Pretty much the most interesting blog on the Internet.— Prof. Steven Landsburg

Once you get past the title, and the subtitle, and the equations, and the foreign quotes, and the computer code, and the various hapax legomena, a solid 50% English content!—The Proprietor

## Friday, September 18, 2015

### In Defense of the Sunk-Cost Heuristic

One of the perennial bugaboos of introduction to economics courses1 is the sunk-cost fallacy. The fallacy goes generally like this:

We have invested an amount $$a$$ in a project. Now we are faced with the following choice: Abandon the project and recover nothing; or, complete it at an additional cost $$b$$ and recover $$c$$. Now, $$b$$ may be larger than $$c$$, but if we do not complete the project, we lose all of $$a$$. In order to save $$a$$, we better press on and complete the project.

Stated in such terms, this is obviously a fallacy and economists rightly delight in pointing it out. $$a$$ is already lost, regardless of the choice and is therefore irrelevant to it. All that matters is whether $$b$$ or $$c$$ is larger.

But that is only sometimes, not always, the situation in which sunk-cost-fallacy-like reasoning occurs. A common, similar, but subtly different, situation is:

We are faced with the option of cancelling a project into which we have already invested $$a$$. When we made the decision to launch the project, we carefully thought about the circumstances and concluded that the full cost $$a+b$$ would be less than the full benefit $$c$$. It is possible that the circumstances have changed and that is no longer true. But if $$a$$ is large and $$b$$ is small, circumstances would have had to change dramatically for $$c$$ not only to fallen below $$a+b$$, but even below $$b$$ alone. Unless we see such a dramatic change, it is probably wise to proceed without too much further analysis.

This is a perfectly good heuristic. Like any heuristic, it can fail. But as our mental capacity is limited, we often must and do rely on heuristics to make decisions.

1 Or so one is told.