What Is Survivorship Bias? (Definition and Examples)

Survivorship bias is the logical error that occurs when a selection process with attrition produces a sample containing only the survivors — and you draw conclusions about the full population from that filtered sample.

The bias gets its name from the literal sense: the survivors survived. They're visible. The ones who didn't survive dropped out of the sample. You can only see and study the survivors, but you draw conclusions as if you'd studied everyone.

Why the Bias Produces Systematic Errors

The survivors are not a random sample of the population that started. They're a selected subset — selected specifically by the process that eliminated everyone else. The characteristics they share may be the cause of their survival, or they may simply be correlated with other variables (timing, market conditions, early luck) that drove the selection.

The World War II version: Allied analysts studied bullet-hole patterns on returning bombers to add armor. Statistician Abraham Wald recognized: reinforce the areas where returning planes had no holes. The planes hit in those spots didn't return. The survivors showed only the wounds that were survivable, not the wounds that were fatal.

The business book version: Every success book written about high-performing companies studies companies that are currently high-performing. The initial cohort that tried the same strategies included many failures. The common traits of the surviving companies are confounded with the survival itself.

The financial version: Market indices quietly remove companies that go bankrupt or delist. The historical return series for "the market" excludes these companies retroactively. The survivorship-adjusted return is lower than the published number.

How to Correct for It

Before drawing any inference from a visible success, ask: - What was the initial population? How many started? - How many dropped out, and why? - Is the trait I'm observing in the survivors something that selected for their survival, or just something survivors happen to share?

The higher the attrition in the selection process, the more distorted the visible sample — and the more skeptical you should be of inferences drawn from the survivors alone.

For the full framework, read Fooled by Randomness: How Luck Masquerades as Skill.