The Much-Maligned Sunk Cost Fallacy

I recently came across a blog post criticizing the sunk cost fallacy for often being invoked fallaciously itself. The main argument seems to be this: Many people who learn about the fallacy are subsequently tempted to use it as justification to give up on long-running, costly projects, before incurring some huge payoff that would have made it worth it to stick with the project. Unfortunately, this hinges on a misunderstanding of the fallacy. Since I think it is a common/tempting one, I figured it was worth outlining what it got wrong, and how to reason about sunk costs rationally instead.

The post lists a few large-scale projects that repeatedly faced pressure to be abandoned but then worked out nevertheless. Examples include Elon Musk’s entrepreneurial endeavors, the James Webb Space Telescope, or a hypothetical stock investment that went down for a while before finally soaring to previously unseen heights. In all of these cases, the reasoning goes, people could have gotten burned and missed out on a huge payoff, had they naively appealed to the sunk cost fallacy and cut their losses because “you can’t recover the sunk costs by toughing it out”. Another illustrative example that wasn’t in the post but comes to mind is the urge to abandon your studies and drop out of university if you hate it, even shortly before getting your degree. Wouldn’t it be wrong to think that you shouldn’t give up just because you had already invested semesters worth of your time, because that time won’t be “recovered” if you do stick around until you graduate?¹

All these cases, however, hinge on a limited understanding of the fallacy that only looks at one part of the equation, while failing to apply it rigorously to the whole cost/benefit calculation. I am not entirely sure why this is so common. One possible reason could be that the classic examples of the fallacy are usually quite simple and extremely one-sided. One of them is that you hate a movie or book and continue spending your time on it instead of doing something you’d enjoy more, just because you’ve already spent some money and time on buying and consuming it. But in this case, there is indeed no potential payoff around the corner: All you get for keeping at it is more time of “negative enjoyment” (plus the opportunity cost of missing out on the “positive enjoyment” you could get from something else).

But to get the full picture in more complicated cases, it is perhaps easier to re-state what the sunk cost fallacy is trying to tell you in slightly different terms. For anything you are (and have been) doing at a given point in time, you face a decision between continuing to do it and stopping. And if you want to evaluate the costs and benefits of either of those two options to determine which one you should pick, you should only look forward in time. If you want to tally up the costs, you should only consider the costs you will incur in the future if you choose that option. Similarly, if you look at the benefits of the option, you should only consider the benefits you will get from it in the future.

Looked at like this, the fallacy, as classically stated, would be to confuse past costs as future benefits — if I stick with this project and throw good money after bad, the bad money I’ve spent on it in the past will somehow become more “worth it” and give me some positive utility in my accounting. You could also turn this around, then the fallacy would be assuming that abandoning the project right now would “cost” you what you had previously invested into it, and sticking with it would leave you roughly at zero (when, in fact, these costs have already been “sunk”).

But it turns out that if you don’t fall for this and apply the correct reasoning, many of the examples meant to illustrate how learning about the sunk cost fallacy could lead you astray are actually reversed! Take the case of finishing your studies: The correct accounting will tell you that you should absolutely stick it out for one more semester. From the point of view of having almost graduated, the degree becomes an absolute bargain for you! For just one semester of your time, you get all of the benefits that a full degree offers over having completed most of the coursework but not having graduated.²

The obvious caveat here is that for any large and complicated undertaking like the ones mentioned at the beginning, the remaining costs that lie ahead, and the benefits you get from completing the project (be that to the next useful milestone or all the way to the end), are extremely hard to estimate accurately. The probabilities of succeeding are even more so. In many cases, that’s precisely what gets you into the mess of having to reason about sunk costs in the first place — if you knew all of this perfectly at the outset, you wouldn’t have to re-evaluate the project in the first place. But like with most things we can’t fully know or predict, we can still learn a lot and do better by not throwing our hands up and going for the naive approach, trying instead to be a bit more explicit about our assumptions and the uncertainties involved, and a bit more rational in how we think about them and what kinds of conclusions we draw from them.

So looking at all this, the notion of “sunk costs” shouldn’t give you any sort of general bias for or against sticking with projects at all, in either direction. It is merely a very specific point about how to do rational cost/benefit accounting for this kind of decision. As such, it is a very valid point that can easily be missed if you’ve never encountered it! But once you apply it the right way, the difficulty in deciding whether or not to stick with something is often more about uncertain future outcomes than thinking about past costs the right vs. wrong way.

1: Any similarities to the author’s current situation are purely coincidental ;-)

2: I couldn’t find the correct source for this again, but IIRC earnings data shows that a pretty substantial portion of the higher wages you get with a degree vs. without one only happens when you get the degree, and you can’t get 95% of the benefit by doing 95% of the work but not getting the degree, etc. Learning outcomes are another story, of course.