Moral Luck and Outcome Bias

The idea in plain words

Two product teams make the same bet with the same information. One launches the week a competitor stumbles; the other launches the week a competitor ships something better. A year later the first team is promoted and the second is reorganised. Nobody in the building would say the first team thought more clearly. Yet that is how the decision gets remembered: as a good decision, because it worked.

Outcome bias is the name psychologists give to this habit of grading a choice by its result. Moral luck is the much older philosophical puzzle underneath it: we say we judge people only for what was in their control, but we keep praising and blaming them for things that were not.

The two ideas belong together. The psychology shows that the habit is real and measurable. The philosophy asks whether it is merely a mistake, or whether it reveals something we cannot give up.

Where it comes from

The philosophical root goes back to Kant. In the first section of the Groundwork (1785), he argued that a good will is good in itself, whatever it achieves. If bad fortune prevented a well-meant action from producing anything at all, its worth would still shine, as he put it, like a jewel. On this view, results are morally irrelevant. Only the quality of the willing counts.

In 1976 Bernard Williams and Thomas Nagel, writing paired papers for the same symposium, showed how hard it is to live by that rule. Williams told the story of a painter, loosely modelled on Gauguin, who abandons his family to pursue his art. Whether we think he was justified, Williams argued, depends in part on whether he succeeds as a painter, something he could not know or control when he left. Nagel gave a plainer example: two equally careless drivers, one of whom happens to hit a child who runs into the road. We blame the second far more, although the difference between them is entirely luck.

Nagel distinguished several kinds of luck. Resultant luck concerns how things turn out. Circumstantial luck concerns the situations we happen to face: a manager who never meets a crisis is never tested by one. Constitutive luck concerns the temperament and abilities we happen to have. His unsettling conclusion was that if we strip out everything that depends on luck, very little is left to judge.

The psychological evidence arrived in 1988. Jonathan Baron and John Hershey gave people descriptions of medical and financial decisions in which the information available to the decision-maker was held constant and only the outcome varied. Subjects rated the very same decision as better, and the decision-maker as more competent, when the outcome was good. They did this even when they were told explicitly that the outcome should not count.

What it does to the mind

Outcome bias feels like common sense because outcomes are vivid and processes are invisible. You can see a product that failed. You cannot see the probability distribution it was drawn from. The mind substitutes the question it can answer (did it work?) for the one it cannot easily answer (was it a reasonable bet given what was known?). Daniel Kahneman called this kind of swap attribute substitution, and it happens without our noticing.

There is also a social reason. Judging by results is cheap, public and hard to argue with. Judging by process requires reconstructing what people knew, which is laborious and invites dispute. Organisations under pressure take the cheap route, and people inside them learn the lesson: protect yourself from bad outcomes, even at the cost of good decisions. That is how outcome bias produces timidity. If a bold bet that fails is punished and a cautious bet that fails is forgiven, people stop making bold bets.

Where it shows up

This site’s own decision boxes carry a verdict, Paid off, Mixed, Backfired or Too early, which describes outcomes. That is useful, but it is exactly the judgement this idea warns about. For that reason the newer sections add a second judgement where the evidence allows: a process grade that asks how good the decision was given what could be known then. The two often disagree, and the disagreement is where the learning is.

Nokia’s 2011 choice of Windows Phone, discussed in the Burning Platform memo, is a hard case. It backfired. But the reasoning, that joining Android would make Nokia one manufacturer among many, was serious, and the alternatives carried their own large risks. Netflix’s Qwikster runs the other way: the strategic reading of the DVD business was right, while the way the change was made was poor. Calling either simply a “bad decision” loses most of what there is to learn. The blind cases on the Decide page are built to separate the two judgements.

What it asks of someone deciding

Write down, before you decide, what you know, what you are assuming, and what would change your mind. This is the only reliable defence against rewriting the record afterwards, because it gives your later self something to compare with.

When reviewing a decision, separate four cases: good process and good outcome (deserved success), good process and bad outcome (bad luck, keep the process), bad process and good outcome (dumb luck, fix the process before it fails), bad process and bad outcome (deserved failure). Most reviews only notice the first and last.

Williams’s point should keep this honest. We cannot fully detach ourselves from outcomes, and perhaps we should not. A leader who bets the company and loses has lost something real for many people, whatever the quality of the reasoning. The aim is to give results their proper weight as evidence, while refusing to let them be the whole verdict.