Survivorship Bias
You can only study what is still around to be studied. When the failures quietly drop out, what remains tells a flattering lie. Two short steps: armour a bomber, then read a fund advert.
Armour the bomber
It's 1943. Every dot below is a bullet hole on a bomber that made it back from a mission. Armour is heavy, so you can only add it to one part of the plane. Where do you put it?
The same trick, in your inbox
A hundred investment funds start with $100 each. None has any skill: every year each one goes up or down at random. Any fund that falls below $80 loses its investors and quietly closes. Ten years on, an advert shows you the funds still open.
The missing data is the data.
Nothing here has any skill, yet the funds you get to see look brilliant, because the failures were removed before anyone looked. Before learning from the winners, ask what happened to everyone who started.
Worth knowing
Abraham Wald, 1943
Wald, a statistician in a wartime research group at Columbia, wrote the analysis of damage to returning bombers. The popular telling is simplified, but the core point is his: account for the planes that never came back.
Winners are easy to find
Dropout billionaires, old buildings that were 'built to last', customer surveys answered only by customers who stayed. Each is a sample that the failures had already left.
Your data has survivors too
A churn model trained only on current customers, a credit model that only sees loans that were approved: both learn from whoever made it through, and miss exactly the cases that matter most.
The bullet holes and the funds are simulated in your browser. Hits land evenly across the plane, and every fund is a random walk with no edge.
Try another puzzle →