Most behavioral findings in this hub did not survive scrutiny. The sunk cost fallacy did --- across labs, species, and a 2025 Journal of Finance paper that tracked the bias inside real corporate acquisitions. Here is why this one is different, and why every strategist should know it.

If you have been reading through this hub, you have watched a procession of canonical behavioral-science findings get dismantled. Power posing collapsed under its own author’s recantation. Ego depletion died in Hagger 2016. Money priming evaporated in preregistered replications. The marshmallow test shrank to a fraction of its claimed effect once family background was controlled. Stereotype threat, the Mozart effect, facial feedback, Bargh’s elderly-walking, the entire family of social priming --- finding after finding has either failed to replicate, shrunk dramatically, or required so many conditional moderators that the original “this is how human behavior works” framing no longer survives.

By now, a reasonable reader might suspect that most of behavioral economics is rotten. That conclusion would be wrong, and this article exists to explain why.

Because in the same forty years that produced all those replication failures, one cluster of findings kept holding up --- in laboratories, in cross-species comparisons, in field studies, and most strikingly in a 2025 Journal of Finance paper that used quasi-random cost shocks inside completed stock-merger transactions to measure the bias inside real corporate decisions worth billions of dollars. The bias is the sunk cost fallacy. It is one of the most empirically secure findings in all of behavioral economics, it is among the highest-leverage mental models a strategist can carry, and unlike many of the so-called biases your business-school professor invoked, this one does what it says on the tin.

This is the second anti-example article in a hub mostly full of takedowns (the first was the default effect; a third, Fitts’s Law, lives next door). It exists for the same three reasons. First, calibration --- the correct lesson from the replication crisis is not “behavioral science is broken” but “behavioral science produced a small number of robust, large, mechanism-grounded findings and a much larger number of fragile, contextually narrow findings, and the field’s main failure was treating those two categories the same.” Second, decision-usefulness --- for any executive deciding which mental models to actually carry into a strategy review, the sunk cost fallacy is one of perhaps five effects that pays back the time spent learning it many times over. And third, intellectual honesty --- if you spend a hub criticizing behavioral economics, you owe readers the parts that worked.

Here is the case for the sunk cost effect, as honest as I can make it, including the legitimate critiques.

What Arkes & Blumer 1985 Actually Tested

The sunk cost effect did not begin with Arkes and Blumer, but the paper that put it on the map and gave it its modern operational definition is Arkes, H. R., & Blumer, C. (1985). “The Psychology of Sunk Cost.” Organizational Behavior and Human Decision Processes, 35(1), 124—140. DOI: 10.1016/0749-5978(85)90049-4. It is one of the cleanest demonstrations in the social-science literature of an effect that almost everyone has now heard of but few people have read the original on.

The paper reported ten experiments. They were not flashy or politically charged, which is part of why they replicated --- they were small-stakes hypothetical scenarios delivered to a few hundred undergraduates and theater patrons, with the kind of within-subjects comparisons that let an effect be measured directly without complicated statistical machinery. The genius of the design was that Arkes and Blumer constructed pairs of scenarios that were identical in every economically relevant respect except for one feature: whether or not the decision-maker had already spent non-recoverable money on the option in question.

The single most famous scenario is the ski-trip vignette. Participants were asked to imagine they had purchased a $100 ticket for a ski trip to Michigan and, weeks later, a $50 ticket for a different ski trip to Wisconsin. They discovered the two trips were scheduled for the same weekend, the tickets could not be sold or refunded, and they had to choose one. Critically, they were also told they expected to have a better time on the Wisconsin trip. Under any forward-looking decision rule --- the $150 was gone regardless, and the question was just which weekend they would enjoy more --- they should have picked Wisconsin. Roughly 54 percent picked Michigan. The dollar magnitude of the prior commitment, not the expected enjoyment of the future activity, was driving the choice.

The other experiments stacked on top of that core finding. One showed that subjects who had paid full price for a season’s worth of theater tickets attended significantly more plays than subjects who had been randomly assigned a discount; the productions were identical, but the people who had committed more money showed up more often. Another, the airplane-prototype scenario, asked subjects to imagine they were a company president who had spent $9 million of a $10 million R&D budget on a stealth aircraft, only to discover that a competitor had just released a better and cheaper version of the same plane. Subjects who saw this version of the scenario were dramatically more likely to recommend spending the final million to finish the project than subjects who were told they were starting fresh with $1 million in unallocated budget facing the same competitor --- even though, looking forward, the two situations were economically identical and one was clearly a losing bet. Across these ten experiments, the modal participant violated the forward-looking expected-value rule whenever the sunk cost made the violation feel like an effort to recoup an investment rather than a forward-looking judgment.

What made the Arkes and Blumer paper enduring is not just that the effect was large --- it was --- but that the operational definition was sharp enough that any subsequent investigator could run essentially the same experiment and get a comparable answer. That sharpness is one of the most important predictors of whether a behavioral finding survives. Studies that hinge on a subjective construct like “ego depletion” or “implicit bias” can be re-operationalized in subtly different ways across labs, and the literature drifts apart. The sunk cost effect has a hard operational definition --- manipulate prior non-recoverable investment, hold forward-looking value constant, measure willingness to continue --- and that operational clarity is part of why it replicates.

Staw 1976 --- Escalation Of Commitment In Corporate Settings

The other foundational paper, predating Arkes and Blumer by a decade, is 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. DOI: 10.1016/0030-5073(76)90005-2. The title is a Pete Seeger lyric about a Vietnam-era platoon led by a captain who refused to turn back from a river crossing that was about to drown them all. Staw’s choice of title was not subtle. He was talking about LBJ and Vietnam as much as he was talking about corporate decision-making.

Staw’s methodology used 240 business-school students in a role-playing exercise. They were asked to imagine themselves as a corporate vice president allocating R&D budget between two divisions of a (fictional) company. After making an initial allocation decision, half the subjects (the “high responsibility” group) were told the division they had funded had performed poorly. The other half (the “low responsibility” group) were told the same poor result, but were told that the original allocation decision had been made by their predecessor --- not by them. Both groups were then asked to allocate a second round of R&D funding between the two divisions.

The result was sharp. Subjects who had personally made the initial losing allocation poured significantly more of the second-round budget into the same losing division than subjects who had inherited the bad decision. The same negative outcome led to different behavior depending on whether the subject was the one who owned the original mistake. Staw labeled this escalation of commitment and connected it directly to a class of organizational pathology --- managers who cannot abandon a project they championed even after the evidence has turned against them, governments who cannot exit a war they started even when victory is no longer plausible, executives who keep doubling down on a strategy that has visibly failed because admitting failure would mean admitting that the original decision was wrong.

The Staw paper is, in some respects, more important than Arkes and Blumer for the strategist’s purposes, because it identifies the specific mechanism --- personal responsibility for the prior decision --- that makes the sunk cost effect dangerous in organizations. The economic textbook version of the bias treats it as a kind of arithmetic error: people fail to write off non-recoverable costs. The Staw version reframes it as an identity-protection mechanism: people defend the original decision because the original decision was theirs, and walking away from it would require acknowledging error. That mechanism predicts where and when sunk cost reasoning will be most severe --- in highly visible projects, in decisions tied to the reputations of specific executives, in organizations where being wrong has career consequences. Most of the corporate examples of the sunk cost fallacy that you have ever read about are escalation of commitment in the Staw sense.

What The Modern Evidence Shows

Forty years on, the sunk cost literature is unusually well-developed for a behavioral-economics finding. Three streams of evidence are worth understanding.

The first is the meta-analytic stream. Roth, S., Robbert, T., & Straus, L. (2015). “On the Sunk-Cost Effect in Economic Decision-Making: A Meta-Analytic Review.” Business Research, 8(1), 99—138. DOI: 10.1007/s40685-014-0014-8 synthesized 98 effect sizes from monetary sunk-cost experiments. The headline finding was that the effect is present and reliably non-zero across decades of studies, with effect-size estimates in the moderate range. More usefully, the meta-analysis differentiated between utilization decisions (will I use this thing I already paid for) and progress decisions (will I keep funding this thing that has already absorbed investment), and found that the effect appears in both classes but is moderated by features like the specificity of the decision, the visibility of the prior commitment, and the nature of the forward-looking information. The meta-analysis did not find evidence that the effect is an artifact of publication bias or small-sample p-hacking; the pattern is structural.

The second is the cross-species stream, which is the kind of evidence that is essentially impossible for the social-priming literature to produce. Magalhães, P., & White, K. G. (2016). “The Sunk Cost Effect Across Species: A Review of Persistence in a Course of Action Due to Prior Investment.” Journal of the Experimental Analysis of Behavior, 105(3), 339—361. DOI: 10.1002/jeab.202 synthesized the laboratory work on whether non-human animals also escalate commitment in response to prior investment. The earlier (1999) review of this question had concluded that the cross-species evidence was inconclusive, but a more recent line of operant-conditioning research has produced experimental designs in which pigeons, rats, and mice show patterns broadly consistent with sunk-cost-like persistence. Sweis, B. M., et al. (2018). “Sensitivity to ‘Sunk Costs’ in Mice, Rats, and Humans.” Science, 361(6398), 178—181. DOI: 10.1126/science.aar8644 is the cleanest demonstration: in a tightly controlled foraging-decision paradigm, all three species --- including humans tested on a structurally parallel task --- were more likely to wait out the remainder of a delay if they had already invested time waiting. This is not the same as human cognitive sunk-cost reasoning (no one thinks the mice are worrying about reputational consequences), but it suggests that the behavioral tendency to persist in a course of action because of prior investment is more biologically deep than the elaborate cognitive-bias framing alone would suggest. The bias is downstream of decision architecture that long pre-dates verbal self-justification.

The third --- and for strategists, the most important --- is the field-evidence stream from corporate finance. Guenzel, M. (2025). “In Too Deep: The Effect of Sunk Costs on Corporate Investment.” Journal of Finance, 80(3), 1593—1646. DOI: 10.1111/jofi.13430 is one of the most carefully identified empirical studies of a behavioral bias in real corporate data ever published. Guenzel exploits the fact that in fixed-exchange-ratio stock mergers, the final dollar cost of the acquisition depends on aggregate stock-market movements between the signing of the merger agreement and the close of the deal --- movements that are essentially exogenous to the specific quality of the underlying business being acquired. This gives a quasi-random source of variation in the “sunk cost” the acquirer has paid for the target, holding constant the actual quality of what they bought. Guenzel finds that an interquartile increase in this quasi-random cost shock reduces the probability that the acquirer subsequently divests the acquired business by 8 to 9 percent. Acquirers who happened, for reasons unrelated to the target’s actual quality, to pay more for the deal then hold the deal longer and more reluctantly walk away from it. Crucially, the effect is concentrated in firm-years in which the acquiring CEO who signed the deal is still in office --- exactly the pattern Staw 1976 would predict if the mechanism is personal responsibility for the original decision. When the CEO leaves and a successor takes over, the original sunk cost stops binding. The empirical fingerprint matches the theory.

To put a fine point on this: there are not many behavioral biases for which we have (a) a clean laboratory demonstration with a sharp operational definition, (b) a 98-study meta-analysis showing the effect is robust, (c) cross-species evidence in pigeons, rats, and mice, and (d) a quasi-experimental field study using billion-dollar corporate decisions and a credible exogenous instrument. The sunk cost effect is one of those few. The evidentiary depth here is in a different class from the typical social-psychology citation.

A couple of qualifications are honest to make. First, not every preregistered replication has produced effect sizes as large as the originals. The Soman 2001 sunk-cost-time-versus-money result, for instance, replicated at smaller effect sizes in a 2023 preregistered registered report --- the basic phenomenon held, but the time-versus-money distinction was less clean than originally reported. This is the pattern that has held for the better-replicating behavioral findings generally: the direction and qualitative phenomenon survive, but effect sizes settle at something smaller than the original. Second, modern theoretical work has identified moderators that affect the magnitude: the personal-responsibility variable from Staw, the visibility and identity-tie of the project, the presence or absence of explicit kill-criteria, the framing of the continuation decision as completion-versus-abandonment versus restart-from-scratch. The sunk cost effect is not a constant that fires identically across all contexts. It is a structural tendency whose magnitude depends on circumstances we can identify.

But the headline --- that decision-makers systematically continue investing in losing courses of action because they have already invested in them --- is one of the most secure findings in the behavioral-economics canon. It is real. It is large. It is observed across humans and other species. It shows up in laboratory experiments, in historical case studies, and in carefully identified corporate field data. If you are going to carry around behavioral biases as mental models for strategy, this is one of the few you can actually rely on.

The Conditions That Amplify Sunk Cost Effects

Knowing that the bias exists is the easy part. Knowing when it bites hardest is what makes it useful. The combined evidence from Arkes and Blumer, Staw, the Roth meta-analysis, and the Guenzel corporate-finance work points to a fairly clear set of amplifying conditions.

Personal responsibility for the original decision is the single largest amplifier. Staw 1976 established this and it has held up. Executives are dramatically more reluctant to abandon projects they personally championed than projects they inherited. This is why the most reliable mechanism for terminating a failing project is to bring in a new decision-maker who has no career equity in the original choice. Guenzel’s finding that the corporate sunk-cost distortion is concentrated in firm-years when the original-deal CEO is still in office is the field-data version of this lab effect.

Visibility of the prior commitment is the second amplifier. Projects that are publicly known --- that have been announced to the board, to analysts, to customers, to the press --- are harder to terminate than equally bad projects that have lived quietly inside one division. The reason is that public commitment puts the executive’s reputation on the table in a way that private commitment does not. The decision to walk away is no longer just an admission that the project was wrong; it is an admission that the executive’s earlier confident announcement was wrong.

Identity investment is the third amplifier and the one most relevant to founder-led companies. When a project is part of how a leader sees themselves --- “we’re the company that is doing X,” “I am the executive who brought us into Y,” “this acquisition is the core of my strategic legacy” --- the cost of abandonment is not just career-reputational but psychological-identity-level. This is the mechanism behind some of the most spectacular corporate refusals to kill obviously failing initiatives in the historical record.

Sunk-cost ratio matters. Garland 1990 --- Garland, H. (1990). “Throwing Good Money after Bad: The Effect of Sunk Costs on the Decision to Escalate Commitment to an Ongoing Project.” Journal of Applied Psychology, 75(6), 728—731 --- established that the fraction of a budget already spent matters more than the absolute dollar amount. Subjects told that 90 percent of a project’s budget had been spent were dramatically more willing to fund the remaining 10 percent than subjects told the same dollar amount was 10 percent of a much larger project. This is why the rhetorical move “we are 90 percent of the way there” is so dangerous in project reviews; it triggers exactly the cognitive frame that maximizes the sunk cost bias.

Absence of explicit kill-criteria is the silent amplifier. Projects that were funded with no pre-specified conditions for termination have nothing to compare against, so the choice to continue versus abandon becomes a pure cost-benefit re-evaluation at every checkpoint --- and the prior investment colors that re-evaluation. Projects that were funded with explicit kill-criteria up front (“if user adoption is below X by month six, we shut it down”) give the decision-maker an external commitment device that can override the sunk cost frame.

When you see a corporate decision where several of these amplifiers stack --- a high-visibility project championed by a still-in-office executive who has tied their identity to it, with no pre-specified kill-criteria and a “we’re almost there” rhetorical frame --- you are looking at a near-certain candidate for escalation-of-commitment behavior. The empirical regularity here is robust enough that it can be used predictively.

How Sunk Cost Shows Up In Corporate Strategy

The Guenzel paper makes the M&A version of this concrete with field data, but the pattern shows up across most of the high-stakes decision categories that strategists actually deal with.

Failed acquisitions are the canonical case. The acquirer pays a premium, integrates poorly, sees the synergies fail to materialize, and then --- exactly as the sunk-cost theory predicts --- refuses to divest for years longer than the forward-looking value of the asset would justify. Guenzel’s 8-to-9-percent reduction in divestiture probability is a direct measurement of this distortion at scale. The Time Warner/AOL deal, the HP/Autonomy deal, the eBay/Skype deal, Microsoft/Nokia --- most of the storied M&A disasters of the last twenty-five years featured years of post-deal escalation before the eventual write-down.

Ongoing strategic initiatives past their breakeven are the second category. A new market entry that the company has been investing in for three years, with steadily-disappointing results, will keep getting funded long past the point at which the same investment in a different opportunity would obviously have higher expected return. The “we are 70 percent of the way to scale” rhetorical frame is doing exactly the work Garland 1990 predicted.

Software rewrites are a particularly clear case. A two-year platform-rewrite project that is running over budget and behind schedule almost never gets killed at month thirty; the project team and the engineering leadership have too much identity-equity in the work, the prior investment looks too painful to write off, and “we are months from being done” becomes the rhetorical drumbeat. The actual finish date keeps sliding to the right; the cost-to-finish keeps growing; the project absorbs years of organizational opportunity cost. The forward-looking choice --- kill the rewrite, accept the legacy stack, redirect the team --- is almost never made, even when it is obviously correct.

Sales-pipeline mismanagement is the everyday version. A sales rep who has spent six months working a deal will keep working it long past the point at which the same effort applied to a fresh pipeline would have higher expected value. Sales managers who set up CRM workflows that surface time-in-stage as a kill-criterion --- explicitly designed to combat sunk-cost reasoning --- consistently report meaningful pipeline-quality improvements.

Hiring decisions are the rarely-discussed version. A manager who spent months recruiting a hire who is not working out will defer the termination longer than a manager who inherited the same underperformer. The “but we invested so much in onboarding them” frame is precisely the sunk-cost frame.

Long-running R&D programs in pharma, semiconductors, energy, and aerospace are perhaps the most consequential corporate version, because the dollar amounts are so large and the time horizons are so long. The Concorde program --- which gave its name to “the Concorde fallacy” in the literature --- is the historical exemplar, but every major corporate R&D laboratory has its own version. The structural problem is that the kill-criteria for a ten-year program have to be pre-committed at year zero and defended through a decade of project-team pushback; very few organizations have the governance architecture to do this well.

The strategic point is not that sunk cost reasoning is universally bad. There are honest reasons to consider prior investment --- implicit option value, learning-curve effects, switching costs at the customer level. The strategic point is that the bias systematically over-weights the prior investment relative to the forward-looking value, and the magnitude of that over-weighting is now well-enough characterized that you can spot it in real decisions before the write-down arrives.

What’s Honest To Say About Sunk Cost Now

The honest summary, after forty years of research, looks like this. The sunk cost effect is real, robust, large, replicable, observed across humans and other species, and detectable in carefully identified field data using actual corporate decisions. The Arkes and Blumer 1985 demonstrations replicate. The Staw 1976 escalation pattern replicates. The Roth 2015 meta-analysis confirms the literature is structural rather than the product of publication bias. The Guenzel 2025 corporate-finance evidence is some of the cleanest empirical work on a behavioral bias ever published. The cross-species evidence in Magalhães 2016 and Sweis 2018 shows the underlying tendency is more biologically deep than human reputation-protection alone would explain.

Where the literature has been refined is around moderators, not the basic effect. Modern researchers have identified the conditions that amplify the bias --- personal responsibility, visibility, identity investment, sunk-cost ratio, absence of kill-criteria --- and the conditions that attenuate it (clear forward-looking information, separation of decision rights from project ownership, external pressure to justify continuation). Effect sizes in modern preregistered replications have sometimes been smaller than in the originals, which is the pattern across all well-replicating behavioral findings, but the direction and qualitative phenomenon are stable.

The sunk cost effect did not get its reputation through hype the way power posing or grit did. It got its reputation through a steady accumulation of converging evidence from independent research traditions --- laboratory cognitive psychology, organizational behavior, behavioral economics, comparative animal behavior, and corporate finance. That kind of multi-tradition convergence is one of the most reliable signals that a finding is real.

So when you read elsewhere in this hub that some celebrated behavioral finding has collapsed, you should not generalize that to all of behavioral economics. The replication crisis was a crisis of methodology in certain kinds of studies --- small-sample laboratory experiments with subjective constructs, single-paper effects without follow-up, findings that depended on flexible analytic decisions. Findings with sharp operational definitions, large effects, and convergent evidence from multiple traditions survived. The sunk cost effect is the cleanest example of what behavioral economics looks like when it works.

What This Means For Strategists

If you take one thing from this article and apply it to your work, take this: the sunk cost fallacy is the most reliable predictor of corporate decision-making failure in the empirical literature, and the structural conditions that amplify it are visible from the outside before the failure arrives. You can use this. Several tools follow directly from what the research shows.

Pre-commit kill-criteria when you fund a project. Before any significant initiative is funded, write down --- in advance, before any sunk costs accumulate --- the specific metrics that, if not hit by specific dates, will trigger termination. Make these criteria binding governance commitments, not aspirations. The empirical research is unambiguous that pre-specified kill-criteria are dramatically more effective than post-hoc evaluation. The reason is structural: you cannot make a forward-looking commitment to be objective about your own prior decisions, so you commit instead to an external criterion that does the objective evaluation for you.

Separate decision rights from project ownership. For significant kill/continue decisions, the person evaluating the decision should not be the person who originally championed the project. This is the Staw 1976 finding applied to governance. Boards that have an independent investment committee or strategy committee make better termination decisions than boards that ask the CEO to evaluate the CEO’s own initiatives. Companies that bring in a new general manager to assess an inherited business unit make better divestiture decisions than companies that ask the original GM.

Run the new-CEO thought experiment. Whenever you are evaluating whether to continue a major commitment, force yourself to write down what a freshly hired CEO with no prior involvement would do with the same information. This is the technique Andy Grove credited with breaking Intel out of the DRAM business, and it works because it explicitly invokes the counterfactual that the sunk cost bias suppresses. Most executives, when they actually do this exercise honestly, find that the new-CEO answer differs from the current-CEO answer on at least one major initiative.

Use base-rate comparisons, not project-specific re-evaluation. When deciding whether to continue funding a project, compare its forward-looking expected return to the base-rate return on alternative uses of the same capital --- not to the project’s own history. The forward-looking comparison is exactly what the bias suppresses; the project-specific re-evaluation is exactly what triggers the bias.

Conduct formal pre-mortems before major commitments. A pre-mortem --- Klein’s technique of asking the team to imagine, before the project starts, that it has failed and to write down why --- surfaces the kill-criteria that should be built into the funding commitment. The technique works because it forces the team to do the cognitive work of imagining failure while the prior investment is still zero. Doing the same exercise later, after months of investment, is dramatically harder.

Audit your portfolio for “concrete amplifier” combinations. For each significant ongoing initiative in your portfolio, score it on the five amplifiers: was the original decision personally championed by someone still in office, is it highly visible externally, is it identity-tied to the executive, is its sunk-cost ratio above 50 percent, and is it operating without pre-specified kill-criteria. Initiatives that score high on three or more of these are statistically likely to be receiving more investment than their forward-looking value warrants.

These tools are not exotic. They are well-established management practices that show up in the literature on capital allocation, project governance, and corporate strategy. What the behavioral-economics research adds is the reason they work and the prediction that organizations that fail to use them will systematically over-invest in losing projects. The sunk cost fallacy is the mechanism. Pre-commitment and decision-rights separation are the antidotes. The empirical record on both sides of that equation is now strong enough to act on with confidence.

What This Anti-Example Tells Us About Behavioral Science

The reason this article exists in a hub mostly devoted to replication failures is that the contrast is itself the lesson. The sunk cost effect did not need to be rescued by elaborate auxiliary hypotheses, redefined in narrower contexts, or qualified into near-meaninglessness to survive the replication crisis. It survived because of three structural properties of the original research.

First, sharp operational definitions. Arkes and Blumer’s experiments specified exactly what counted as a sunk cost, exactly what counted as a continuation decision, and exactly what counted as the forward-looking alternative. Subsequent investigators could implement the same operations and get comparable answers. The bias did not depend on a subjective construct that could be measured one way in one lab and another way in another.

Second, large detectable effects. The 54 percent versus alternative-choice asymmetry in the ski-trip scenario, the 86 percent attendance change in Staw’s escalation experiment, the 8-to-9 percent divestiture-rate change in Guenzel’s corporate data --- these are not d-equals-0.2 effects that disappear in a slightly larger sample. They are large enough that any properly powered study will detect them.

Third, convergent evidence from multiple traditions. The same phenomenon shows up in undergraduate-scenario experiments, organizational role-plays, animal-operant-conditioning studies, capital-allocation field data, and historical case studies of corporate failure. Cross-tradition convergence is much harder to fake or to produce by selective publication than within-tradition consistency.

When behavioral findings have these three properties --- sharp operational definitions, large detectable effects, convergent evidence --- they tend to survive replication. When they have none of these properties --- fuzzy constructs, small effects, evidence from a single research group --- they tend not to. The replication crisis was a hard sorting between these two categories, and it should have left intact a small core of robust findings. The sunk cost effect is one of the largest and most useful members of that core.

The broader lesson for anyone evaluating behavioral-science claims in 2026 is that the question to ask is not “is the source prestigious” or “is the finding widely cited” or “is the headline compelling.” Those signals were exactly what failed during the replication crisis. The questions to ask are: how sharply is the construct operationalized, how large is the effect, and how many independent research traditions converge on the same conclusion. By those criteria, the sunk cost effect passes with room to spare. Most of the headline-grabbing behavioral findings of the 1990s and 2000s did not. That is not a damning verdict on behavioral economics. It is just the right way to read its evidence base going forward.

Sources

  • Arkes, H. R., & Blumer, C. (1985). The psychology of sunk cost. Organizational Behavior and Human Decision Processes, 35(1), 124—140. DOI: 10.1016/0749-5978(85)90049-4
  • 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. DOI: 10.1016/0030-5073(76)90005-2
  • Garland, H. (1990). Throwing good money after bad: The effect of sunk costs on the decision to escalate commitment to an ongoing project. Journal of Applied Psychology, 75(6), 728—731. URL: https://psycnet.apa.org/record/1991-11275-001
  • Thaler, R. H. (1980). Toward a positive theory of consumer choice. Journal of Economic Behavior & Organization, 1(1), 39—60. DOI: 10.1016/0167-2681(80)90051-7
  • Roth, S., Robbert, T., & Straus, L. (2015). On the sunk-cost effect in economic decision-making: A meta-analytic review. Business Research, 8(1), 99—138. DOI: 10.1007/s40685-014-0014-8
  • Magalhães, P., & White, K. G. (2016). The sunk cost effect across species: A review of persistence in a course of action due to prior investment. Journal of the Experimental Analysis of Behavior, 105(3), 339—361. DOI: 10.1002/jeab.202
  • Sweis, B. M., Abram, S. V., Schmidt, B. J., Seeland, K. D., MacDonald, A. W., Thomas, M. J., & Redish, A. D. (2018). Sensitivity to “sunk costs” in mice, rats, and humans. Science, 361(6398), 178—181. DOI: 10.1126/science.aar8644
  • Guenzel, M. (2025). In too deep: The effect of sunk costs on corporate investment. Journal of Finance, 80(3), 1593—1646. DOI: 10.1111/jofi.13430

FAQ

Is it really irrational to consider sunk costs?

The strict economic answer is yes: from a forward-looking decision-making perspective, sunk costs are by definition non-recoverable and should not influence the choice between future options. The behavioral answer is more nuanced. Sunk costs can carry useful information --- they can signal commitment to counterparties, they can lock in optionality, they can preserve learning-curve effects. The bias is not that people sometimes consider prior investment; the bias is that people systematically over-weight the prior investment relative to the forward-looking expected value, in a way that produces predictably bad decisions. The well-identified field evidence in Guenzel 2025 measures exactly this distortion in real corporate acquisitions and confirms it operates at meaningful magnitude.

How do I actually avoid the sunk cost fallacy in my organization?

The single highest-leverage intervention is pre-committed kill-criteria. Before any significant initiative is funded, specify in advance the metrics and timelines that would trigger termination, and make those criteria binding governance commitments. The second is decision-rights separation --- the person evaluating continuation decisions should not be the person who originally championed the project. The third is the new-CEO thought experiment --- explicitly ask what a freshly hired executive with no prior involvement would do with the same information. All three are well-supported by the empirical literature and are cheap to implement.

What about emotional investments --- relationships, careers, life choices?

The behavioral mechanism is the same. The sunk cost bias does not care whether the prior investment was money, time, effort, emotion, or reputation; the same identity-protection machinery that drives corporate escalation also drives the tendency to stay in failing relationships, persist in careers we have outgrown, and continue pursuing degrees that no longer match our goals. The same antidotes apply --- pre-committed criteria (you should have decided in advance what would constitute “this is not working”), the equivalent of decision-rights separation (talking to someone who has no prior investment in your choice), and the new-life thought experiment (“if I were starting fresh today, would I make this choice”). The research literature does not differentiate meaningfully between monetary and non-monetary sunk costs at the cognitive level.

What about long-term R&D projects where breakthroughs require sustained investment?

This is the legitimate version of the question. Some R&D programs genuinely require long sustained investment past the point at which short-term metrics would suggest termination, and treating those as sunk-cost-fallacy candidates would mean killing breakthrough science prematurely. The honest answer is that this is a governance design problem rather than an argument against sunk-cost discipline. The correct design is to specify the long horizon explicitly at funding time, identify the technical milestones that would update the probability of eventual breakthrough, and pre-commit to evaluating those milestones --- not the short-term commercial metrics --- at each checkpoint. The organizations that do the best long-horizon R&D (some pharma companies, Bell Labs in its heyday, certain defense research programs) are precisely the ones that do this governance work formally rather than relying on executive intuition.

Does the sunk cost effect apply to time the same way it applies to money?

The evidence is mixed and the question is more interesting than the textbook version. Soman 2001 originally argued that the bias applies less cleanly to time than to money, but the 2023 preregistered replication of Soman found sunk cost effects for both time and money in modern samples. The fair reading of the current literature is that the time-versus-money distinction is real but smaller than the original work claimed, and that the practical implication for strategy is that time investments --- engineering hours, executive attention, organizational bandwidth --- should be treated as sunk costs in roughly the same way that monetary investments are.

How do I tell whether a project I’m involved in is suffering from sunk cost reasoning?

Run the diagnostic: was the original decision personally championed by someone still in the role, is the project highly visible externally, is it identity-tied to the leadership, is more than 50 percent of the projected budget already spent, and is it operating without pre-specified kill-criteria. If three or more of these are true, the project is statistically likely to be receiving more investment than its forward-looking value justifies. The second diagnostic is the new-CEO test: write down honestly what a freshly hired CEO with no prior involvement would do with the same information. If your answer differs materially from what your organization is actually doing, that gap is approximately the magnitude of the sunk-cost distortion.

Is the sunk cost effect actually present in non-human animals, or is that a stretch?

The cross-species evidence is genuinely interesting and the Sweis 2018 Science paper is the strongest case for it. In a carefully designed foraging-decision paradigm, mice, rats, and humans (tested on a structurally parallel task) all showed greater willingness to continue waiting out a delay if they had already invested time in the wait. The fair interpretation is not that mice are doing the same cognitive reasoning humans are; it is that the behavioral tendency to persist in a course of action because of prior investment is more biologically deep than the elaborate cognitive-bias framing alone would suggest. The decision architecture is older than the elaborate self-justification machinery humans have built on top of it. This is one of the reasons the effect replicates so well --- it is not entirely dependent on culturally specific cognitive habits.

Why does this article exist in a hub that mostly debunks behavioral findings?

Calibration, decision-usefulness, and intellectual honesty. The replication crisis broke a lot of behavioral-science claims, and a hub that documents those failures owes readers an honest accounting of what survived. The sunk cost effect is one of the cleanest, best-replicated, highest-leverage findings in the surviving core. Including it as an explicit anti-example to the takedowns lets a reader leave the hub with the right calibration: most of the loud, headline-grabbing behavioral findings of the 1990s and 2000s did not hold up, but a small core of well-operationalized, large-effect, multi-tradition-convergent findings did. The sunk cost effect is one of those, and for any strategist trying to use behavioral economics in actual decisions, it is one of the few you can rely on.

replication-crisis sunk-cost behavioral-economics strategic-decisions evidence-evaluation

Free Tool

Built for Experimentation Teams

GrowthLayer is the experimentation platform I built for CRO teams --- test management, AI-powered insights, and pattern recognition across your entire program.

Explore GrowthLayer → (opens in new tab)

· Start Free →

Share this article

LinkedIn (opens in new tab) X / Twitter (opens in new tab)

Copy link

Go deeper

Methodology The PRISM Method Case Studies $30M+ in Results Work Together Services & Mentoring

Experimentation and growth leader. Builds AI-powered tools, runs conversion programs, and writes about economics, behavioral science, and shipping faster.

About LinkedIn Newsletter

← Previous

Door-In-The-Face: The Persuasion Technique That Actually Replicates (With Specific Conditions)

replication-crisis sunk-cost behavioral-economics strategic-decisions evidence-evaluation
Share this article
LinkedIn (opens in new tab) X / Twitter (opens in new tab)
Atticus Li

Experimentation and growth leader. CXL-certified CRO practitioner, Mindworx-certified behavioral economist (1 of ~1,000 worldwide). 200+ A/B tests across energy, SaaS, fintech, e-commerce, and marketplace verticals.

About LinkedIn Newsletter