind is computing the loss in a frame that the math does not require.

A caution from the more recent literature. Several authors have argued that what looks like sunk cost reasoning is sometimes a forward-looking response to reputational, signaling, or commitment-device concerns that the simple expected-value model does not capture. A leader who continues a failing project to preserve their reputation for resolve is making a forward calculation, even if the calculation is troubling. The Codex's framing is that this qualification matters: not all persistence in the face of losses is the bias, and the watching has to distinguish forward-justified continuation from past-anchored continuation. The fresh-start question is the cleanest discriminator: if you would not begin from current conditions, the continuation is not forward-justified, whatever story is being told about why.

Cross-references

Within the category. Noticing catches the moment the past investment is doing the decisional work — the felt resistance to walking away, the urge to reframe stopping as betrayal of the prior effort. Motivated Reasoning is the close sibling: the past investment supplies the motivation, and the reasoner constructs arguments for why continuing is justified. The two together are most of the actual cognitive operation of the sunk cost fallacy; the watching has to catch both. Affect Heuristic operates as well — the affective tag on the prior commitment shapes the assessment of the continuation, often without the reasoner noticing.

Within the Foundation. Holding Beliefs Without Identity addresses the deepest fuel for the fallacy: when a course of action has fused with identity, walking away from the action becomes walking away from the self, and the cost of the corrective rises beyond what most people can pay. Identity decoupling, when its profile is written, is the upstream work that makes sunk-cost recovery possible. Revising Beliefs Under Evidence is the home of the larger discipline: a belief about a course of action is a belief, and the practice of revising beliefs when evidence warrants applies to commitments as much as to claims. The Update Protocol is the structured method for the actual move once the fresh-start question has produced its answer.

Across to Knowledge. Reading What's Operating carries the systems-level analogue: institutions and movements continue failing courses of action for the same structural reasons individuals do — past investment, identity, public commitment, the asymmetric cost of acknowledging prior error. A practitioner who has watched the bias firing in themselves can read it in the systems around them, and the reading is often more useful than the introspection.

Limitation. The bias is hardest to defeat in exactly the cases where it matters most: long-term investments, public commitments, identity-fused pursuits. The fresh-start question is straightforward to ask and brutal to answer honestly when the honest answer requires walking away from years of one's own life. Practitioners who have caught the bias in low-stakes contexts often still miss it in the high-stakes contexts that define them. The honest stance is that some sunk-cost recovery is only possible with outside help — a trusted dissenter who can see the situation without carrying the investment — and that part of the watching is knowing when to ask.