During the Second World War, the American military had a practical problem. Bombers were coming back from raids over Europe riddled with bullet holes, and adding armour plating to the aircraft would improve their survival odds. But armour is heavy, and heavy planes fly slower, carry less, and burn more fuel. The military could only afford to reinforce a few areas. So officers did the sensible thing: they examined the returning aircraft, mapped where the damage clustered, and prepared to armour those spots.
A statistician named Abraham Wald, working with a group of mathematicians assigned to military problems, pointed out that this was exactly backwards. The damage map showed where a plane could be hit and still make it home. The areas with fewer holes were not safer — they were the places where a hit meant the aircraft never returned to be counted. The armour belonged where the data was thin.
Wald's actual work was considerably more technical than the story usually told, and he was estimating aircraft vulnerability rather than literally sketching bullet holes on a diagram. But the core insight is real, and it has a name: survivorship bias. It is the error of drawing conclusions from a group that has already been filtered, while forgetting that the filter removed the most informative cases.
The Shape of the Error
Survivorship bias is not a failure of intelligence. It is a failure of availability. We reason with the examples in front of us, and the examples in front of us are systematically not random.
Consider how business advice is generated. A founder builds a company that becomes enormously valuable. Journalists interview them. They write a book. They give a talk about the habits and instincts that got them there — the decision to ignore the market research, the refusal to take outside money, the insistence on shipping before the product was ready.
Every one of those decisions was probably also made by hundreds of founders whose companies collapsed. Those founders do not get interviewed, because failure is not a story anyone buys. So the trait that actually separated the winner from the losers — which might have been timing, capital, a well-placed contact, or nothing more than luck — never enters the analysis. What we get instead is a list of behaviours that correlate with taking a big risk, presented as a list of behaviours that cause success.
Where It Shows Up in Ordinary Life
**Old buildings look better than new ones.** Walk through a historic district and it is easy to conclude that people used to build more beautifully. Some of that is true. But the ugly, cheap, poorly built structures of the same era were demolished decades ago. You are looking at a curated exhibition of the best of the past compared with all of the present.
**Old things seem more durable.** The cast-iron pan that lasted seventy years is proof of good manufacturing, people say. It is also true that the pans which cracked in year three went to the scrap heap and are not in anyone's kitchen to be admired.
**Investment strategies look reliable in hindsight.** Fund performance tables often quietly exclude funds that closed. A strategy tested against companies currently in an index has never been tested against the companies that were dropped from it for going bankrupt.
**Advice from people who quit their jobs to travel.** The ones for whom it worked out post beautiful essays about courage. The ones who came back broke and demoralised post nothing at all.
**Alternative medicine testimonials.** People whose condition improved write reviews. People who got worse, or died, do not.
Why the Brain Falls For It
Two mechanisms combine. The first is the availability heuristic — we judge how common something is by how easily examples come to mind, and success stories are amplified while failure is quietly buried. The second is our appetite for narrative. A story requires causation: this person did X, and therefore Y happened. "This person did X, and so did nine hundred others, and one of them got lucky" is not a story anyone can use.
There is also an uncomfortable social dimension. Survivorship bias flatters the successful, because it lets achievement be read purely as the product of character. And it is harsh on the unsuccessful for the same reason. Any framework that explains outcomes entirely through virtue will always find an audience among those who did well.
How to Actually Correct For It
You cannot eliminate survivorship bias, but you can systematically interrogate it. Three questions do most of the work.
**First: what would I need to see that I am not seeing?** Before accepting that a habit produces success, ask what proportion of people with that habit failed. If you cannot even estimate the denominator, you do not have evidence — you have an anecdote.
**Second: who is missing from this sample, and why?** Every dataset is the output of a selection process. Ask what that process removed. Customer reviews exclude people who never bought. Alumni surveys exclude dropouts. Employee satisfaction surveys exclude everyone who already quit.
**Third: seek out the failures directly.** This is the most useful and least practised habit. Interview the people whose businesses closed. Read the post-mortems of failed projects rather than the launch announcements of successful ones. In fields with genuine feedback loops — aviation safety, medicine, engineering — progress comes overwhelmingly from studying failures in detail, which is precisely why those fields have improved so much faster than fields that only study winners.
What It Does Not Mean
Survivorship bias is not an argument that success is entirely luck, or that expertise does not exist, or that you can learn nothing from people who did well. Skill is real and it shifts the odds. The bias is narrower than that: it says the *visible* difference between the successful and everyone else is not reliable evidence about *which* factors mattered, because you are only ever looking at the half of the picture that survived the filter.
The practical upshot is modesty about causation. When someone tells you what made them successful, they are giving you an honest report of what they remember doing. They are not giving you a controlled experiment, and they genuinely cannot — because the control group was never asked.
The Bottom Line
The returning bombers had holes in the wings and the fuselage. The engines looked fine. The reason the engines looked fine is that a hit to the engine meant the plane went into the North Sea. Whenever you are handed a pattern drawn from winners, ask what happened to the aircraft that never came back — and notice that the answer is almost always the part of the story nobody collected.
Be the first to share a thought.