In a classic experiment, psychologists Arkes and Blumer, randomized theater ticket prices to test for the existence of a sunk-cost fallacy.  Patrons who bought season tickets at the theater box office were randomly given discounts, some large some small.  At the end of the season the researchers counted how often the different groups actually used their tickets.  Consistent with a sunk-cost fallacy, those who paid the higher price were more likely to use the tickets.

A problem with that experiment is that it was potentially confounded with selection effects.  Patrons with higher values would be more likely to purchase when the discount was small and they would also be more likely to attend the plays.  Now a new paper by Ashraf, Berry, and Shapiro uses an additional control to separate out these two effects.

Households in Zambia were offered a water disinfectant at a randomly determined price.  If the price was accepted, then the experimenters randomly offered an additional discount.  With these two treatment dimensions it is possible to determine which of the two prices affects subsequent use of the product.  They find that all of the variation in usage is explained by the initial offer price.  That is, the subjects revealed willingness to pay was the only detrminant of usage and not the actual payment.

This is the cleanest field experiment to date on the effect of past sunk costs on later valuations and it overturns a widely cited finding.  On the other hand, Sandeep and I have a lab experiment which tests for sunk cost effects on the willingness to incur subsequent, unexpected, cost increases.  We show evidence of mental accounting:  subjects act as if all costs, even those that are sunk, are relevant at each decision-making stage.  This is the opposite effect found by Arkes and Blumer.

(Dunce cap doff:  Scott Ogawa)