The sunk cost fallacy is actually not always a fallacy.
Many of us might have have heard of the sunk cost fallacy—the fallacy where we allow past investments to influence our decisions. Social scientists, including economists and psychologists, tell us that honoring sunk costs is irrational. Thomas Kelly (2004) argues against this claim, which raises the question for any initial reader: how could the sunk cost fallacy not be a fallacy?1
Kelly is actually arguing against two claims which together he calls the “conventional wisdom” about sunk costs (61).
(1) is meant to set up the sting of (2). If honoring sunk costs were a rare phenomenon, there wouldn’t be much impact in saying that honoring sunk costs is irrational.
On (2), Kelly frames the claim as saying that giving any weight to sunk costs in our decision-making is irrational. In other words, (2) says that continuing some course of action or being more likely to continue some course of action because of past investments is irrational.2 On the conventional wisdom, we usually give weight to many different factors when we are making decisions, but sunk costs should not be one of those factors.
Kelly wants to make sure that we are targeting cases that are actually about honoring sunk costs.3 As he explains, imagine that you have bought an expensive ticket to see a theater performance, but you would actually prefer to stay home tonight to read a book.
If you decide to go see the theater performance because you want to avoid the regret of wasting the ticket, this is crucially different from going to see the performance because you have already spent so much money on the ticket. In the first case, you are giving weight to how much you don’t want to feel regret unlike the second case where you are giving weight to how much you previously invested.
But I think you would be right to object here that in the first case you would only feel regret if you were giving weight to your previous investment. As soon as you don’t give any weight to how much you spent on that ticket, you wouldn’t feel any regret at staying home.
A better example Kelly offers is from a famous study by Hal Arkes and Catherine Blumer where the authors propose that we continue some course of action we have previously invested in because we don’t want to appear wasteful to others.
Kelly points out that this hypothesis actually “competes” with the hypothesis that we continue some course of action we have previously invested in because we are honoring sunk costs (64). Unlike the earlier example, it is false that you don’t want to appear wasteful to others because you are honoring sunk costs. Even if you don’t honor sunk costs, you might continue a course of action you have previously invested in because you don’t want to appear wasteful.
The lesson here is that while the action may be the same the reason you perform this action may be different. This means we need to be careful when discussing cases of honoring sunk costs that the reason for performing an action is actually honoring sunk costs and not another reason.
Though not central to his argument, Kelly offers an illuminating clarification of the position of someone who believes honoring sunk costs is irrational.
Someone of this position can still believe that knowledge of sunk costs is useful in decision-making. Sunk costs may indicate whether a course of action is likely to succeed or fail. Kelly gives the example of someone deciding to become a medical doctor. How much this person has previously invested in becoming a doctor as well as how far she has come would be useful in deciding how likely it is that she will get into medical school, complete her residency, etc.
All that the defender of the sunk cost fallacy needs to concede then is that knowledge of sunk costs is valuable “only insofar as such knowledge enhances one’s ability to predict those future consequences” (71). In other words, the defender of the sunk cost fallacy maintains that we should use knowledge of sunk costs only to the extent that they help us predict the future.
To bolster the strength of his argument against such a position, Kelly asks us to consider a case where you know how all your actions will play out. In essence, you are to imagine that you can predict the future. Even in this case, Kelly argues that knowledge of sunk costs can be essential to deciding what to do, meaning it can be rational to honor sunk costs.
To show this, Kelly introduces the notion of a redemptive action. A redemptive action is one that prevents “the past efforts of others from having been in vain” (74). You might prefer a redemptive action out of a “respect for agents and their projects” so there could be good reasons for preferring redemptive actions (78). Thus, if you have a preference for performing redemptive actions, knowledge of sunk costs is useful for decision-making even when you can predict the future. Namely, it would help you decide which actions are redemptive.
For example, say there are two actions A and B and you have a preference for redemptive actions. Based on your predictions, B would have a slightly better outcome than A. But based on your knowledge of sunk, costs if A is a redemptive action and B is not, then this might actually tip the scale in favor of doing A even though solely in terms of outcomes B wins out.
It turns out then that knowledge of sunk costs have a use beyond just helping us predict the future. This means that honoring sunk costs can be rational when you have a preference for redemptive actions.
To complete his argument against defenders of the sunk cost fallacy, Kelly presents a dilemma for them. It follows from two ways of defining honoring sunk costs as mentioned in a previous note.
The first is a wide definition that almost automatically makes (1) of the conventional wisdom true. This particular one is from Arkes and Blumer: “a greater tendency to continue an endeavor once an investment in time, effort, or money has been made” (124). (1) is then true because this definition captures a broad phenomenon.
If a defender of the sunk cost fallacy goes for the first definition, then (2) of the conventional wisdom looks false. Under this definition, preferring redemptive actions counts as honoring sunk costs because you would be continuing an endeavor for the reason of preventing the past efforts of others from having been in vain. This means that (2) is false because honoring sunk costs is not always irrational. Performing redemptive actions can be rational.
The second is a narrow definition that tacks on the restriction that you do not have an “independent preference for finishing already-begun projects” (79-80). Under this definition, preferring redemptive actions would not count as honoring sunk costs because it would be an independent preference for finishing an already-begun project.
If a defender of the sunk cost fallacy goes for this second definition, then (2) is a lot more defensible since Kelly’s argument does not apply anymore. However, (1) starts to look false. Paradigmatic cases of honoring sunk costs don’t count as honoring sunk costs under this definition.
Take for example, the involvement of the United States in the Vietnam War. The argument that the U.S. should stay in Vietnam because of the lives already lost is taken to commit the fallacy of honoring sunk costs. But under this second definition, the preference for redeeming these lost lives is an independent preference so the Vietnam War case is not a case of honoring sunk costs. The defender of the sunk cost fallacy cannot go for this definition too.
Thus, the defender of the sunk cost fallacy is faced with a dilemma. On the one hand, honoring sunk costs wouldn’t always be irrational and on the other, many cases we consider to be honoring sunk costs turn out not to be genuine cases.
You might object that this entire time Kelly has never argued for the claim that performing or preferring redemptive actions is rational. It is strange that Kelly seems to merely assert that it is rational.
There is a good reason that Kelly does not defend the rationality of performing or preferring redemptive actions. Coming back to the Vietnam War case, you might argue that it would be irrational to prefer redeeming those lost lives at the cost of many more future lives. But, as Kelly points out, once we have come to this point, the issue is not about whether someone is honoring sunk costs but that someone is valuing the redemption of lost lives over future lives. This is not a criticism on the grounds that someone is committing the sunk costs fallacy so Kelly can go either way on this issue while maintaining that his argument against the conventional wisdom about sunk costs still works.
All Kelly needs is to show that honoring sunk costs is not always irrational. This merely requires the claim that performing or preferring redemptive actions can be rational and not that it always is or is so in some particular case.
You may still object that Kelly doesn’t argue for why honoring sunk costs can be rational. Kelly writes that “no particular preference can be intrinsically irrational” and that even if some preferences were irrational “it is unclear why preferences for redemption should be included” (74).
Kelly, Thomas. “Sunk Costs, Rationality, and Acting for the Sake of the Past.” Noûs 38, no. 1 (2004): 60-85. ↩
This definition of honoring sunk costs is a “wide definition” and a “narrow definition” would add that honoring sunk costs also means that we do not have an “independent preference for finishing already-begun projects” (79-80). ↩
Kelly criticizes psychologists for not doing this, pointing out that much of the literature is “methodologically flawed” due to this mistake (64). ↩