Summary
Sometimes a decision that seemed sensible ex ante proves to be not worthwhile, and standard rationalist advice is to exit the investment, ignoring any costs you've already sunk into it. However, sometimes sticking with the original plan can be a way to punish your bad judgment and so reduce the chances of similar errors in the future. I discuss a few examples to illustrate.
Contents
Introductory example
Veronica the vegan is going out to lunch at a new restaurant. She is trying to avoid meals containing meat or eggs, but she isn't familiar with the words in the menu.
She considers asking the waiting staff for advice on which options are vegan, but she's shy and ultimately decides to just pick something that looks like it has no animal products. She places her order and waits for the food to arrive. When it comes 25 minutes later, she's horrified to see that it contains a large pile of about 15 prawn that suffered through death to produce this meal.
Veronica's stomach is growling, and she doesn't want to wait another 25 minutes to order a different dish. She would also feel embarrassed telling the waiting staff that she didn't mean to order her current meal. Moreover, she reasons, the harm done by ordering the prawn dish is a sunk cost: Nothing she does now can change it, because obviously the restaurant won't take it back, and no one else would be willing to eat it. Ordering something else would only waste more money, more time, and more willpower. So Veronica decides to eat the prawn even though she feels a little bad for misplacing her order.
Evaluation
Was Veronica's decision to eat the prawn rational? In some sense it was, because she made the best of her current situation. On the other hand, maybe it was a little too easy for her to accept the cost of ordering the prawn. Perhaps she should have taken additional care to ask the waiting staff beforehand what the meal would contain. If Veronica had adopted a more absolutist stance against eating animal products, then she would have suffered through the embarrassment and hunger of ordering a second meal, and this might have been enough to incentivize her to overcome her shyness and ask for a vegan meal from the beginning, or at least make it more likely she does so next time.
When you treat past costs as sunk, it can be easy to pretend that they weren't a big deal and so avoid the psychological disappointment that comes with incurring those costs. This may mean that you don't incorporate those costs into your decisions to the appropriate degree. This is especially likely in cases where the costs fall on other parties, because then you don't viscerally experience the harm, so your brain may not adequately adjust its behavior in response to what has been learned.
Punishment
In other words, feeling the sunk-cost fallacy can be a form of self-punishment to ensure that you pay the consequences of your bad decisions. If you order a ticket to a movie that you discover you dislike, if you force yourself to stay, you're teaching yourself via self-punishment, "Don't be so stupid next time! Read the reviews and watch a preview before going to a movie."
It's true that you already paid some monetary and time cost to go to the movie, and maybe this is enough punishment already. Maybe you don't need to lump on additional psychological annoyance. But, as with eating prawn, wasting money on movie tickets imposes an altruistic cost that you may not fully internalize unless you experience it in other ways, and sitting through a bad movie can be one such way.
Now, I'm not suggesting we necessarily should feel the sunk-cost fallacy in order to give ourselves extra punishment. Sometimes that's not necessary or healthy. Moreover, we needn't punish ourselves by sitting through a bad movie; we could instead go home early and do some unpleasant chore as punishment, and this would be more productive on balance. However, I am pointing out that there can be a legitimate psychological rationale behind feeling the sunk-cost fallacy that can sometimes be overlooked in the naive rationalist approaches to the issue.
This idea is not unusual. In fact, there's a mundane example where we neglect sunk costs all the time because of the longer-term reinforcement effects: Namely, the case of criminal punishment. Once a crime has been committed, the cost is sunk, and the damage may not be repairable. So why impose a further cost by punishing the offender, especially in an expensive, unproductive manner like jail time? Well, sometimes such punishment is not a good idea, but when it is, the reasoning has to do with pre-commitments to punishment so as to maintain the right incentives. If, whenever someone committed a crime, we took it to be "in the past" and "sunk," there would be less incentive against further crimes. Similarly, by treating the cost of ordering prawn as sunk, Veronica reduces her incentive to take the right actions in future cases.
Sunk-cost dilemma
The "sunk-cost dilemma" describes a situation in which someone continues in a venture that ends up having overall negative value, even though at each step, it looked like continuing would have positive value ignoring sunk costs. This article describes a concrete example in the case of a business where the management consistently underestimated the total costs that would be incurred. One way to approach the situation is just to say that ignoring sunk costs is rational. But another perspective is to realize that if you consistently incur more costs than you anticipated, maybe there's a problem with your projections. Maybe you should take a step back, feel the losses that are happening at a deeper level, and reassess your predictions rather than blithely ignoring the costs because they don't matter anymore to future decisions.
Other cases where sunk costs matter
In this article I approached the sunk-cost fallacy as a way to increase the punishment of bad decisions, but there can be other cases when the fallacy is not necessarily a fallacy. Another is described in "Is the Sunk Cost Fallacy Actually Smart Business?"