What Is the Sunk Cost Fallacy?
The sunk cost fallacy is a pattern of judgment in which an irrecoverable past investment influences a current decision even though the investment cannot alter the comparative future value of the available alternatives. People may persist with a disappointing project, attend an unwanted event, keep using an inferior service, or allocate additional resources because they have already committed money, time, or effort. The scientific question is whether changing the amount of past investment changes subsequent choice while the relevant future consequences remain equivalent.
A sunk cost is an expenditure or commitment that cannot now be recovered. A nonrefundable ticket is a sunk monetary cost; the weeks already spent on a completed but unsuccessful preliminary design are sunk time and effort. A refundable deposit, an asset with resale value, a transferable license, or training that improves future performance has recoverable value or continuing consequences and must be evaluated differently. The key property is irreversibility at the moment of the new decision.
Consider someone who has paid $90 for a concert but now expects to enjoy a free gathering with friends more. If the $90 is nonrefundable regardless of the choice, spending the evening at the concert cannot retrieve it. The live options differ in anticipated enjoyment, travel, fatigue, and other future costs. Choosing the concert exclusively to avoid “wasting the ticket” is a clear candidate for the sunk cost fallacy. If attending would yield more enjoyment, fulfill a genuine obligation, or permit resale of another ticket, the same choice could be justified for reasons unrelated to sunk costs.
The classic research definition concerns an increased tendency to continue after an investment of money, time, or effort. Arkes and Blumer's foundational studies connected the pattern to the desire to avoid feeling wasteful and investigated paid theater subscriptions, hypothetical purchases, and judgments about continuing projects. Arkes and Blumer (1985)90049-4)
The term “fallacy” refers here to an empirical departure from a specified forward-looking decision criterion. It is not a formal logical fallacy in the strict sense of an invalid deductive argument, and displaying the pattern in one task is not a clinical symptom or psychiatric diagnosis. A cognitive bias describes a systematic feature of judgment under particular conditions, rather than a permanent property of an individual.
Why Should Irrecoverable Costs Be Excluded From a New Choice?
A normative decision compares the prospective consequences of the alternatives available now. For a simplified financial choice, the expected net value of continuing equals the expected future benefit minus the expected additional cost of continuing. The expected net value of stopping equals the expected future benefit of stopping or switching minus its future costs. An identical past payment appears on both sides of the comparison and cancels out.
Suppose a company has spent $4 million developing a product. It now estimates that completion requires another $1 million and that expected future receipts, appropriately adjusted for risk, are $700,000. Continuing has a prospective expected net value of negative $300,000 relative to stopping if stopping has no further cost or benefit. The historical $4 million cannot improve that forecast. With the same information, the correct calculation would be identical if initial development had cost $400,000 or $40 million.
A mathematically explicit model is to choose the available action a that maximizes E[U(future outcomes | a, current information)], subject to present constraints. Under a risk-neutral monetary model, U can be linear in future wealth; with risk aversion, liquidity constraints, uncertainty, and nonfinancial consequences, the utility function must reflect those factors. Sunk expenditures do not directly change the comparison unless they alter a relevant present state or future consequence. Thus “ignore sunk costs” is a conditional decision rule, not an injunction to ignore the entire history of a project.
This distinction matters because prior actions can create genuine continuing value. An unfinished project might now be highly likely to succeed, a substantial part of the necessary work may already be completed, or exiting might trigger contract penalties. Those facts affect future costs and benefits. Some information about previous investment also helps infer project quality, market commitment, or the cost of replacing a functioning system. Researchers must therefore distinguish the psychological effect of the irrecoverable expenditure itself from rational adaptation to information produced by that expenditure. Roth, Robbert, and Straus (2015)
Two Major Forms: Utilization and Continued Investment
Utilization decisions after payment
A utilization decision concerns whether to consume something already paid for or choose another available experience. Examples include attending a prepaid event, eating beyond comfort at an all-you-can-eat buffet, or using a service more intensively because its membership fee was high. The original payment may become psychologically coupled to usage, making nonuse feel like losing money twice.
The normative question is what additional satisfaction and costs each option entails from now onward. Sometimes consumption requires no further monetary outlay but carries opportunity costs, fatigue, travel, or reduced enjoyment of a better alternative. Since these consequences may be substantial, prepaid status alone cannot determine the choice. Research on payment depreciation suggests that the motivational force of a prior payment often diminishes as the interval between payment and consumption increases. Gourville and Soman (1998)
Progress decisions and commitment escalation
A progress decision concerns whether to invest still more resources in something already underway. Examples include completing an over-budget project, maintaining a failing advertising campaign, or extending a software rollout after adverse testing. The new decision requires a new investment and may have uncertain returns. The sunk cost fallacy is one possible reason decision-makers allocate further resources, but escalation of commitment is the broader behavioral phenomenon: persistence can also arise from incentives, responsibility, reputation, optimistic forecasts, and organizational politics.
Barry Staw's experimental research showed that negative feedback could increase investment in an initially chosen business direction, especially when participants were personally responsible for earlier decisions. That finding helped establish the self-justification research program, but personal responsibility is not identical to a controlled monetary sunk-cost manipulation. Staw (1976)90005-2)
The distinction is crucial for interpreting effect estimates. A study of prepaid attendance and a study of organizational funding may both reveal path dependence, yet the incentives, feedback structure, and possible causes are different. A meta-analytic review explicitly separated utilization from progress decisions rather than treating every continuation as the same psychological event. Roth, Robbert, and Straus (2015)
Origins of the Scientific Research
Economic reasoning about sunk costs predates the psychological experiments. Behavioral economists challenged the assumption that real consumers invariably evaluate options using only prospective utility. Thaler's work on mental accounting described how people create subjective accounts for purchases and mentally connect the pain of paying with the pleasure or obligation of consumption. This offers one explanation for why a prepaid ticket exerts pressure long after the purchase decision. Thaler (1980)90051-7)
Staw's 1976 study of commitment escalation highlighted the causal role of personal responsibility under negative outcomes. Arkes and Blumer's 1985 article then consolidated the sunk-cost phenomenon across experimental vignettes and a field context. Their theater-subscription study found greater subsequent attendance among subscribers who initially paid more, consistent with a payment-linked influence on behavior. The result is historically important, while its field design requires attention to possible alternative explanations and the context-specific character of attendance. Arkes and Blumer (1985)90049-4)
Subsequent decades introduced experimental decision games, consumption studies, organizational escalation paradigms, theoretical models, and evidence syntheses. Some experiments found that participants largely disregarded historical spending under transparent incentives. This produced a more discriminating scientific question: which representations of irreversible cost, counterfactual opportunities, time intervals, and tasks reliably cause persistence? Roth, Robbert, and Straus (2015)
Psychological Mechanisms
Mental accounting and the feeling of waste
Mental accounting organizes transactions into informal psychological budgets. Payment may open a subjective account that feels unresolved until the associated product or experience has been used. If someone declines to attend a prepaid performance, the transaction can feel unfinished and the unrecovered payment becomes psychologically salient. Using the ticket may close that mental account even when another activity would be more pleasant.
This is an explanatory model rather than a universal mechanism. Individuals differ in how readily they remember expenditures, connect particular payments to particular experiences, and treat use as a symbolic recovery of value. The time separating payment from use changes that connection. Experimental work calls this payment depreciation: the behavior associated with an earlier payment may weaken over time. Gourville and Soman (1998)
Reference points and anticipated losses
A completed purchase or substantial investment can establish a reference point against which cancellation is coded as a loss. A continuation option may retain a psychologically appealing chance of recovering the initial investment, while stopping feels like accepting failure. Prospect theory offers an influential framework for evaluating gains and losses around reference points, although demonstrations of the sunk-cost effect do not themselves establish a particular loss-aversion parameter. Kahneman and Tversky (1979)
Reference dependence creates different predictions depending on the framing of the options. The fact that loss aversion is a prominent account should not be translated into the claim that all sunk-cost behavior has been causally traced to loss aversion. The more specific empirical task is to manipulate reference points or gain–loss framing while holding relevant payoffs constant.
Self-justification, dissonance, and public responsibility
A decision can become linked to perceived competence and personal identity. Canceling a self-chosen project may feel like admitting that the earlier decision was mistaken, particularly if others know who advocated for it. Continuing can defer that judgment. Such motives are prominent in escalation-of-commitment studies that vary responsibility or accountability. Staw (1976)90005-2)
The relevant causal paths can coexist. A manager might overvalue past spending, anticipate embarrassment, and exaggerate prospects of success. Even when observed continuation is economically poor, a single behavioral observation cannot identify which of these processes produced it. Controlled manipulations and process measures are necessary to distinguish them.
Optimistic beliefs about success
Past investments can shift beliefs about the likelihood of eventual success. Instead of explicitly arguing that past expenditures ought to count, a decision-maker may conclude that a heavily funded project must be close to success or worth preserving. This reasoning can be mistaken when the amount already spent contains no new information about future quality. Arkes and Blumer reported inflated success estimates in some invested-participant conditions, making belief distortion a plausible accompanying process. Arkes and Blumer (1985)90049-4)
Researchers should keep belief formation distinct from preferences. A person who genuinely assigns a higher probability to success makes a different subjective calculation from someone who believes the prospects are poor but continues to protect the perceived value of earlier spending. Both pathways can generate the same action and require different interventions.
Effort, identity, and the meaning of commitment
Time and effort have several functions. They can build competence, create sunk emotional attachment, produce personal meaning, and establish real future obligations. The mere fact that someone values an activity after working hard on it does not prove a bias. To demonstrate one, an experiment must separate the productive consequences of effort and the current value of the activity from the irrecoverable effort itself.
This is particularly important for personal goals, creative work, and educational decisions. The relevant assessment includes the value of acquired knowledge, future opportunities, remaining learning, and the possibility of transferring skills elsewhere. The slogan “you have already put years into it” offers insufficient evidence for either continuing or stopping.
What Do Experiments Actually Measure?
Randomized sunk-cost contrasts
A direct experiment varies a prior, nonrecoverable investment while keeping expected future outcomes and other causal inputs constant. One condition might receive a high-priced prepaid ticket and another an equivalent low-priced ticket; both then confront the same available alternatives. The estimated causal effect can be expressed as the difference in continuation rates or in expected time spent on the original activity.
The manipulation must be interpreted carefully. A higher price may also signal superior quality, prestige, or a different opportunity cost. If participants were allowed to select what to purchase, differences in preexisting preferences can contaminate estimates. Credible experiments control information, assignment, recoverability, and comparability of options.
Hypothetical vignettes and consequential decisions
Questionnaire vignettes ask participants to imagine a business project, trip, membership, or other commitment. They permit controlled comparisons at relatively low cost and can investigate beliefs and justifications. Their outcome is stated choice under hypothetical conditions, which should be distinguished from actual spending or behavior with material incentives.
Consequence-bearing laboratory tasks permit participants to invest real effort or money, then choose how long to continue. Field studies observe behavior under naturally occurring or randomized payments. Each method has trade-offs: vignettes offer experimental precision, laboratory tasks offer behavioral outcomes in restricted settings, and field evidence adds realism with additional potential confounding.
Correct benchmarks, outcome measures, and effect sizes
For a simple choice, investigators may compare the proportion choosing to continue in high-cost and low-cost conditions. For continuous outcomes, they may report minutes of persistence, dollars reinvested, or standardized mean differences. Some studies use willingness to pay or forecasts of success. The behavioral estimate is interpretable only alongside the task's benchmark: additional expected monetary value, subjective utility, or experimentally controlled preference.
A statistically reliable change in persistence establishes sensitivity to the manipulated history. Calling it irrational additionally requires showing that the history changed neither useful information nor relevant prospective consequences. Without that second step, the study identifies a sunk-cost-related effect more securely than a normative error.
Classical Findings and the Meta-Analytic Evidence
Arkes and Blumer's original program remains the defining experimental reference. Their theater-subscription field component tied an initial payment difference to attendance over subsequent months, while their scenarios investigated continuation, use, and judgment after irreversible commitment. The experiments helped turn a familiar anecdote into a testable behavioral phenomenon. Arkes and Blumer (1985)90049-4)
Roth, Robbert, and Straus synthesized research published from 1976 through 2013. Their meta-analysis ultimately included 45 studies and 100 analyzable effect sizes across monetary sunk-cost paradigms. The pooled standardized mean effect was approximately 0.50 under their effect-size convention, and heterogeneity was substantial. That value summarizes a selected body of paradigms rather than the probability that any particular individual will make a sunk-cost error in daily life. The paper's abstract describes 98 effect sizes, while its methods and final pooled analysis discuss 100 usable effects; the unit and eligibility details must therefore be read in context rather than collapsing all counts into one unsupported statement. Roth, Robbert, and Straus (2015)
Their results supported a general sunk-cost effect while emphasizing differences between use/consumption decisions and further-investment decisions. Payment effects in utilization choices tended to attenuate as time elapsed; age and sampling characteristics also moderated estimates. Education in economics did not produce a consistent protection across the included settings. Crucially, the pooled results incorporated heterogeneous designs and in some instances net effects that could contain other mechanisms, particularly in project-progress tasks. Roth, Robbert, and Straus (2015)
A separate meta-analytic literature studies escalation of commitment, where organizational, social, project, and psychological drivers jointly predict continuing investment after unfavorable feedback. Such studies illuminate real-world persistence but should not be substituted for trials manipulating sunk expenditures alone. Sleesman, Conlon, McNamara, and Miles (2012)
What Newer Research Has Changed
Preregistered real-effort and digital-field studies in 2026
A 2026 paper by Beknazar-Yuzbashev, Ichiba, and Stalinski tested sunk-cost predictions in two preregistered experiments and developed a reference-dependent account of why standard discount manipulations may miss the mechanism. In an online real-effort study of 1,806 participants, experimentally varying a ticket price by $2 increased participation time in a 14-minute task by approximately 1.3 minutes, a reported effect of 0.12 standard deviations. The authors also conducted a YouTube-based field experiment involving 11,328 videos and variation in when a pre-video advertisement became skippable. The results speak to specific forms of time and monetary commitment rather than establishing a constant effect across all settings. Beknazar-Yuzbashev, Ichiba, and Stalinski (2026)
These designs are valuable because real effort and digital behavior can address some limitations of vignette-only evidence. At the same time, the online task's small time horizon, experiment-specific incentives, and platform-defined behavior limit direct extrapolation to expensive corporate projects or personal relationships. The appropriate reading is that carefully designed causal tests can identify sunk-cost responses, and their measured magnitude is tied to a particular intervention and decision environment.
Childhood and development across the lifespan
A 2025 study used two experiments, with samples of 682 and 378 participants spanning ages 2 through 97, to investigate sunk-cost judgments with the same measurement families across ages. Adolescents and adults showed sunk-cost judgments more consistently than children, and different vignette measures generated different age patterns. The study challenges any simple claim that the fallacy must begin early in childhood or that susceptibility decreases monotonically throughout life. Hamzagic, Mah, Derksen, and Bernstein (2025)
Another 2025 paper compared younger and older adults using more consequential and hypothetical decisions to revisit previously reported age differences. Its findings did not support a straightforward interpretation that older adults are always more economically rational; stimulus content and choice format mattered. Age is therefore best treated as a possible moderator rather than a diagnostic shortcut for predicting an individual's behavior. Pighin, Timberlake, and Tentori (2025)
Does spending time function like spending money?
Irrecoverable time is conceptually a sunk cost, but its mental representation may differ from money. Soman's experiments found that previous time investments did not generate the same effect under some direct time framings, whereas converting the time into monetary terms restored an effect. This cautions against treating one dollar spent, one hour invested, and one unit of physical effort as interchangeable manipulations. Soman (2001)
Time also has productive and personally meaningful consequences that are difficult to hold fixed. A person who has practiced piano for years may now play well; the acquired skill is a current asset even though the hours cannot be recovered. Empirical measurement must isolate the psychological weight of past effort from the genuine future benefit produced by practicing.
Why the Effect Varies Across Situations
The amount of money previously spent, whether the amount is framed as a discount or a loss, and whether the decision concerns use or continued investment can all influence measured persistence. Delay between payment and choice, proximity to completion, beliefs about quality, available alternatives, and whether the original actor is personally accountable also matter. Cross-study variation is a scientific finding in itself, rather than an inconvenience that can be removed by averaging. Roth, Robbert, and Straus (2015)
In an organization, further spending can be affected by who approves budgets, the incentives for admitting failure, and how project information is presented. In everyday consumption, the salient variables may instead be enjoyment, inconvenience, guilt about waste, and the elapsed time since payment. A decision aid designed for one setting should therefore be evaluated in that setting before being generalized.
National or cultural differences are plausible because social responsibility, norms concerning waste, face-saving, and money management vary. However, common cross-country anecdotes cannot establish the direction or magnitude of cultural effects. Strong comparative tests require equivalently translated instruments, comparable incentives, representative samples, and explicit analysis of measurement invariance.
Sunk Cost Fallacy Versus Neighboring Concepts
Escalation of commitment
Escalation is the allocation of additional resources to a prior choice despite unfavorable feedback. Sunk costs may contribute, but continuation can also express responsibility, reputation management, or genuinely updated expectations. A project may escalate even when its manager pays little attention to historic spending. The constructs overlap empirically and have different explanatory scope.
Loss aversion and the endowment effect
Loss aversion concerns how losses and gains are evaluated around a reference point; the endowment effect concerns valuation changes associated with possession. Both can make abandoning a paid-for option unattractive, but neither is synonymous with sensitivity to irrecoverable expenditure. Identifying loss aversion requires appropriately designed valuation or choice comparisons, rather than inferring it from persistence alone.
Status quo bias and switching costs
Status quo bias is a tendency to favor an existing option under certain conditions. Switching costs are real prospective costs—learning another tool, migrating data, renegotiating contracts, or losing access during a transition. A decision to remain with a familiar service may combine both influences. Any analysis that categorically calls continued use a sunk-cost error while overlooking actual transition expenses fails its normative benchmark.
Present bias and the planning fallacy
Present bias concerns disproportionate preference for outcomes close in time; the planning fallacy concerns forecasts that underestimate the demands of future work. They can affect a project with sunk spending, but they are separate mechanisms. An optimistic completion estimate can make further investment attractive even if the decision-maker is not motivated by previous spending.
Cognitive dissonance and clinical symptoms
Cognitive dissonance describes psychological tension among conflicting cognitions and may contribute to self-justifying persistence. It is an explanatory construct and does not prove that every sunk-cost decision arises from discomfort. Sunk-cost behavior itself is a judgment phenomenon, not a diagnostic category. Excessive persistence may accompany many personal circumstances, but it does not establish depression, obsessive-compulsive disorder, mania, addiction, or any other condition.
Everyday Examples and Their Decision Criteria
Education and careers
A student may consider completing a degree because several semesters of tuition and effort have already been spent. A forward-looking assessment compares the remaining tuition, likely educational benefits, career opportunities, alternatives, and personal goals. Credits that transfer, professional networks, accumulated knowledge, and degree eligibility have ongoing value. The past tuition payment in itself has no recoverable benefit. An immediate withdrawal can be just as poorly reasoned as automatic continuation if it ignores the remaining value of the program.
Employment and entrepreneurship
A founder may continue funding a product to vindicate years of development. A valid review updates expected revenue, remaining production costs, competitive position, and other opportunities, and records uncertainties rather than hiding them in a narrative about sacrifices. Staff welfare, contracts, and investor commitments can create real future costs. Decision quality depends on acknowledging those constraints without treating past unrecoverable spending as a reason to spend more.
Subscriptions, entertainment, and travel
Someone may keep watching a disliked series because several seasons are complete, or attend an unpleasant prepaid event because the ticket was expensive. The decision turns on expected enjoyment, remaining time, alternative activities, fatigue, and practical obligations. Recognizing a sunk cost may change a choice, but the desired result is better matching of behavior to current values, not a universal recommendation to abandon any paid-for activity.
Relationships and shared commitments
Past investments in a relationship can influence perceptions of what is owed, what might improve, and what leaving would mean. Current responsibilities, mutual care, safety, financial interdependence, and possible future changes are consequential. An analogy to project economics must be used with care because interpersonal relationships have moral and social obligations that cannot be reduced to financial returns. The sunk-cost idea clarifies one possible cognitive influence; it does not decide for someone whether to remain or leave.
Public policy and large projects
Governments may continue infrastructure or procurement programs long after circumstances change. Evaluation requires up-to-date demand forecasts, remaining costs, externalities, contractual exit terms, institutional responsibilities, and competing uses of public funds. Announcing the total previously spent can be legitimate public accountability; allowing that figure alone to determine an additional appropriation is a different matter. Independent review may help keep those functions separate.
Consequences for Human–Technology Decisions
Digital services can make accumulated effort particularly visible. Progress bars, completion counts, paid feature histories, streaks, platform-specific data, and learning investments may influence users' willingness to switch or stop. Some of these investments create authentic switching costs or future utility; others serve chiefly as reminders of past effort. Product research should measure both rather than asserting that all retention is irrational.
In human–AI decision support, an assistant may summarize a project as “already 80% complete” without explaining the costs and success probabilities that remain. A scientifically responsible decision tool should expose the comparison between continuing, modifying, and stopping using present information. It should also label uncertainty and avoid treating the original recommendation as a commitment that must be defended.
Recommendations from algorithms can generate another source of path dependence: a team may maintain a system because it invested in procurement or integration, while a different system might now perform better. Evaluation should incorporate migration costs, verification, security, accessibility, governance, and expected benefits. The broader psychology of deference to AI has its own scientific literature; it is related to sunk costs through the decision context rather than through an identity of mechanisms.
How to Make Better Decisions When Past Costs Are Salient
Explicitly reconstruct the choice from today
Describe the alternatives as they exist now: continue, modify, postpone, terminate, or switch. For each, list the expected future benefits, additional costs, risks, nonfinancial outcomes, and opportunity costs. Place prior nonrecoverable spending in a separate historical accounting line so it remains visible for auditing without being counted again as a future expense.
The useful question is: “If this project and its current capabilities were offered to us today with no psychological ownership of the earlier decision, would its remaining benefits justify its remaining costs?” This counterfactual can help reveal attachment, but it should not conceal real assets and obligations inherited from the past.
Apply the marginal decision test
Compare each additional investment with the expected incremental value that the investment can still produce. A $20,000 development milestone may be sensible after an expensive failure if it genuinely unlocks $80,000 in expected net value; the same milestone may be irrational if it is expected to yield only $5,000. The historical project's profitability and the marginal decision's profitability are separate questions.
Use independent reviews and premade stopping criteria
For repeated decisions, an outside reviewer can evaluate updated evidence without the same personal stake in the original proposal. Predetermined criteria—technical performance, minimum revenue thresholds, safety triggers, or bounded testing periods—can reduce the opportunity for narratives to shift after disappointing results. These are defensible decision-design practices, but their effectiveness should be assessed empirically in the actual domain.
Examine thoughts and feelings without assuming that awareness is enough
Strough and colleagues randomly assigned members of a U.S. lifespan panel to different reflective instructions before a hypothetical project-completion choice. A condition directing attention toward thoughts and feelings reduced sunk-cost bias relative to an improvement-focused condition and a control. This is evidence for a specific intervention under a specific scenario, not proof that introspection reliably eliminates the fallacy in high-stakes real-world decisions. Strough and colleagues (2016)
The intervention also shows why generic advice can miss the mechanism. Asking only “how can we salvage this?” may encourage an elaborate rescue scenario. A better review asks what the decision-maker is anticipating, which future claims are supported, and whether the option would still be selected absent the historical payment.
Distinguish training effects from durable transfer
Recognizing the sunk-cost fallacy in a vignette is useful, but correctly naming a bias does not establish that a person will resist it in a later organizational or emotionally demanding situation. Education, feedback, process aids, incentive alignment, and independent review are candidates for targeted testing. Long-term behavioral change and transfer across domains demand stronger evidence than immediate performance on a similar questionnaire.
A Worked Decision Example
Imagine that a nonprofit has already spent $120,000 on a digital scheduling platform. Additional development will cost $45,000. Its updated analysis estimates that the completed platform would deliver $30,000 in future benefits relative to the present alternatives. Terminating the project costs $5,000 in unavoidable contract exit charges; switching to an available service would deliver the same essential function at a net cost of $12,000 over the relevant period.
One decision frame says that abandonment would waste $120,000. Another asks which choice is best from today. In that frame, continuation requires an additional $45,000 to receive $30,000 of benefit, yielding a net negative $15,000 relative to doing nothing, before accounting for any other effects. Termination includes the $5,000 exit charge, while switching must include $12,000 in real prospective expenses and the value of the service delivered. Proper comparison uses a common outcome baseline and a consistent time horizon for all alternatives.
The numerical answer depends on the actual benefits and costs of the switching and stopping alternatives; the historical $120,000 is the same in every branch and so does not differentiate them. A separate audit can assess whether the original development decision was reasonable. The governance task is to preserve accountability for that earlier choice while evaluating the next choice independently.
Reproducibility, Interpretation, and Open Questions
Experimental evidence supports sensitivity to sunk expenditure in many settings, and meta-analysis finds a positive average effect. It also shows considerable between-study variability. More recent preregistered tests establish effects under specific consequential manipulations and identify why common price-discount designs may be insufficient. A single general statement that “people always throw good money after bad” is therefore less accurate than a task-specific account of choice under irreversible commitments. Roth, Robbert, and Straus (2015) Beknazar-Yuzbashev, Ichiba, and Stalinski (2026)
Important open research problems include the effects of realistic stakes, preregistered replications across national samples, the measurement of long-term intervention transfer, and distinguishing loss framing from rational information about quality. Studies also need to separate sunk monetary expense, unproductive time, acquired knowledge, and emotional investment with more consistent measurements.
Research on developmental differences illustrates why broad population statements require caution: different tasks can generate different age profiles, and findings using hypothetical and consequential choices may diverge. Evidence about the average response of a research sample is not a clinical instrument and is inadequate for assessing the competence or motives of a particular individual. Hamzagic and colleagues (2025) Pighin, Timberlake, and Tentori (2025)
Frequently Asked Questions
What is the simplest sunk cost fallacy example?
A person attends an event they no longer want to attend solely because the nonrefundable ticket was expensive. The price has already been paid regardless of attendance. The live decision depends on anticipated enjoyment and other future consequences.
Is continuing a project after spending a lot always irrational?
No. Continuing can be justified when the remaining expected benefits exceed the additional costs, when stopping creates significant future liabilities, or when prior work generated valuable assets or information. A fallacy occurs when irrecoverable spending itself improperly changes the comparison.
Is the sunk cost fallacy the same as escalation of commitment?
They overlap, but escalation of commitment encompasses a wider class of continued investment after negative feedback. Sunk expenditures are one possible causal contributor alongside accountability, incentives, and optimistic beliefs.
Can sunk costs include time and effort?
Yes, irrecoverable time and effort are conceptually sunk. Their psychological effects can differ from monetary spending, and the knowledge or ability gained from past effort can remain valuable in the future.
Why does quitting feel like wasting previous effort?
People may represent an unfinished investment as an open mental account, experience cancellation as an acknowledged loss, or feel pressure to justify their previous choice. Those mechanisms have experimental support in some contexts, and their relative contribution varies by task.
Does everyone show the same level of sunk cost bias?
No. Studies report substantial heterogeneity across decisions, measurements, incentives, age groups, and experimental designs. An isolated choice offers too little information to assign a stable personal tendency.
Does learning economics eliminate the sunk cost fallacy?
Research does not support a consistent universal protection. Recognizing the normative principle can help structure a decision, but situational motives and task representations may still influence behavior.
Can an AI decision assistant prevent sunk cost errors?
An assistant can separate historical costs from future consequences, compare options, and make assumptions explicit. The effectiveness of a particular AI intervention must be tested; reliance on an automated recommendation can introduce additional judgment problems if information or calibration is poor.
Related Articles
Why People Give Up on Goals: Persistence, Disengagement, and Recovery examines commitment and disengagement in personal goal pursuit, including when persistence continues to serve current values.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making examines how algorithmic recommendations influence judgment and the independent problem of calibrated trust.
Forensic Evidence and Cognitive Bias: How Context Can Contaminate Judgment explains the mechanisms by which prior information can alter professional evaluations in forensic settings.
Neuroticism and Decision-Making: Risk, Uncertainty, and Negative Emotion discusses emotional and dispositional influences on risk judgment while retaining a distinct trait-psychology focus.
Analysis Paralysis: Why Some Decisions Become Difficult — Explores choice delay, uncertainty, and difficulty committing to an option.
Maximizing and Satisficing: Decision Styles and Well-Being — Examines search and choice strategies and their relation to well-being.
Escalation of Commitment: Why Failing Projects Continue — Studies continued investment despite unfavorable project feedback.
References
Arkes, H. R., and Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124–140. https://doi.org/10.1016/0749-5978(85)90049-490049-4)
Beknazar-Yuzbashev, G., Ichiba, S., and Stalinski, M. (2026). To the depths of the sunk cost: Experiments revisiting the elusive effect. Journal of Economic Behavior and Organization, 241, 107352. https://doi.org/10.1016/j.jebo.2025.107352
Gourville, J. T., and Soman, D. (1998). Payment depreciation: The behavioral effects of temporally separating payments from consumption. Journal of Consumer Research, 25(2), 160–174. https://doi.org/10.1086/209533
Hamzagic, Z. I., Mah, E. Y., Derksen, D. G., and Bernstein, D. M. (2025). Sunk-cost judgments across the child to adult lifespan. Psychonomic Bulletin and Review, 32, 1827–1838. https://doi.org/10.3758/s13423-025-02656-y
Kahneman, D., and Tversky, A. (1979). Prospect theory: An analysis of decision under risk. Econometrica, 47(2), 263–291. https://doi.org/10.2307/1914185
Pighin, S., Timberlake, B., and Tentori, K. (2025). Sunk-cost fallacy across real and hypothetical decision making in younger and older adults. Current Psychology, 44, 14619–14626. https://doi.org/10.1007/s12144-025-08059-5
Roth, S., Robbert, T., and Straus, L. (2015). On the sunk-cost effect in economic decision-making: A meta-analytic review. Business Research, 8, 99–138. https://doi.org/10.1007/s40685-014-0014-8
Sleesman, D. J., Conlon, D. E., McNamara, G., and Miles, J. E. (2012). Cleaning up the big muddy: A meta-analytic review of the determinants of escalation of commitment. Academy of Management Journal, 55(3), 541–562. https://doi.org/10.5465/amj.2010.0696
Soman, D. (2001). The mental accounting of sunk time costs: Why time is not like money. Journal of Behavioral Decision Making, 14(3), 169–185. https://doi.org/10.1002/bdm.370
Staw, B. M. (1976). Knee-deep in the big muddy: A study of escalating commitment to a chosen course of action. Organizational Behavior and Human Performance, 16(1), 27–44. https://doi.org/10.1016/0030-5073(76)90005-290005-2)
Strough, J., Bruine de Bruin, W., Parker, A. M., Karns, T., Lemaster, P., Pichayayothin, N., Delaney, R., and Stoiko, R. (2016). What were they thinking? Reducing sunk-cost bias in a life-span sample. Psychology and Aging, 31(7), 724–736. https://doi.org/10.1037/pag0000130
Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior and Organization, 1(1), 39–60. https://doi.org/10.1016/0167-2681(80)90051-790051-7)