The sunk cost fallacy

You paid for the ticket, the film is bad, and you stay to the end to get your money's worth. The project is a year in and plainly not working, and you argue for one more quarter because of everything already invested. From inside it feels like loyalty or persistence; the tell is that the argument is about the past. Arkes and Blumer defined the sunk cost effect as a greater tendency to continue an endeavour once an investment of money, effort or time has been made. Economically the past spending is gone whatever you do, so only future costs and benefits should matter. The ticket you paid for and the ticket you were given should lead to the same decision about driving through the snowstorm.

Arkes and Blumer's explanation is that people do not want to appear wasteful, to themselves as much as to others, and that stopping would mark the earlier spending as a loss. That ties it to loss aversion: quitting turns a paper loss into a realised one. The effect shows up wherever the past is visible and the future uncertain. Governments finish roads they would not start today; companies keep products alive to avoid writing off the development; engineers keep the system they built rather than adopt the one they would choose fresh. Arkes and Blumer tested it in the field by selling the same theatre season at three random prices and counting who turned up.

How large the effect is outside a questionnaire is contested. Friedman and colleagues built a laboratory game in 2007 designed to produce the effect and varied eleven features of it; they report a surprisingly small sunk cost effect that was largely insensitive to their manipulations. The Many Labs project in 2014 replicated a scenario version across 36 samples and found an effect, but a small one, with Cohen's d around 0.3. And persisting is sometimes right: the past spending may have bought information, options or reputation that count as future benefits. So the test is not 'have we spent a lot' but 'would we start this today, at today's prices, knowing what we now know'.