Escalation of commitment occurs when a decision maker keeps allocating time, money, attention, or institutional support to an existing course of action despite adverse feedback and despite the availability of a better prospective alternative. The phenomenon is particularly important in long-running projects: each new setback may make withdrawal more psychologically difficult because earlier decisions have become connected to personal identity, credibility, contractual relationships, and organizational routines. Scientific analysis asks whether additional investment is justified by information available now, not whether the project has already incurred losses. Staw (1976)90005-2); Brockner (1992).
Research distinguishes the repeated act of recommitment from a one-time sunk-cost choice. It investigates how negative feedback is detected, interpreted, communicated, and translated into successive resource-allocation decisions. Escalation may arise from self-justification, preference for an originally chosen option, expected recovery, loss framing, competition, or organizational incentives. Some continued investments are economically defensible; the judgment of bias requires a specified decision benchmark and credible estimates of future outcomes. The evidence therefore supports a conditional, mechanism-based account rather than a universal claim that persistence is irrational. Sleesman et al. (2012); Schulz-Hardt et al. (2009).
What Escalation of Commitment Means
The scientific definition
Escalation of commitment is a dynamic pattern in which a decision maker maintains or intensifies commitment to a previously selected course of action after receiving information that undermines its expected success. The decisive feature is the sequence: an initial choice establishes a commitment, subsequent evidence challenges it, and an opportunity to revise or terminate the course of action is met with renewed investment. Resources can include funding, employee time, managerial attention, reputational capital, and further exposure to risk. In experimental studies, the dependent variable may be an additional investment amount; in organizations, it may be delayed termination, repeated budget extensions, or the continued use of a failing system. Brockner (1992).
An observable continuation decision becomes evidence of biased escalation only when its prospective costs and benefits, appropriate alternatives, and available information have been evaluated. Unexpected setbacks are compatible with a project that remains valuable. A rescue investment can be rational when additional spending purchases sufficiently probable benefits, preserves a valuable option, or prevents a larger avoidable future loss. Research designs therefore need to separate a decision maker's unwillingness to disengage from favorable private information, delayed payoffs, and genuine strategic commitments. Bowen (1987); Schulz-Hardt et al. (2009).
A concrete project example
A development team has spent eighteen months building a platform. Pilot users report poor task performance, the cost of maintaining the architecture rises, and an independent alternative can deliver the same outcomes more reliably. Managers authorize another quarter of work, explaining that abandoning the platform now would invalidate the effort already expended. A second review brings additional unfavorable evidence, but the team increases the budget again to reach the next milestone. The pattern contains the defining stages of escalation: prior choice, disconfirming feedback, identifiable alternatives, and repeated recommitment. The outcome is not established by a single late project or cost overrun; an analysis must compare realistic forward-looking options.
Suppose finishing the existing platform requires an additional $4 million and is expected to generate $3 million in future net benefits before those additional costs. A replacement requires $1 million and is expected to generate $3 million in benefits. On these simplified assumptions, continuation has prospective net value of −$1 million, while replacement has prospective net value of $2 million. The $8 million already spent on the existing platform is irrecoverable in either case and therefore cancels out of the comparison. If the assumptions change—for example, replacement involves substantial unmodeled switching costs—the recommendation may change. The example illustrates a normative criterion, not an empirical estimate of typical projects.
Escalation, sunk costs, and ordinary persistence
A sunk cost is an expenditure that cannot be recovered and is unaffected by the current decision. The sunk-cost effect refers to greater willingness to proceed because such expenditure occurred. Escalation of commitment encompasses a longer process involving feedback, ownership of previous choices, social pressures, and additional rounds of investment. The terms overlap, but they occupy different levels of analysis. Research on sunk costs provides an important foundation; research on escalation adds decision history and evolving organizational context. Arkes and Blumer (1985)90049-4); Staw (1976)90005-2).
Persistence means continuing effort toward a goal. Persistence can express patience, perseverance, learning, or a rational response to temporary poor outcomes. Escalation refers to a suspected or demonstrated mismatch between investment and current prospects. The relevant question is whether the project still deserves resources compared with feasible alternatives, using an appropriate forecast and explicit decision criteria. The distinction also prevents treating dedication, organizational loyalty, or grit as automatic symptoms of error.
The Historical Development of Escalation Research
Staw's 1976 investment experiment
Barry Staw's foundational experiment placed 240 business-school students in a simulated investment decision and varied personal responsibility for the initial choice along with the consequences attributed to that choice. The strongest further investments occurred among participants who were personally responsible for an unfavorable outcome. This counterintuitive result suggested that negative feedback could sometimes intensify commitment by creating a need to defend the initial decision. The experiment launched a sustained research tradition on the interaction of responsibility, feedback, and reinvestment. Staw (1976)90005-2).
The design had a clear experimental advantage: responsibility and feedback could be manipulated while the investment environment was held relatively consistent. Its limitations also matter. Participants made decisions in a simulated business setting; outcomes were tied to instructions and hypothetical assumptions; and the task did not reproduce all the incentives, budgets, and political constraints of actual organizations. The original findings identify one possible causal pathway under those conditions, not the frequency of escalation in every workplace or industry.
Sunk-cost experiments and field observations
Arkes and Blumer documented how past expenditure can affect future participation. Their research included a theater-subscription field study in which customers paying more for subscriptions attended more performances during the following months, alongside questionnaire experiments. The result is relevant because waste aversion can become part of the rationale for remaining involved in an endeavor. Yet a theater subscription is a different decision ecology from a multinational project's repeated financing, and its findings cannot be substituted for a direct estimate of organizational escalation. Arkes and Blumer (1985)90049-4).
The research program subsequently expanded from isolated funding decisions to technological projects, organizational exit, competitive auctions, group decisions, and the recognition of failing systems. These settings introduced alternative explanations: people may genuinely prefer what they originally selected; teams may share incentives that reward apparent consistency; and managers may interpret negative feedback differently according to their access to information. Keil et al. (2007); Whyte (1993).
From single explanations to an integrated framework
Brockner's influential theoretical review described the self-justification explanation and organized alternatives that could complement it or challenge its centrality. The later meta-analysis by Sleesman, Conlon, McNamara, and Miles evaluated determinants across an accumulated research literature, giving the field a broader quantitative foundation for assessing competing accounts. Scientific progress here has consisted in specifying interacting mechanisms and boundary conditions rather than naming one motive that explains all failed project continuations. Brockner (1992); Sleesman et al. (2012).
Why People Continue Failing Projects
Self-justification and responsibility
The self-justification account begins with an uncomfortable implication of failure: the decision maker's earlier judgment may have been mistaken. Additional investment offers a psychological path toward vindication, particularly when success could retroactively transform a criticized choice into a defensible one. Personal responsibility makes the outcome self-relevant. The mechanism is close to cognitive dissonance and motivated reasoning, but it should be operationalized through changes in choices, perceived responsibility, and evaluations rather than inferred solely from an observer's impression. Staw (1976)90005-2); Brockner (1992).
Responsibility can also change the underlying preference structure. When people originally chose a project, they may have preferred its goals, technology, or strategic direction before any subsequent feedback. A person inheriting the project was never selected on that preference basis. Schulz-Hardt, Thurow-Kröning, and Frey experimentally demonstrated that original choice can create a preference-based difference and that responsibility did not add an independent effect beyond the measured preferences in their studies. This interpretation prevents automatically labeling every difference between project initiators and successors a desire to save face. Schulz-Hardt et al. (2009).
Reputation, accountability, and public commitments
A decision made in public has audiences. Managers may expect colleagues, investors, or elected officials to interpret project cancellation as evidence of incompetence. If organizational promotion, political standing, or professional identity depends on appearing decisive and consistent, continuation may become individually attractive even when it reduces organizational welfare. Accountability therefore needs precise specification: being answerable for decision quality, being required to justify a previous choice, and facing punishment for visible failure create different incentives. Brockner (1992); Simonson and Staw (1992).
Public commitments can also coordinate many stakeholders. A formal promise to deliver a bridge, a digital platform, or a scientific facility alters expectations and may produce genuine legal or coordination costs if withdrawn. A normative evaluation should include avoidable obligations and future penalties. These costs are prospective consequences of terminating the project, while the mere fact of past expenditure remains sunk. Careful accounting preserves the distinction between reputational discomfort and a real future contractual liability.
Negative framing and loss aversion
When decision makers code a struggling project as a loss relative to an established reference point, an additional investment may look like a chance to return to the break-even position. Prospect-theory-based accounts predict that framing and the perceived shape of risk can influence willingness to gamble for recovery. Whyte's experiments with individual and group decisions found patterns consistent with a prospect-theory explanation while also finding support for self-justification. This evidence supports multiple routes to escalation; it does not show that one universal loss-aversion coefficient determines every continuation choice. Whyte (1993).
A project that is described as “90 percent completed” can invite recovery-oriented thinking, even when the remaining ten percent requires expensive and uncertain work. Conversely, terminating the project may be framed as recognizing a loss already incurred. Neither expression directly specifies future value. Explicitly re-anchoring the analysis at the present decision point encourages attention to remaining costs, credible payoffs, and the probability of successful completion.
Selective perception and the failure to recognize failure
An organization's feedback is rarely a single unequivocal number. Project reports may contain encouraging usage figures next to poor retention, technical progress alongside low customer value, or short-term losses alongside plausible learning. Decision makers must first recognize the significance of these patterns. Keil, Depledge, and Rai found that problem recognition was inversely associated with escalation in their study, while selective perception and illusion of control influenced recognition and investment choices. Their findings connect escalation with the interpretation of evidence as well as with the motivation to continue. Keil et al. (2007).
Information architecture therefore matters. Dashboards that emphasize completed tasks while omitting opportunity costs can make a project appear healthier than a comparison of alternative outcomes would suggest. At the same time, a dashboard that highlights one disappointing quarter while ignoring strong leading indicators can promote premature exit. Better feedback does not mean maximizing bad news; it means giving decision makers evidence about the variables that actually govern expected success.
Illusion of control and optimism about recovery
Managers are often asked to make uncertain outcomes better through action. This legitimate agency can blend with overestimating one's control over market demand, technological feasibility, or external shocks. Illusion of control may make a further intervention appear more effective than available evidence warrants. Unwarranted optimism can likewise shift recovery probabilities upward, particularly when earlier setbacks are attributed to temporary external causes while successes are credited to internal skill. The strength of these mechanisms depends on the realism of the task, the quality of feedback, and the decision maker's actual influence. Keil et al. (2007).
Optimism is not by itself evidence of escalating commitment. A forecast can be confident and correct. The empirical issue is calibration: do events occur at the rates stated by the forecast, and do forecasts incorporate comparable cases? In a decision review, probability claims should be written down before the next outcome is observed, then compared with subsequent evidence.
Competition and the fear of conceding
Escalation also occurs in interactive conflict. In dollar-auction paradigms, participants bid for a fixed prize while earlier bids become sunk. Continuing to bid can be driven by fear of losing one's existing stake, competition, social comparison, and the emotional intensity of an unfolding contest. Hafenbrädl and Woike experimentally examined competitive escalation and different interventions, including learning through direct experience. Their results illustrate that interpersonal rivalry produces a different causal environment from the solitary appraisal of a company's project. Hafenbrädl and Woike (2018).
The distinction matters for intervention. A manager facing a competitive auction may need limits on bidding rules and escalation triggers, while a software project requires independent technical review and explicit opportunity-cost comparisons. A recommendation that works for a hypothetical investment choice may have limited effect when rivals continually provoke reactive decisions.
Organizations can reward continuation
Organizational structures may separate the people who benefit from continuing from those who bear its eventual costs. Budget owners can preserve headcount, teams can avoid disruptive redeployment, and executives can defer a visible write-off beyond their tenure. Even if every actor recognizes the economic problem, the incentive structure can maintain collective commitment. Escalation is therefore a multilevel phenomenon: individual cognition, social processes, governance, accounting rules, and political incentives may operate together. Brockner (1992); Sleesman et al. (2012).
A productive analysis separates individual responsibility from the system in which that responsibility is exercised. Blaming a particular manager for a poor outcome can itself strengthen the need to justify the choice. Changing reporting incentives and decision ownership may alter behavior more effectively than exhorting people to be less biased.
Normative Rationality: When Should a Project Continue?
The forward-looking comparison
A standard economic benchmark is to choose the feasible option with the highest expected incremental net value from the present decision point. The decision maker compares each alternative's future benefits, future costs, risk, timing, and relevant constraints. Past costs that will remain unchanged regardless of today's decision are excluded. Formally, for option i, an illustrative risk-neutral criterion is EV(i) = Σ_s P(s | information, i) × [future benefits(i,s) − future costs(i,s)]. Continue when the expected value of continuation exceeds that of the best feasible alternative, including orderly shutdown. This equation states a normative model under defined probabilities and risk-neutral valuation; actual organizations may require risk adjustment, nonfinancial objectives, and portfolio constraints.
The project assessment can also use discounted cash flow, expected utility, real-options reasoning, or constrained optimization. These frameworks can disagree when probabilities are poorly known, outcomes are nonlinear, or objectives are multi-dimensional. The rationality benchmark must be specified before describing observed persistence as a mistake. A loss-making quarter is a descriptive outcome, whereas a negative forward-looking value relative to viable alternatives is evidence bearing on the choice standard.
Irrecoverable costs, recoverable assets, and switching costs
An organization may have spent $10 million on research, but that historical expenditure cannot be recovered through either continuation or abandonment. It is sunk. If selling equipment would bring $2 million, however, the potential sale proceeds are a future benefit of exit. If replacing a software platform requires retraining, the new training expenditure is a real future switching cost. A valid comparison includes these recoverable and avoidable consequences even though an unsound sunk-cost argument may accompany them.
This classification often resolves apparent contradictions in conversations about escalation. The same sentence, “We have already invested too much to stop,” may conceal either a faulty appeal to irreversible spending or a legitimate concern about future contractual penalties and integration costs. A good analyst requests disaggregated estimates rather than deciding from the rhetoric alone.
Uncertainty, options, and the value of learning
A project that has suffered a setback may still warrant a small experiment because new evidence can improve the decision about a larger future investment. This is a value-of-information argument. An appropriately limited pilot, provided its cost is below the expected benefit of learning, is distinct from extending a failing course merely to defend previous expenditures. Waiting or staging investment can also preserve a real option under uncertainty, though delay itself may have costs.
A staged decision should specify what evidence the next stage is intended to produce. Without such a specification, a series of “one last tests” can become an indefinite commitment that shifts its success criteria after each disappointing result. The analytical object is an explicitly time-bounded learning investment with precommitted interpretation rules, not an open-ended license to continue.
Why eventual success does not prove earlier decisions were optimal
A project that eventually succeeds may have been a poor investment when earlier decisions were made, just as a rationally selected project can fail by chance. Outcome evaluation and decision-quality evaluation operate at different levels. Ex ante judgment requires assessing what information and alternative options were available at the decision date. Outcome bias can make successful persistence appear brilliant and unsuccessful persistence appear foolish regardless of the decision's original probabilistic merits.
This distinction also prevents a selective sample of dramatic failures from establishing a psychological theory. Projects attract attention when budgets explode or leaders refuse cancellation, but an empirical claim about the general rate of escalation needs a defined sampling frame, appropriate comparisons, and records of the projects that ended quietly or recovered after justified investment.
How Escalation Is Studied and Measured
Sequential resource-allocation tasks
A common laboratory design asks participants to select or inherit a project, presents unfavorable performance information, and offers a second opportunity to allocate resources. Researchers manipulate personal responsibility, prior spending, feedback severity, and sometimes the existence of alternative projects. Additional investment is compared across randomized conditions. Such tasks can identify causal effects of experimental prompts under controlled assumptions, especially when participants understand the payoff structure. Staw (1976)90005-2).
A single second allocation is only a partial simulation of the real escalation process. Projects unfold through repeated reviews, changing stakeholders, and feedback whose diagnostic value varies over time. Multi-round experiments can study whether an intervention's effect persists after additional negative feedback and whether participants respond differently to temporary reversals versus structural project failure. Schultze and Schulz-Hardt (2022).
Experimental dependent variables and comparisons
Investigators measure dollars allocated, the binary decision to continue, the time before withdrawal, the number of rounds continued, or willingness to fund an alternative. These outcomes should be interpreted in their specific units and incentive context. A larger amount invested is evidence of escalation only if the manipulation and benchmark justify that interpretation. When one project has greater expected value than another, choosing it more often can be rational despite its negative past performance.
Hypothetical financial scenarios offer experimental control and lower cost, while incentivized studies can better align participants' payoffs with their stated choices. Neither format perfectly recreates legal obligations, employment incentives, or the distribution of personal and organizational costs. Construct validity improves when a study verifies that participants understand the feedback and can perceive the available exit options.
Observational organizational evidence
Field researchers may examine delayed write-offs, project termination rates, leadership changes, budget revisions, and records of performance reports. Such observations reveal ecological complexity that laboratory scenarios omit, but they introduce confounding: the firm that continues a project may possess favorable private information, face switching costs, or have strategic obligations that researchers cannot observe. Causal interpretation requires credible comparison groups, longitudinal records, and attention to the timing of managerial transitions.
A rigorous field test should reconstruct which options and forecasts were available at each decision point. Researchers also need to guard against survivorship bias: abandoned projects may disappear from records, whereas surviving projects produce a longer history of revisions. Prolonged duration can be an indicator for investigation without being sufficient proof of irrational escalation.
Measuring mechanisms separately
Self-justification, loss framing, preference strength, perceived control, problem recognition, and organizational incentives are distinct variables. Studies can measure them with pre-registered questionnaires, experimentally manipulated feedback, revealed choices, and mediation designs. A statistical mediator alone rarely settles mechanism because measurement error and omitted confounding remain possible. Experimental designs that directly vary the hypothesized mechanism provide more discriminating evidence.
The distinction between a construct and a task effect is essential. Escalation of commitment is a descriptive decision pattern in an appropriate design; it is not a psychiatric diagnosis, a symptom inventory, a stable personality label, or a general measure of intelligence. High scores on an investment task should never be interpreted as proof that an individual has a clinical condition.
What the Evidence Supports
The 2012 meta-analytic synthesis
Sleesman and colleagues' meta-analytic review synthesized determinants of escalation across the organizational-science literature and compared theoretical perspectives. Its contribution was to integrate diverse constructs—rather than assume that sunk costs or self-justification account for the entire phenomenon. Meta-analysis can improve precision and assess patterns across studies, but pooled estimates remain dependent on the included tasks, constructs, populations, study quality, and whether publication practices selected unusually strong results. The review supports a multifactorial framework without yielding a universal escalation probability for arbitrary real-world projects. Sleesman et al. (2012).
Because studies operationalize persistence in different ways, pooling their effect sizes does not automatically translate into a statement that a given percentage of projects will fail due to bias. An effect observed in a student business simulation is expressed on a different scale from delayed cancellation in public infrastructure. Readers should ask which moderator, comparison condition, and outcome unit underlies any quoted effect size.
A failed replication of the classic effect
Białek, Węgrzyn, and Meyers attempted to replicate and extend Staw's responsibility effect while examining numeracy and cognitive reflection. In their sample of 365 participants, they found no evidence for the predicted escalation pattern; participants tended to put more resources into projects with better prospects, and numeracy and cognitive reflection did not predict escalation. This result directly limits claims that the classic phenomenon appears reliably whenever a person owns a prior bad decision. Białek et al. (2021).
A failed replication does not logically establish that escalation never occurs. It implies that the historical effect cannot be considered invariant across operationalizations, incentives, samples, or instructions. The scientific response is to identify moderators and report unsuccessful tests alongside positive findings, rather than selectively cite experiments that resemble a preferred narrative.
A replication and extension of responsibility change
Schultze and Schulz-Hardt reassessed changing decision responsibility as an intervention and highlighted two conditions for a genuinely useful remedy: temporal stability across subsequent decisions and specificity to projects that continue to fail, as opposed to those that temporarily decline and subsequently recover. Their study challenges simple prescriptions that changing the manager will automatically produce better decisions. The mechanism must be assessed over time and relative to project quality. Schultze and Schulz-Hardt (2022).
An intervention can lower reinvestment in an experimental task and still reduce welfare if it causes people to abandon promising projects after a single setback. Evaluations therefore need both sides of the error distribution: unjustified continuation and unjustified termination. This makes well-designed de-escalation a problem of calibrated responsiveness to evidence.
Cultural, individual, and situational variation
Studies of escalation span organizational roles and decision settings, but estimates from one population should not be generalized to all cultures or institutions. Norms of saving face, hierarchy, accountability, risk distribution, and group dissent may change how negative feedback is received. Differences in expertise, familiarity with the decision domain, and incentives can also modify choices. Evidence for a cultural pattern requires matched decision tasks and measurement equivalence, not stereotypes about national character.
Individual cognitive reflection or numeracy is not a guaranteed protective factor; the 2021 failed replication found no association in that experimental setting. Domain expertise can help interpret feedback, yet experts may also possess stronger attachments to an approach they designed. A plausible moderator must be tested in a defined design before being recommended as a reliable predictor. Białek et al. (2021).
Collective and Institutional Escalation
Teams and group decisions
Groups pool information but can also distribute responsibility ambiguously or magnify a shared loss frame. Whyte's investment scenarios found escalation among both individuals and groups, with group decisions amplifying certain tendencies in that design. These findings warrant concern about team processes, yet they do not justify the generalization that groups always escalate more than individuals. Decision rules, dissent, incentives, and the distribution of knowledge shape the direction of group effects. Whyte (1993).
A review committee may reinforce a project sponsor's preferred narrative if all members depend on the project's continuation. Another committee may be overly conservative if none bears the cost of abandoning valuable opportunities. Independence, relevant expertise, and a transparent criterion matter more than the mere number of reviewers.
Technology and software projects
Complex software efforts are susceptible to uncertain requirements, invisible technical debt, delayed user feedback, and shifting definitions of completion. Completion milestones can create a misleading sense of progress when they count delivered features rather than tested user value. Research on information-systems escalation addresses how problem recognition and biases influence continuation decisions. Keil et al. (2007).
Organizations can improve diagnostic feedback by separating technical progress, adoption, reliability, total lifecycle cost, and counterfactual alternatives. A deployed feature can be technically complete and commercially unsuccessful. A higher budget may be defensible if it solves a clearly identified bottleneck; it may also disguise reluctance to accept a mistaken architectural commitment. These competing interpretations require evidence.
Public programs and infrastructure
Large public projects combine uncertain engineering estimates with election cycles, multi-party contracts, distributional politics, and budget approval routines. Costs may increase over years while responsibility rotates among institutions. Escalation research suggests asking whether political and reputational incentives favor repeated continuation, but a delayed or expensive public project should not be classified as psychologically biased without a full comparison of its feasible alternatives and public objectives.
Public decision audits should document revised benefits, financing costs, maintenance obligations, environmental effects, and opportunity costs. Accountability improves when the rationale for continuation can be examined independently and when public objectives are stated clearly enough to permit real evaluation rather than moving targets.
Healthcare decisions and ethical constraints
Medical settings can involve sequential decisions under uncertainty, prior investments of professional effort, and reluctance to change a treatment pathway. Descriptive accounts of escalation in surgical intensive care have highlighted self-justification and accountability as possible mechanisms, although case-based evidence cannot establish prevalence or a causal effect. Clinical decision making also includes patient preferences, prognostic uncertainty, ethical duties, and the risk of withdrawing beneficial care. Braxton et al. (2017).
The appropriate response is an evidence-based multidisciplinary reassessment consistent with clinical guidelines and individual patient goals. A psychological concept does not authorize categorical clinical recommendations. Treatment changes must be based on medical evidence, shared decision making, and professional assessment.
Artificial intelligence and decision support
Decision-support systems can make alternative forecasts more visible, improve documentation, and expose inconsistencies across review rounds. They can also facilitate escalation when generated reports create a false impression of certainty, repeatedly endorse the sponsor's assumptions, or present polished explanations for an existing choice. Human oversight should therefore examine inputs, uncertainty, independent benchmarks, and the incentives governing when recommendations are accepted or overridden.
The psychological topic here is how human decision makers assign authority to technology. It is separate from claims about a system's subjective experience. AI tools should be assessed by whether their forecasts and recommendations improve observed decision quality under specified conditions, rather than by the fluency of their explanations. See the separately published analysis of AI as Authority.
Interventions: How to Reduce Unjustified Recommitment
Independent review and rotation of responsibility
Separating the original project sponsor from the next funding decision can reduce the immediate pressure to vindicate an earlier judgment. Independent reviewers may evaluate alternatives from the present forward, and responsibility rotation can bring fresh attention to negative feedback. Yet recent replication research shows why this remedy cannot be assumed to work permanently or selectively for truly failing projects. Monitor both error types and examine whether the benefit survives later review rounds. Schultze and Schulz-Hardt (2022).
A project review should preserve accountability for learning rather than invite scapegoating. A culture that punishes candid termination may make managers conceal evidence or seek continual extensions. The institutional objective is accurate resource allocation, not the public humiliation of someone who made an uncertain initial choice.
Predefined exit criteria and decision checkpoints
Before a major investment, teams can define observable performance milestones, acceptable spending boundaries, and the conditions for another assessment. They should distinguish a temporary setback from structural failure and identify what evidence would reverse the continuation recommendation. Written rules reduce the opportunity to reinterpret a missed threshold after the fact, although rigid thresholds can themselves cause mistakes when new information changes the environment.
A robust checkpoint records its baseline forecast, the forecast's confidence or credible interval, and the alternatives available. It should permit an explicit exception when justified by new evidence and document the reason for the exception. This procedure builds a traceable sequence of decisions instead of relying on post hoc narratives about whether stopping would mean defeat.
Compare options without historical-cost cues
An intervention grounded in the sunk-cost distinction asks reviewers to evaluate the remaining project and its replacement as if deciding now which prospective option to finance, while preserving real future transition costs. A practical worksheet separates historical irrecoverable expenditure from prospective cash flows and nonfinancial consequences. The objective is not to erase institutional memory; prior spending can reveal information about project complexity and learning, even when it no longer has an independent place in the value calculation. Arkes and Blumer (1985)90049-4).
A useful question is: “If another team had arrived today with this evidence, which project would we select?” This counterfactual can weaken identity-based ownership, but its practical effectiveness depends on truthful forecasts and a viable alternative. Asking the question is a decision aid, not a validated universal cure.
Challenge assumptions with genuinely diagnostic evidence
Teams can commission an external technical audit, compare project performance with similar completed projects, or run a pilot designed to discriminate between recovery and structural failure. Such reviews should define the predictions of competing hypotheses in advance. If every possible result is interpreted as proof that the project deserves another investment, the review has no diagnostic value. The importance of problem recognition in escalation studies supports treating evidence interpretation as a central intervention target. Keil et al. (2007).
Independent evaluation is most useful when evaluators have access to complete information and are rewarded for accuracy. An outsider who sees only a selective executive summary can reproduce the same reasoning problem while adding an appearance of objectivity.
Motivational reframing
Molden and Hui tested a regulatory-focus intervention in two studies of hypothetical financial decisions. Activating growth-and-advancement-oriented motivation reduced recommitment relative to comparison conditions in those experiments. The result suggests that making attractive alternatives cognitively salient may help break a loss-recovery frame, but the designs do not establish enduring reductions in large organizations. Molden and Hui (2011).
Such findings are best understood as mechanisms to test locally rather than management slogans. A new frame can affect attention, but if incentives or contractual constraints still reward continuation, the organizational decision may remain unchanged.
The potential benefits and risks of self-affirmation
Sivanathan, Molden, Galinsky, and Ku studied whether affirming a sense of personal worth can reduce the need to defend a failing decision. Across their experiments, affirmation on a value or trait unrelated to the original judgment sometimes reduced escalation, while affirming decision-making ability could increase it. This nuanced pattern demonstrates that an intervention's wording and relevance to a threatened identity can reverse the intended effect. Sivanathan et al. (2008).
The practical lesson is to evaluate psychological interventions with experimental outcome measures and appropriate control conditions. It would be premature to conclude that generic confidence-building workshops or positive statements reliably improve termination decisions.
Learning to recognize competitive traps
Hafenbrädl and Woike studied interventions involving prior experience with structurally identical or similar competitive situations. Their work shows that the structure of an interaction can itself foster escalation and that interventions can be tested against measurable losses. Findings from dollar auctions should nevertheless be interpreted within competitive tasks; they do not prove that exposure to a game will improve organizational governance or ordinary project budgeting. Hafenbrädl and Woike (2018).
Institutions can reduce competitive traps by setting binding financial limits and clarifying in advance when a bid, partnership, or negotiation no longer creates value. Such rules can change the payoff structure, not merely the participant's interpretation of it.
Evaluate interventions with the right outcomes
A successful intervention should improve prospective decision quality over a series of decisions. Key outcomes include future realized value, avoidance of unjustified extensions, preservation of profitable recovery opportunities, forecast accuracy, and appropriate responsiveness to new evidence. Merely reducing continuation rates can look successful while causing premature abandonment. This specificity requirement is central to interpreting recent intervention research. Schultze and Schulz-Hardt (2022).
Evidence-based organizational trials can randomize review formats, use comparable project portfolios, and preregister decision criteria. When randomization is infeasible, careful longitudinal comparisons and transparent reporting of selection bias are preferable to uncontrolled testimonials. The general transfer of laboratory debiasing into complex real-world projects remains an empirical question.
A Structured Decision Audit for Failing Projects
Step 1: Establish the current decision and its alternatives
Identify the choice being made today, its decision deadline, and the feasible set of options. Options commonly include continuing unchanged, restructuring, funding a limited learning experiment, pausing, selling a recoverable asset, replacing the project, or terminating it. Specify ownership and implementation costs for each option. The audit should not treat the current project as the default against which every alternative must justify itself.
Step 2: Separate future consequences from historical investment
Document what resources have already been irretrievably spent and place them outside the forward-looking payoff calculation. Then identify remaining expenditures, sale value, transition costs, contractual penalties, employee redeployment costs, and additional benefits under each option. Check that the calculation neither smuggles past losses back into a “need to recover” term nor overlooks genuine future costs created by exit.
Step 3: Make forecasts explicit and compare them with benchmarks
For each option, estimate probability distributions or bounded scenarios for costs, benefits, completion, and failure. Explain data sources, base rates, and uncertainties. Compare with similar projects where possible, and distinguish internally negotiated targets from externally observed outcomes. Recalculate expected value or the appropriate policy objective under realistic scenarios rather than presenting a single precise but unsupported figure.
Step 4: State a continuation rule before new feedback arrives
Define which observable outcomes would justify additional investment, revision, or withdrawal, and set the next review date. A legitimate exception should be supported by documented new information. Precommitment reduces opportunities to move the threshold after every disappointing result, but periodic reassessment preserves flexibility if the environment changes materially.
Step 5: Build independent disagreement into governance
Ask reviewers to present both the strongest continuation case and the best alternative case, noting what would falsify each. Provide confidential channels for dissent and ensure that people presenting negative information are not punished for revealing it. Require written reasoning when reviewers choose to override agreed criteria. Compare ex ante decisions with realized outcomes over time to improve calibration rather than rewarding mere persistence.
Step 6: Review the process after termination or success
Conduct a decision-quality review based on the evidence known at each milestone. Document the forecast errors, hidden costs, useful early warnings, and missed opportunities. Praise timely termination when it frees resources for a stronger option, and acknowledge justified continuation that was unlucky in its realized outcome. Such reviews gradually reshape the incentives that make defensive persistence attractive.
Common Misinterpretations
Every failed project was an escalation error
Failure is an outcome; escalation is a pattern in sequential decisions. Ex ante uncertainty means projects with sound expected value can fail. A postmortem based only on the final result risks hindsight bias and outcome bias. Examine when information became available and what alternatives were feasible at the time.
Every additional investment after bad news is irrational
Temporary setbacks, learning curves, and legitimate option value may support additional investment. It is possible to rationally invest in a project that is currently losing money when remaining prospective benefits exceed costs. The scientific claim of bias requires evidence of a departure from a stated normative standard.
The phenomenon is explained entirely by sunk costs
Sunk costs can contribute, but responsibility, preferences, framing, social accountability, problem recognition, rivalry, and institutional incentives can influence escalation. Reducing it to a single numerical error obscures the interventions required for multilevel decisions. Brockner (1992); Sleesman et al. (2012).
A change in leadership is always a solution
New leaders may have less identity investment but also less domain information, stronger incentives to differentiate themselves, or an incentive to terminate a good predecessor's project. Research on responsibility change supports testing temporal stability and specificity rather than assuming a reliable one-direction effect. Schultze and Schulz-Hardt (2022).
Escalation demonstrates mental illness or deficient intelligence
Escalation of commitment is studied as a context-sensitive decision phenomenon, not as a psychiatric diagnosis. The failure of one replication to associate behavior with numeracy or cognitive reflection further cautions against treating the phenomenon as a simple index of personal cognitive capacity. Białek et al. (2021).
Frequently Asked Questions
What is an example of escalation of commitment?
A manager continues financing a product after repeated negative market tests because stopping would expose the earlier decision as costly. The example becomes a demonstrable decision error when analysis shows that a feasible alternative had higher expected prospective value using the information available at the time.
What causes escalation of commitment?
Several mechanisms can contribute: self-justification, preference for one's original choice, loss framing, selective perception of feedback, optimism about control, reputation, rivalry, and incentives to defend an organization's earlier decisions. Their influence varies by study design and decision environment. Sleesman et al. (2012).
Is escalation of commitment the same as the sunk-cost fallacy?
They overlap but refer to different levels of decision making. Sunk-cost effects concern the inappropriate influence of irrecoverable prior investment on a current choice. Escalation of commitment concerns repeated persistence or reinvestment after adverse feedback, potentially involving several additional social and organizational mechanisms. Read the separate scientific article on Sunk Cost Fallacy.
Is escalation always a cognitive bias?
Escalation is a descriptive behavior pattern whose irrationality depends on the benchmark. A continuation decision can be defensible when new evidence, switching costs, or real option value make continuation more attractive than alternatives. Bias is established by inappropriate responsiveness to factors that should not determine the forward-looking comparison.
Can a project recover after repeated setbacks?
Yes. That possibility is one reason a good de-escalation intervention must discriminate between structural failure and temporary difficulty. The challenge is to use calibrated forecasts and informative tests without treating every possibility of recovery as sufficient reason for unlimited investment. Schultze and Schulz-Hardt (2022).
How can teams decide when to stop?
Teams can compare prospective values of all feasible options, preserve alternative estimates, predefine review criteria, invite independent scrutiny, and monitor whether forecasts remain calibrated. Termination and continuation both warrant explicit evidence. There is no scientifically validated universal threshold that fits every industry and project.
Does personal responsibility reliably increase escalation?
Staw's original task found stronger recommitment in personally responsible participants after negative outcomes, but a later replication with 365 participants did not reproduce the expected pattern. Alternative explanations emphasizing selection and initial preferences also qualify the responsibility account. Staw (1976)90005-2); Białek et al. (2021); Schulz-Hardt et al. (2009).
Do groups make escalation worse?
Some experiments have found group amplification, but the direction depends on group composition, shared goals, responsibility, information, and decision rules. Group participation alone is not a guarantee of better or worse outcomes. Whyte (1993).
Can training or AI prevent escalation?
Structured reviews, exposure to alternatives, and decision-support systems can assist diagnosis and accountability. Specific experimental interventions have reduced recommitment under controlled conditions, while transfer to repeated consequential decisions requires separate testing. AI recommendations must themselves be checked for error, false precision, and overreliance. Molden and Hui (2011); Hafenbrädl and Woike (2018).
Conclusion
Escalation of commitment is a sequential decision phenomenon in which a prior choice retains influence after adverse feedback. It becomes a bias when continued investment is explained by irrecoverable losses, defensive motives, or distorted information processing rather than by a justified comparison of prospective options. Research connects the phenomenon to self-justification, initial preferences, loss framing, problem recognition, competition, and organizational governance, while recent replication work demonstrates meaningful limitations on the generality of classic findings. Brockner (1992); Białek et al. (2021).
The most useful practical response is a disciplined series of forward-looking reviews with explicit alternatives, diagnostic evidence, independent judgment, and criteria that distinguish recoverable difficulty from persistent failure. Such a system improves decisions by making commitment responsive to future value and changing information. Its effectiveness should be judged by observed decision quality across cases, including whether it avoids unnecessary termination as well as unnecessary continuation.
Related Articles
Sunk Cost Fallacy: Why Past Investments Shape Future Choices — Examines how irrecoverable spending affects subsequent choices and why sunk costs matter for escalation.
Planning Fallacy: Why We Underestimate Time and Costs — Explores inaccurate prospective estimates of project duration and costs, a related but independent phenomenon.
Decision Fatigue: What Research Actually Shows — Reviews contested mechanisms of repeated decision making and how they differ from commitment-driven persistence.
Cognitive Debiasing Strategies: Which Interventions Work? — Examines evidence for training, structured review, and decision-quality interventions.
AI as Authority: Trust, Expertise, Automation Bias, and Human Decision-Making — Explores trust, expert claims, and oversight when automated systems influence human decisions.
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)90049-4
Białek, M., Węgrzyn, M., and Meyers, E. A. (2021). Escalation of commitment is independent of numeracy and cognitive reflection: Failed replication and extension of Staw (1976). Economics and Business Review, 7(2), 5–16. https://doi.org/10.18559/ebr.2021.2.2
Bowen, M. G. (1987). The escalation phenomenon reconsidered: Decision dilemmas or decision errors? Academy of Management Review, 12(1), 52–66. https://doi.org/10.5465/amr.1987.4306470
Braxton, C. C., Robinson, C. N., and Awad, S. S. (2017). Escalation of commitment in the surgical ICU. Critical Care Medicine, 45(4), e433–e436. https://doi.org/10.1097/CCM.0000000000002261
Brockner, J. (1992). The escalation of commitment to a failing course of action: Toward theoretical progress. Academy of Management Review, 17(1), 39–61. https://doi.org/10.5465/amr.1992.4279568
Hafenbrädl, S., and Woike, J. K. (2018). Competitive escalation and interventions. Journal of Behavioral Decision Making, 31(5), 695–714. https://doi.org/10.1002/bdm.2084
Keil, M., Depledge, G., and Rai, A. (2007). Escalation: The role of problem recognition and cognitive bias. Decision Sciences, 38(3), 391–421. https://doi.org/10.1111/j.1540-5915.2007.00164.x
Molden, D. C., and Hui, C. M. (2011). Promoting de-escalation of commitment: A regulatory-focus perspective on sunk costs. Psychological Science, 22(1), 8–12. https://doi.org/10.1177/0956797610390386
Schultze, T., and Schulz-Hardt, S. (2022). Does change of responsibility reduce escalating commitment? A replication and theoretical extension. Journal of Experimental Psychology: Applied, 28(1), 189–204. https://doi.org/10.1037/xap0000366
Schulz-Hardt, S., Thurow-Kröning, B., and Frey, D. (2009). Preference-based escalation: A new interpretation for the responsibility effect in escalating commitment and entrapment. Organizational Behavior and Human Decision Processes, 108(2), 175–186. https://doi.org/10.1016/j.obhdp.2008.11.001
Simonson, I., and Staw, B. M. (1992). Deescalation strategies: A comparison of techniques for reducing commitment to losing courses of action. Journal of Applied Psychology, 77(4), 419–426. https://doi.org/10.1037/0021-9010.77.4.419
Sivanathan, N., Molden, D. C., Galinsky, A. D., and Ku, G. (2008). The promise and peril of self-affirmation in de-escalation of commitment. Organizational Behavior and Human Decision Processes, 107(1), 1–14. https://doi.org/10.1016/j.obhdp.2007.12.004
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
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)90005-2
Whyte, G. (1993). Escalating commitment in individual and group decision making: A prospect theory approach. Organizational Behavior and Human Decision Processes, 54(3), 430–455. https://doi.org/10.1006/obhd.1993.1018