The sunk cost fallacy
The sunk cost fallacy is the mistake of continuing something because of what has already been spent on it, and it is a claim about the past standing in for a judgment about the future. A sunk cost is money, time or effort that is already gone and cannot be recovered by any choice you make now. Since no future decision can bring it back, it carries no information about which option is better, and yet it reliably changes what people choose.
What the sunk cost fallacy is
The sunk cost fallacy is a decision fault with one foot in logic and one in psychology, which is why it carries a name from each side. Judged as an argument it is a fallacy: the premise "we have already spent two years on this" does not support the conclusion "therefore we should spend a third", because the relevant comparison is between the value of continuing and the value of stopping, and both of those lie in the future. Judged as a behavior it is a bias, because the error runs in one direction, appears in most people, and can be produced in an experiment on demand.
Economists usually say sunk costs are irrelevant to a rational decision, and psychologists usually say they are among the most reliable influences on a real one. Both statements are true, and the distance between them is the whole subject. Among the members of the cognitive bias family, this one is unusual precisely because it has a clean normative standard: you can say exactly what the correct answer is, which you cannot always do for the halo effect or the Dunning Kruger effect.
The one sentence version, and how to use the term in a sentence
In one sentence: a cost you cannot get back should not appear anywhere in your reasoning about what to do next. That is the entire rule, and it is short enough to apply while someone is still talking.
Used correctly in a sentence, the term names the structure of a decision, not a person's character. "We are keeping the old system because of the sunk cost fallacy: the four hundred hours of migration work are gone whichever system we run next year." That sentence identifies the irrelevant premise. Compare the lazy version, "you are committing the sunk cost fallacy", which names no premise and can be said about any decision to persist, including a correct one.
The evidence: season tickets and an unfinished aircraft
The evidence behind the sunk cost fallacy is a set of studies by Hal Arkes and Catherine Blumer, published in 1985 as "The psychology of sunk cost". In the best known of them, people buying a theater season ticket were randomly given either the full price or a discount at the point of purchase. Everyone received the same tickets to the same plays. Those who had paid the full price attended more of the plays in the first half of the season than those who had received a discount, and the pattern faded later in the year.
Randomization is what makes this decisive. The groups did not differ in how much they liked theater, since a coin decided who paid what. The only difference was how much had already been spent, which is precisely the quantity that should have had no effect on whether an evening out was worth the trip.
Arkes and Blumer also used a written scenario in which a company has spent most of a budget developing an aircraft and then learns a competitor is about to release a better one. Respondents who were told about the money already spent were far more willing to finish the project than respondents given the same future prospects with no spending history attached. The pattern has a nickname from a different field: biologists studying continued investment in a failing venture have called it the Concorde fallacy, after the airliner project that ran long past the point where its economics were in doubt.
Where the sunk cost fallacy shows up
The sunk cost fallacy shows up wherever effort accumulates and the record of it is visible. Six ordinary settings:
- Software projects. A rewrite is eighteen months in, everyone agrees the design was wrong, and the eighteen months are the argument for month nineteen.
- Study and training. A person finishes a qualification they no longer want because of the terms already paid for.
- Home renovation. A contractor's estimate doubles, and the money already sunk into the foundations is used to justify the rest.
- Reading and watching. Finishing a bad book at page 300 because of the first 299, when the only question is whether the next hour is worth spending.
- Queues and waiting. Staying in a slow line because of the twenty minutes already given to it.
- Hiring decisions. Keeping a poor arrangement in place because of the effort spent setting it up, which is a cost that recruitment will not refund.
What the sunk cost fallacy is confused with
The sunk cost fallacy is confused with two things, and one of them is not an error at all: the gambler's fallacy, and a perfectly sound decision to continue because the remaining work is still worth doing.
The gambler's fallacy, which misreads the past as a prediction
The sunk cost fallacy and the gambler's fallacy are both errors about the past, and they are opposite errors. The gambler's fallacy treats past outcomes as though they change the odds of the next independent event, so a run of losses feels like evidence that a win is due. The sunk cost fallacy treats past expenditure as though it changes the value of the next choice. One misreads history as a prediction, the other misreads history as an obligation.
The question that separates them: is the past being used to forecast an outcome, or to justify a commitment?
A sound decision to continue, which is the confusion that does real damage
The more costly confusion runs the other way: treating every decision to persist as the fallacy. Continuing is correct whenever the value of the work still to come exceeds the cost still to be paid, and that calculation has nothing to do with what was spent already. A project 3 years in, with 2 months of work left and a buyer waiting, should be finished, and the 3 years are no part of the reason. A project 3 years in with 4 more years to go and no buyer should stop, and again the 3 years are no part of the reason.
The question that separates them: if I strike out every mention of what has already been spent, does the argument for continuing still stand up? If it does, this is not the fallacy, whatever it looks like from outside. If nothing survives the deletion, the past is doing all the work.
| Error | What the past is wrongly said to do | Sentence that gives it away |
|---|---|---|
| Sunk cost fallacy | Justify continuing | We have come too far to stop now |
| Gambler's fallacy | Predict the next outcome | It has to turn around soon |
| Confirmation bias | Supply supporting evidence only | Everything we have looked at says it is working |
| Loss aversion | Make the loss feel larger than the equivalent gain | I cannot bear to write it off |
There is also a way to misuse the label. If someone argues for continuing on grounds of future value, contractual penalties, or what the team has learned that will pay off next quarter, that is a live argument about the future and answering it with "sunk cost fallacy" attacks a case they did not make. That move is close to a straw man fallacy, and a label is never a rebuttal.
Is the sunk cost fallacy always bad, and what reduces it
It is bad in the narrow sense that the reasoning is invalid, and not always bad in its effects, which is a distinction worth keeping. A rule of finishing what you start has real benefits: it makes you predictable to other people, it stops you abandoning hard things at the first difficulty, and it protects you from the opposite error of restarting forever. The fallacy is the reasoning, not the persistence. Persistence justified by future payoff is not the fallacy at all.
To overcome the sunk cost fallacy, restate the decision so the spent resources cannot enter it. Five moves that work:
- Ask the newcomer question: if I arrived today, with this project at this stage, would I fund it?
- Set the stopping rule in advance, written down at the start, naming the result that ends the effort.
- Give the decision to someone with no history in it, since the fallacy is weakest in a person who spent nothing.
- Separate the accounting from the choice, recording the loss honestly in one document and deciding in another.
- Name what continuing costs, because the money not yet spent is the only money still under your control.
The test to run before you carry on
The test to run before you carry on is the question that removes the past from the frame: if the work done so far had been done by someone else, for free, would I choose this option now? If the answer is no, the past effort is doing the arguing.
A second question catches the version that hides behind reputation: am I continuing because of what it produces, or because stopping would mean admitting the earlier decision was wrong? The second is a real cost, but it is a cost of embarrassment, and it should be priced as one rather than disguised as strategy.
Sunk cost in relationships and other commitments
Applied to relationships, the sunk cost pattern is the same reasoning with a much heavier context, and it deserves care rather than a slogan. "We have been together seven years" is, as a premise, exactly parallel to "we have spent eighteen months on this rewrite": the seven years are not recoverable and cannot by themselves tell you what the next year will be like.
The honest limit is that a relationship is not a project, and several things that look like sunk costs in it are not. Shared obligations, promises made to another person, and the interests of people who depend on the arrangement are all future facing, and they belong in the decision. What does not belong is the bare quantity of time already spent, used as though it were an argument. This page describes a reasoning error and offers no advice about anyone's life. The test is the same one, asked gently: is the reason to continue something that lies ahead, or something already behind?