Why money already spent keeps getting a vote

The rule, stated plainly

A sunk cost is any resource already committed and impossible to recover: money spent, time invested, effort expended. The formal position is unambiguous. When choosing between courses of action, only the future consequences of each option are relevant, because the past is identical under all of them and therefore cannot distinguish between them.

Almost everybody agrees with this in the abstract and almost nobody applies it consistently. The tendency to continue an endeavour because of what has already been put in is one of the more reliably reproduced findings in decision research, and it appears across very different domains, from small laboratory tasks to large capital projects that everyone involved knows are failing.

What is actually happening

Several mechanisms are proposed and they are not mutually exclusive. One is a reluctance to book a loss: abandoning the project converts an ambiguous situation into a definite failure, and the accounting entry hurts more than the ongoing drain. Another is self-justification, where continuing preserves the story that the original decision was sound. A third is simple aversion to waste, which is a generally useful instinct misapplied to a case where the waste has already occurred.

Underneath some of it sits a plain accounting confusion rather than any psychological attachment. People habitually reason about a project in terms of total cost against total benefit, which invites the money already spent into the sum, when the comparison that governs the decision is remaining cost against remaining benefit. Framed the second way, a project that looked worth finishing often stops looking that way, and nobody has had to overcome anything emotional to see it. Simply changing which column the past sits in does most of the work.

It is worth noting that the relative weight of these explanations is not settled, and the effect is stronger in some framings than others. Where responsibility for the original decision sits with the person deciding whether to continue, escalation tends to be more pronounced — which is a clue about how much of it is about the self rather than the arithmetic.

The version with other people watching

In an organisation, the sunk cost problem acquires a second layer that has nothing to do with individual psychology. Cancelling a project is a visible admission, and the person who cancels it frequently absorbs the reputational cost of the whole endeavour, while the person who quietly continues it distributes that cost across a longer period and a wider group.

This means that escalation can be individually rational within an incentive structure that punishes visible failure more than gradual waste. Calling that a bias is a category error. It is a design fault in the incentives, and the fix belongs to whoever designs them — for instance by making cancellation a normal, expected outcome with a defined process rather than an event requiring an explanation.

Where the rule genuinely does not apply

There are cases where past investment legitimately bears on the present, and confusing them with sunk cost reasoning leads to its own errors. If the prior investment created assets, knowledge or capability that make completion cheaper than starting fresh, that is not a sunk cost argument — it is a claim about future costs that happens to be caused by the past.

Commitments to other people are similar. Having promised something changes the future consequences of abandoning it, because there is a relationship and a reputation that will still exist afterwards. The test is whether the past changes any future outcome. If it does, it belongs in the calculation; if it only changes how the decision feels, it does not.

How to run the check

The cleanest device is to ask what a newcomer would do. If someone arrived today with no history, given the current state of the project, the remaining cost and the expected benefit, would they start? If the answer is no, and nothing about the prior investment changes future costs, then continuing is being driven by the past.

The second device is to set the exit condition before starting, in writing, in terms specific enough to trigger. A stopping rule agreed in advance is easier to follow than a judgement made in the middle, because the person who writes it has none of the commitment that the person facing it will have. This is the same principle as recording a forecast before an outcome: the past self is a more impartial judge than the present one, and the only way to consult it is to have written something down.