During World War II, the Statistical Research Group—a team of mathematicians solving problems for the military—operated at Columbia University. According to a popular version of the story, they were approached with a question about aircraft armor. The military studied bombers that returned from missions and marked where the most bullet holes were: the wings, the fuselage, the tail. It seemed logical to reinforce the armor precisely there.
Mathematician Abraham Wald suggested the opposite. The aircraft the military examined were the ones that made it back. The bullet holes on them showed areas where the plane could take hits and survive. And the areas without holes—the engines, the cockpit—were not the spots where no one shot. They were the ones where being hit meant the plane did not return and never made it into the sample. Armor needed to be placed where there were no holes.
The story itself is notably simplified in popular retellings, but its core idea is exact. If a sample is pre-filtered by the result, it shows not what causes the result, but what is compatible with it. This is called survivorship bias.
In HR, this bias is more common than it seems, because almost all the people data a company works with is data about survivors.
The successful employee profile. A company takes the top performers in a role, looks for their shared traits, and builds hiring requirements based on them. But those same traits could just as easily be present in those who were hired and failed—they are simply absent from the analysis. The profile describes the best, but does not explain why they are the best.
Engagement survey. The most dissatisfied employees have already left and do not participate in the survey. Scores go up, and it looks like an improvement—even though it is the composition of the respondents that changed, not the conditions.
Hiring sources. A manager is convinced that the best employees come from specific universities or companies. But they only see those whom they hired. Candidates who were screened out based on a formal criterion and succeeded elsewhere do not enter their experience.
Others' practices. Companies copy the approaches of successful competitors without seeing the dozens of companies that did the exact same thing and shut down.
What helps. Include in the analysis those who did not make it: those who resigned, failed probation, or were rejected candidates, if there is a way to find out how their future turned out. Track cohorts from the moment of hire, rather than looking only at those working right now. Compare the profile of strong performers not with a void, but with the profile of weak ones: a trait only matters if it distinguishes between them. And before drawing any conclusion about what works, ask Wald's question: what data is missing here because its carriers did not survive to be analyzed?
The best employees tell a lot. But the most important part of the story is usually known by those who are no longer in the room.