Sunk cost
A Socratic walk-through of sunk cost — reasoned out one step at a time, not lectured.
The question we started with
THE QUESTION #Why do people keep pouring money into a project they already believe will fail?
A committee agrees, out loud, that the programme will not deliver. Then it votes to fund another year. Nobody in the room is stupid and nobody is lying about the forecast. So what exactly is being weighed, if not the forecast?
Ask it the other way round. If a stranger walked in this morning holding the same budget and the same evidence, with no history at all, would they start this project today? If the answer is plainly no, and the room continues anyway, then something is entering the decision that a newcomer does not have. What could that something be?
Reasoning it through
REASONING #The only candidate is the past. And here is the awkward observation: a decision is a choice between futures. Whichever option is taken, the money already spent is spent — it appears identically in both branches, so it cannot distinguish them. Anything identical in both branches cancels. That is the whole normative claim, and it is unusually clean: only costs and benefits that differ between the options can bear on the choice.
Notice how that principle cuts. It says to ignore money already gone, yes. But it also says to ignore a future cost that you pay either way — the rent on the building, if you keep the building regardless. And it says to include a cost that is entirely about the past's grip on the future: a contractual penalty for cancelling is a real payment you make only in the stopping branch, so it counts.
So why is the past so hard to drop? One reason is arithmetic that feels like reasoning: "we are eighty percent of the way there" invites the thought that finishing is cheap, when what matters is only the remaining twenty percent measured against what finishing is worth now. Another is that stopping converts a running expense into a stated loss. Nothing changes materially, but the loss becomes a number someone has to say aloud, with a name attached.
And here we should be careful, because a third reason is not a mistake at all. Persisting can be genuinely rational. If a firm that abandons projects halfway finds that suppliers demand better terms and partners stop signing, then finishing buys a reputation for finishing, and that is a real future benefit purchased in the future. Past spending can also carry information: whoever authorised it may have known something, and the trouble encountered since is evidence about difficulty ahead. Escalation can be self-justification, and it can be inference. From outside, the two look identical.
How solid is the effect itself? The classic demonstrations are Arkes and Blumer's in 1985 — theatre subscribers who happened to receive a randomly discounted season ticket attended fewer plays than those who paid full price, which is exactly the wrong direction for anyone weighing only the evening ahead. The finding is real, but the literature since is more mixed than the textbook telling: effect sizes vary by framing and setting, and some replications find it much weaker. It is a genuine tendency, not a law.
The analogy
THE ANALOGY #Think of a tunnel being dug through a mountain. However much rock is already behind you, the question at every morning's start is the same: how much rock remains, what will it cost to move, and is the far side still worth reaching? The tunnel does not care how tired you are.
In a tunnel the remaining distance is measurable, whereas in a failing project the remaining cost is precisely what is in dispute — and the rock already moved genuinely tells you something about the rock ahead, which is why "ignore the past" is a rule about costs, not about evidence.
Clarifying the model
THE MODEL #One distinction is worth making sharply, because the two ideas are routinely merged. Loss aversion is a claim about how outcomes are weighed at the moment of choice: a loss of a given size hurts more than an equivalent gain pleases. Sunk cost is a claim about what enters the calculation at all: an unrecoverable past expenditure changes the choice when it should not. Loss aversion is one plausible explanation of the sunk cost effect — stopping makes a paper loss real — but it is a separate claim, and it is not the only candidate; self-justification and a reluctance to look wasteful do similar work.
The other refinement: "sunk" means unrecoverable, not merely old. If the rig bought last year can be sold, its resale price is live money and belongs in the decision. The test is never when was this spent, but does this amount differ between continuing and stopping.
A picture of it
THE PICTURE #How to readRead left to right for when the money moves, and bottom to top for whether it differs between continuing and stopping. Only the top half may enter the decision. The bottom-left corner is the classic error — money already spent, pulling on a choice it cannot affect — but note the bottom-right corner is an error too: a future cost you pay under either option is just as irrelevant. The top-left point is the reason the rule is about recoverability rather than about the past: an asset you can still sell is live money, however long ago you bought it.
What became clearer
WHAT CLEARED #The principle is almost trivially simple — compare futures, and cancel anything the two futures share — and its simplicity is why the fallacy is so easy to name and so hard to obey. But the honest version has a second half. Sometimes the room is right to continue: reputation for finishing is a real asset, and past effort is evidence as well as expenditure. What separates rational persistence from escalation is whether anyone can say what future benefit the next pound buys, in terms a newcomer with no history would accept.
Where to go next
ONWARD #- Escalation of commitment: why the person who authorised a project defends it harder than their successor does.
- Option value: when paying to keep a bad project alive is really buying the right to decide later.
Key terms
TERMS #| Term | What it means |
|---|---|
| Sunk cost | an expenditure already made that cannot be recovered under any of the options being considered. |
| Sunk cost fallacy | letting such an expenditure influence a choice between futures it cannot distinguish. |
| Escalation of commitment | the organisational pattern of increasing investment in a failing course of action, often by the person responsible for starting it. |
| Loss aversion | the tendency to weigh a loss more heavily than an equivalent gain; a distinct claim, sometimes offered as an explanation of the above. |
Every term the collection defines is gathered in the glossary.