The Sunk-Cost Fallacy
This is the bias that masquerades as virtue. Quitting feels like waste; persisting feels like loyalty, discipline, character. So the sunk-cost fallacy is uniquely dangerous — not because it's hard to understand, but because abandoning it feels like betrayal. You're not just fighting a faulty calculation; you're fighting the part of you that doesn't want to admit the last decision was wrong. Let's pull it all the way down — what's solid in the research, what's overstated, and the seam between the cold math and the human pull that nobody escapes.
What it is (and where it came from)
The sunk-cost fallacy is the tendency to continue an endeavor because of what you've already invested in it — money, time, effort, or pain — rather than because of what it will yield going forward.
A "sunk cost" is, by definition, a cost that has already been incurred and cannot be recovered regardless of what you do next. The non-refundable ticket. The two years in the wrong degree. The $40,000 already sunk into a renovation. The decade in a marriage that died years ago. The defining feature is irreversibility: no future choice can claw the cost back.
The rational principle is brutally simple and not seriously contested in economics or decision theory: a sunk cost should be irrelevant to a forward-looking decision. When you choose what to do now, the only things that should count are the marginal future costs and the marginal future benefits — the additional money/time/effort it will take from this point, weighed against what you'll get from this point. What's already gone is gone. It is data about the past, not a stake in the future. The textbook injunction is "ignore sunk costs." The fallacy is failing to ignore them — letting an unrecoverable past expenditure drive a present decision.
The term comes out of economics, where "sunk cost" has been standard cost-accounting vocabulary for over a century — distinguished from variable or recoverable costs. But the psychological discovery — that ordinary people systematically and predictably violate the "ignore sunk costs" rule — belongs to a specific paper.
In 1985, Hal Arkes and Catherine Blumer published "The Psychology of Sunk Cost" in the journal Organizational Behavior and Human Decision Processes. This is the foundational empirical study, and it's worth knowing what they actually did, because it's cleaner than most of what gets attributed to it.
Arkes and Blumer ran a series of experiments — the paper reports around ten — designed to isolate sunk cost from every rational reason you might have for persisting. Two of them became canonical:
- The theater-ticket / season-pass study. Patrons buying season tickets to a campus theater were randomly given one of three prices: full price, a small discount, or a large discount. Same seats, same shows — the only difference was how much they'd sunk. Arkes and Blumer found that, over roughly the first half of the season, the full-price group attended more plays than the discount groups. The only thing that differed was the sunk amount, and it changed behavior — exactly what the rational model says should not happen. (Worth noting honestly: the effect was concentrated in the first half of the season and faded in the second half. The original finding was real but bounded, not a permanent compulsion.)
- The airplane / radar-blank hypothetical. Subjects imagined they were the president of an aviation company who had spent $9 million (90% of a $10 million budget) building a radar-blank plane, only to learn a competitor's plane is better in every way. Should you spend the last $1 million to finish? A large majority said yes, finish it — throwing good money after bad. A control group asked whether to spend $1 million to start an identical project from scratch, with no prior investment, mostly said no. Same forward-looking decision; the only difference was the $9 million already spent. The sunk cost flipped the answer.
That second study is the heart of it: when the future prospects are held identical, the mere existence of a prior, unrecoverable investment pushes people toward continuing. Arkes and Blumer's contribution wasn't noticing that people throw good money after bad — that's old folk wisdom ("don't cry over spilt milk," "in for a penny, in for a pound"). Their contribution was demonstrating it experimentally with the rational reasons stripped out, and naming the psychology behind it.
The real science / evidence
A few things are well-established and a few things are shakier. Let's keep them separate, because cognitive science has a serious replication crisis and this literature is not exempt.
What's solid:
The basic sunk-cost effect — that prior investment biases people toward continuation in controlled scenarios — is one of the more robustly replicated findings in judgment-and-decision-making. It shows up across the hypothetical-scenario paradigm (the airplane-type problems) repeatedly and across cultures. Arkes and Blumer's airplane-style problems in particular have been re-run many times in classrooms and labs and reliably produce the predicted asymmetry. When you can hold future prospects constant and vary only the sunk amount, people lean toward finishing what they paid for.
The closely related organizational phenomenon — escalation of commitment — has its own deep literature, anchored by Barry Staw, whose 1976 paper "Knee-deep in the Big Muddy" launched the field. Staw showed that decision-makers who had personally made a prior investment that turned out badly were more likely to increase their commitment to it — to double down on a failing course of action — than people who inherited the same situation without having chosen it. This is a crucial refinement: escalation is strongest when personal responsibility is involved. It's not purely about the money sunk; it's about your having sunk it. That points the mechanism toward self-justification, which we'll get to.
What's more contested:
Several things widely repeated as "facts" about sunk cost deserve hedging:
- The money/time asymmetry. It's often claimed that people fall for sunk money but not sunk time — that time doesn't trigger the bias the way dollars do. There's genuine experimental work suggesting time behaves differently from money here, but the picture is mixed and the conditions under which time does or doesn't drive escalation are not settled. Treat "time isn't a sunk cost trigger" as a live hypothesis, not an established law.
- Sunk cost in non-human animals and children. There's a real and interesting line of research reporting sunk-cost-like behavior in rodents and other animals, and debate over whether young children show it less than adults (which would be strange for a "primitive" bias). These are fascinating but unsettled. Do not state as fact that "even pigeons commit the sunk-cost fallacy" — the findings exist but the interpretation is argued over.
- Effect sizes and universality. Like much of the heuristics-and-biases canon, the cleanest demonstrations come from hypothetical questionnaire scenarios. How large the effect is in high-stakes real-world decisions, and how much individual variation there is, is harder to pin down. The direction of the effect is robust; the magnitude in the wild is genuinely uncertain.
So: the existence and direction of the bias — solid. The exact boundary conditions, the time-vs-money story, the animal/developmental claims, the real-world magnitude — varying degrees of "we're still arguing." An honest study app says both.
The mechanism
Why does a rational creature keep paying into a hole? The "ignore sunk costs" rule is so simple a child can follow it for someone else's decision. The fallacy is what happens when it's your own. There isn't one mechanism — there are several reinforcing ones, and the honest answer is that researchers don't fully agree on their relative weight.
1. Loss aversion (the prospect-theory engine).
Kahneman and Tversky's prospect theory (1979) established that losses loom larger than equivalent gains — the pain of losing is roughly felt as more intense than the pleasure of gaining the same amount. (The frequently cited "2-to-1" or "2.25-to-1" loss-to-gain ratio comes from Tversky and Kahneman's 1992 cumulative-prospect-theory paper; it's a rough average from their data, not a universal constant — the real ratio varies a lot by person and context. Cite it as "roughly twice," not as a fixed law.)
Here's how loss aversion drives sunk cost: as long as you keep the project alive, the loss is unrealized — still abstract, still "maybe it works out." The moment you quit, you have to book the loss. You have to look directly at the $40,000 or the two years and call it gone. Quitting converts a vague, deferred loss into a sharp, certain, present one — and loss aversion makes that conversion feel unbearable. So you persist, not because continuing is smart, but because continuing lets you keep not feeling the loss. The fallacy is partly an avoidance behavior: a way of postponing pain.
2. Self-justification (the ego engine).
This is Staw's escalation story, and it connects to cognitive-dissonance theory (Festinger). When you've publicly or privately committed to a course of action, quitting is an admission that the original decision was wrong. That creates dissonance between "I am a competent person who makes good decisions" and "I just wasted $40,000." The cheapest way to resolve the dissonance — cheaper than rewriting your self-image — is to keep going and tell yourself it'll pay off. Persistence becomes proof that the original call was right after all. This is why escalation spikes with personal responsibility: the more the failing project is yours, the more your identity is staked on not abandoning it. The sunk cost isn't only financial; it's reputational and psychological. You're not protecting the money. You're protecting the story that you're not a fool.
3. Waste aversion / the "throwing it away" frame.
Arkes argued that a core driver is a learned aversion to appearing wasteful. We're taught from childhood that waste is bad — finish your plate, don't let things go to waste. Quitting feels like waste ("all that money/effort, for nothing"), while persisting feels like honoring the investment. The trick the mind plays is treating the sunk cost itself as the thing being wasted — when in fact the cost was already spent the moment it became unrecoverable. Quitting doesn't waste it; it was already spent. The only thing quitting "wastes" is the future resources you'd otherwise pour after it. The frame is exactly inverted from reality.
4. Honoring the cost / the moral overlay.
Layered on top of all of this is a quasi-moral sense that the sacrifice already made obligates you. The hours your parents spent on your medical-school dreams. The years your spouse gave. The blood you've already left on the field. Walking away can feel like dishonoring not just your own investment but other people's — a betrayal of the sacrifice itself. This is where sunk cost stops being a clean cognitive error and becomes a genuine moral knot, and it's the seam where the bias is hardest to simply "correct away." More on that below, because it's the part the pop-psychology version completely misses.
Sharp examples
The Concorde. This is the textbook case, to the point that the sunk-cost fallacy is sometimes called the "Concorde fallacy" (a term that originated in evolutionary biology, where it was used to describe animal behavior, before being applied to the airplane). Britain and France poured money into the supersonic jet for years after it was clear it would never be commercially viable. The shared, repeated justification was explicitly sunk-cost reasoning: we've already invested so much, we can't stop now. Two governments escalated commitment to a project they knew was a financial loser, because abandoning it meant booking an enormous, public, embarrassing loss. The plane was a marvel of engineering and a monument to the fallacy at the same time.
The non-refundable ticket / the all-you-can-eat buffet. Small and personal: you've paid $30 for the buffet, you're full, and you keep eating to "get your money's worth" — making yourself uncomfortable to honor a cost that's identical whether you eat one more plate or not. The $30 is gone either way. The only live variable is your comfort, and you're sacrificing it to a number that can't change.
The bad relationship / the decade-in marriage. "We've been together seven years, I can't leave now." The seven years are sunk. They cannot be recovered by staying an eighth. The only question a forward-looking mind can ask is: from today, does continuing make my life better than the alternative? But here the moral overlay bites hardest — the years feel like an investment that staying will "redeem" and leaving will "waste," and there's a real other person whose sacrifice you'd be acknowledging as also-spent. This is where the cold logic is correct and insufficient — see the honest cut.
The failing project at work. A manager championed a software rollout. Eighteen months and a budget overrun in, it's clearly a dog. Killing it means standing in front of the board and saying "I was wrong, and we lost a year and a budget." Continuing means the loss stays hidden a little longer and maybe it turns around. Staw's research predicts exactly what usually happens: the person who chose the project escalates; a fresh manager brought in cold is far more willing to pull the plug. Same data, different responsibility, opposite decision. Companies sometimes exploit this deliberately — bringing in outside leadership precisely because they have no sunk cost to defend.
The PhD / the career. A decade into a track you now hate, the years already spent become the loudest argument for staying. But years can't be un-spent by spending more years. The honest version of the question isn't "can I justify the time I've already put in?" — it's "if I had these exact skills and this exact age today, with no history, would I choose this path forward?" If the answer is no, the history is lying to you.
Where it gets distorted / oversold (the honest cut)
This is where a serious treatment has to push back on the slogan version of the bias, because "ignore sunk costs, just quit" is itself a way of being wrong.
1. Not all persistence is the sunk-cost fallacy. The single most common error in applying this concept is calling any persistence "sunk cost." Sometimes continuing is correct because the future expected value is genuinely positive — you're close to a breakthrough, the marginal cost to finish is small, the learning curve is about to pay off. Grit, deliberate practice, seeing hard things through — these often look like sunk-cost behavior from the outside but are actually sound forward-looking bets. The fallacy is specifically persisting because of the past investment when the future math says stop. If you'd make the same choice with the sunk cost magically erased, it's not the fallacy — it's just a decision. The bias is only present when the past is doing work the future can't justify.
2. Sunk costs are real information, even if they're not stakes. The strict rule says "the past cost is irrelevant." But the fact that something cost more than expected, or is failing, is genuine evidence about the future — it can tell you your estimates were optimistic, the domain is harder than you thought, the team is wrong. The discipline isn't to erase the past; it's to use it as information about future prospects without treating it as a debt the future owes you. A subtle but real distinction the slogan flattens.
3. The commitment-device defense. Sometimes the inability to walk away is a feature. Burning the boats, the marriage vow, the public commitment — humans deliberately create sunk costs and points-of-no-return precisely because our future selves are weak and will want to quit at the first hard moment. Some persistence that violates cold marginal logic is actually a pre-commitment paying off. Not everything that ignores the marginal calculus is a mistake; sometimes the marginal calculus is exactly the thing you wisely tied yourself to a mast against.
4. The cruelty of the strict reading. Told as pure logic, "ignore sunk costs" can become a license to treat every relationship, every promise, every loyalty as a continuously re-optimized transaction — drop anything the moment its forward EV dips. That's not wisdom; it's a recipe for a life with no commitments and no roots. The honoring instinct that drives the fallacy is also the instinct behind loyalty, perseverance, and keeping your word. The mature move isn't to amputate it. It's to know when you're honoring something worth honoring and when you're just refusing to feel a loss.
So the honest cut runs both ways: the slogan version under-warns (most people don't quit nearly soon enough on dead projects) and over-claims (not all persistence is a bias, and some inability-to-quit is a virtue). Hold both.
The debates & frontier
- Is it one bias or several? "Sunk cost," "escalation of commitment," and "waste aversion" overlap but aren't identical, and researchers debate how much they're the same underlying phenomenon versus distinct effects with different mechanisms. Escalation requires personal responsibility and a sequence of decisions; the classic sunk-cost effect can appear in a one-shot hypothetical. They're cousins, not twins.
- Loss aversion under fire. A large part of the standard mechanism story rests on loss aversion, and loss aversion itself has come under serious scrutiny in the last decade. Some researchers argue the evidence for a universal, large loss-aversion effect is weaker and more context-dependent than the famous "2x" framing suggests. Since loss aversion is load-bearing in the usual explanation of sunk cost, that wobble matters. The behavior (sunk-cost persistence) is robust; the favored explanation for it is less settled than textbooks imply.
- The replication crisis backdrop. Judgment-and-decision-making is part of the broader psychology replication reckoning. Many classic effects have shrunk or failed under rigorous re-test. The core sunk-cost asymmetry has held up comparatively well, but specific elaborations (the time/money asymmetry, the animal and developmental claims, precise effect sizes) should be held loosely. The intellectually honest stance is: strong on the central effect, agnostic on the periphery.
- Debiasing — does anything work? A genuine open frontier. Simply knowing about the sunk-cost fallacy provides surprisingly weak protection — you can teach it and people still fall for it on their own decisions. Structural fixes seem more reliable than willpower: reframing the decision as a fresh choice, having someone with no skin in the game make the call, setting kill-criteria in advance (before any cost is sunk). The frontier question is whether the bias can be trained out of an individual or only engineered around.
How to apply it
- Run the fresh-start test. The single most reliable debiasing move: ask "If I were arriving at this exact situation today, with no history and no prior investment — would I choose to start this from here?" If the answer is no, the only thing keeping you in is the past, and the past can't be bought back.
- Name the loss you're avoiding. Most sunk-cost persistence is loss-avoidance in disguise. Ask: what loss am I refusing to book by not quitting? Make it concrete and present. Once you see that continuing doesn't prevent the loss — it already happened — but only adds future cost to it, the spell weakens.
- Separate the sunk cost from the information. Use the failure as data ("my estimates were optimistic, this domain is harder than I thought") without treating it as a debt ("I've put too much in to stop"). Information about the future: keep. Obligation to the past: drop.
- Set kill-criteria before you start. The cleanest defense is structural and pre-commitment-based: before you sink anything, decide in writing what result would make you walk away. A decision made before the cost is sunk is immune to the cost.
- Outsource the cold call. When you can't trust yourself — because it's yours and your identity is staked on it — give the decision to someone with no responsibility for the original choice. This is exactly why fresh leadership kills projects founders can't. Borrow a mind with no sunk cost.
- Distinguish honoring from clinging. Before you quit, ask honestly: is continuing a forward-looking bet, a wise pre-commitment, or a genuine loyalty I'd choose to keep — or is it just refusal to feel a loss? The first three can be right. The last one is the fallacy. The whole skill is telling them apart.
Connects to
The Law of Sacrifice — This is the law sunk cost lives inside. Sacrifice says nothing is gained without something given up; to get, you must release. The sunk-cost fallacy is the shadow of Sacrifice — the inability to release what has already been sacrificed into a dead end. The law's mature reading demands you let go of the spent thing to free your hands for the live one; the fallacy is the white-knuckled refusal to open your hand, treating the past sacrifice as a hostage you can still ransom. Where Sacrifice says release to gain, sunk cost whispers cling to redeem — and the cruelty is that clinging guarantees the very waste it's trying to avoid. The releasing it asks for is the hardest because it requires booking the loss as final.
Loss Aversion (Kahneman & Tversky, prospect theory) — the engine underneath. Quitting forces you to realize a loss, and losses loom larger than gains, so you persist to keep the loss unbooked.
Escalation of Commitment (Barry Staw, "Knee-deep in the Big Muddy," 1976) — the organizational and personal-responsibility version: you double down hardest on the failures you chose.
Cognitive Dissonance (Leon Festinger) — the self-justification mechanism: quitting admits the first decision was wrong, so persistence defends the ego, not the asset.
Cause & Effect — the corrective. Forward-looking marginal reasoning is causal reasoning; the fallacy is letting an effect already in the past masquerade as a cause of present action.
Arkes & Blumer (1985) — the founding empiricists of the psychology of sunk cost · Hal Arkes — the waste-aversion line of explanation · the Concorde fallacy — the cultural touchstone and the link back to evolutionary biology, where the term was born.
The past already took its payment. Every additional dollar, hour, and year you feed it is a separate, brand-new decision — and the only question that decision can honestly ask is "what now," never "what already."