Sunk Cost Fallacy

The sunk cost fallacy is the tendency to continue an activity, investment, or commitment because of what has already been spent on it, rather than because of what it will produce from now on. Hal Arkes and Catherine Blumer named and documented it in "The psychology of sunk cost" in Organizational Behavior and Human Decision Processes 35(1) in 1985, where they define the sunk cost effect as a greater tendency to continue an endeavor once an investment of money, effort, or time has been made.

The logic it violates is straightforward. A decision changes only what happens next, so the only costs and benefits relevant to it are the ones that differ between the options. An amount already paid that no option recovers is identical under every option, which means it cannot distinguish them, which means it carries no information about which to choose. Where the reasoning actually breaks down is one step earlier, in deciding what is genuinely unrecoverable.

Sunk and recoverable are the two categories, and most financial situations are a mixture. People misclassify in both directions, and both directions are expensive.

The commoner error is treating a recoverable amount as sunk. "I have too much in this house to sell now" describes equity, which is money that comes back on a sale net of transaction costs, so it is not sunk at all and it is available to do something else. The same applies to the surrender value of a cash-value insurance policy and to the current market value of an investment: what you paid is history, but what it is worth today is a live asset you are choosing to keep. The rarer error runs the other way, treating a truly unrecoverable expense as though continuing could redeem it. A non-refundable deposit, a front-end sales charge already deducted, a completed repair, an exam fee, a year of tuition: those are gone under every option on the table, including the option of walking away.

So the useful sequence is a bookkeeping question first and a psychology question second. Split the amount already committed into the part that comes back under some available option and the part that does not. Put the first part into the decision and strike the second out of both columns. What remains is a comparison between two futures, which is the only comparison there is.

The reframing that does the work is a change of tense, not a change of resolve. The question "should I keep this?" invites a justification of the original purchase, because you made it. The questions that avoid the trap are ones the past cannot answer: given where things stand today, would I choose this from scratch? If a stranger handed me this position, this project, or this car this morning, would I keep it or sell it? What is the best use of the next dollar? Those formulations are not motivational; they simply omit the variable that has no place in the arithmetic.

Personal responsibility for the original decision intensifies it, which is the most actionable finding in the literature. Barry Staw's 1976 study "Knee-deep in the big muddy," in Organizational Behavior and Human Performance 16(1), found that people commit further resources to a failing course of action more readily when they were responsible for choosing it in the first place. The behavior is often called escalation of commitment, and it identifies the specific mechanism at work: what is being defended is the earlier judgment rather than the asset. It follows that the practical countermeasures are structural. Ask someone with no history in the decision what they would do. Write down in advance the condition under which you would stop, before there is anything to defend. And treat the honest admission that an earlier decision was wrong as the cheap part of the transaction rather than the expensive part.

It is a different error from the two it is most often merged with. Reluctance to sell at a loss is loss aversion working on a reference point, usually the purchase price, and it concerns how the outcome is scored. Treating money in one labeled pot as unavailable to another purpose is mental accounting, and it concerns which account an outcome lands in. The sunk cost fallacy is about the past expenditure functioning as a reason to continue, and it appears where there is no loss and no separate pot at all: a prepaid course you no longer want, an unused membership, a subscription paid annually. Those cases are the diagnostic ones, because nothing there is underwater and the pull is present anyway.

The honest boundary, because "sunk cost" is now used to dismiss reasoning that is sound. Three continuations are not fallacies. Where partly completed work genuinely lowers the remaining cost of finishing, the prior spending has changed the forward-looking numbers and belongs in them. Where abandoning carries a contractual penalty or a real reputational consequence, that penalty is a future cost and belongs in the comparison. And where someone keeps a commitment because they value being a person who keeps commitments, that is a preference about how to live rather than an error in arithmetic. The fallacy is specifically the case where the only argument for continuing is the size of what has already gone.

A hypothetical example, and the arithmetic that matters is which items appear in the columns at all rather than any exchange rate between hours and dollars.

Tomás paid $3,200 for a professional certification course, has completed about a third of it, and now needs 80 more hours of study plus a $400 exam fee to finish. A different credential he has since learned about costs $900 and roughly 50 hours, and it is the one his employer's job postings actually name.

If the $3,200 is genuinely unrecoverable, the comparison is between finishing, at $400 plus 80 hours, and switching, at $900 plus 50 hours. The $3,200 appears in neither column, because he pays it either way; it is the one number that cannot distinguish the options. The reasoning that goes wrong is "but I have already spent $3,200 on this one," which adds the same figure to both sides and then treats it as favoring one.

Now change one fact, because this is where most real cases sit. Suppose the provider refunds $1,100 of the unused portion on withdrawal. That $1,100 is not sunk, it is an asset available only under the switching option, so it belongs in the switching column: switching now costs $900 − $1,100 = −$200, meaning he ends up $200 ahead in cash and needs 30 fewer hours. Nothing about his preferences changed. One line moved from the wrong column to the right one, and it reversed the sign.

Pros

• Naming it points at a specific and checkable step: sort the committed amount into recoverable and unrecoverable before deciding anything.
• The correcting question is concrete rather than motivational, because "would I choose this today from scratch" omits the offending variable by construction.
• Because escalation is strongest for the person who made the original decision, an outside view is a cheap and reliable fix.
• It applies well beyond investing, to courses, memberships, repairs, renovations, and careers, which makes one habit useful in many places.

Cons

• The classification step is where people actually go wrong, and calling something a sunk cost does not make it one.
• It runs together with reluctance to realize a loss, so the same situation often involves two errors that need different corrections.
• The label has become a way to dismiss legitimate reasons to continue, including genuine completion economies and real contractual penalties.
• Applied without judgment it argues for abandoning things at the first difficulty, which is a different mistake with the same rhetoric.
• Recognizing it in yourself requires conceding that an earlier decision was wrong, which is the part the research says people resist.