Why looking only at successes leads to disastrous conclusions
During World War II, researchers analyzed returning bombers to see where armor should be added. They noticed heavy damage on the wings and fuselage, so they proposed reinforcing those areas. But statistician Abraham Wald realized this was a mistake: they were only looking at the planes that survived. The armor actually belonged on the engines and cockpit—the areas where shot-down planes had been hit, causing them to crash.
The Missing Bullet Holes
During the Second World War, military planners wanted to improve the survival rates of bombers flying dangerous combat missions. Armor plating added weight and reduced speed and fuel efficiency, meaning it could only be placed on the most critical sections of the aircraft. To determine where this protection was most urgently needed, researchers cataloged the bullet holes and shrapnel damage on bombers that successfully returned from combat. The inspection teams found that damage was heavily concentrated along the fuselage, wings, and tail sections. The engines and cockpit areas, by contrast, showed remarkably few hits. The initial, intuitive recommendation seemed straightforward: reinforce the wings and fuselage, where the greatest concentration of enemy fire had been absorbed.
The statistician Abraham Wald, working with the Statistical Research Group at Columbia University, recognized a fundamental flaw in this reasoning. The military was only inspecting the airplanes that had survived their missions and made it back to base. Wald realized that enemy fire was likely distributed relatively evenly across the aircraft. If returning planes consistently showed damage to their wings and fuselage, it meant that damage to those areas was survivable. The areas with few or no bullet holes—the engines, cockpit, and vulnerable fuel systems—were missing from the sample precisely because hits in those spots caused the aircraft to crash and be lost. The armor, Wald argued, should be placed where the returning planes showed no damage at all.
How the Filter Warps Reality
The error Wald corrected is known as survivorship bias, a specific form of selection bias. It occurs when a dataset includes only the individuals, objects, or ideas that have passed through a selective filter, while the failures are excluded from observation. Because the failures are removed from view, any conclusions drawn solely from the survivors will misrepresent reality. The human brain naturally focuses on visible evidence and struggles to account for what has vanished, treating the available sample as if it represents the entire original population.
This cognitive blind spot leads people to mistake correlation for causation. When analyzing a group of survivors, observers search for common traits among them and assume those traits are responsible for their success. However, without analyzing the failed population, it is impossible to know whether those same traits were equally present—or even more prevalent—among those who did not survive. If both the successes and the failures shared the exact same characteristics, those traits cannot explain the difference in outcome.
The Distorted Mirror of Finance
Survivorship bias heavily distorts financial data and investment analysis. When researchers or marketing teams evaluate the long-term performance of mutual funds, hedge funds, or investment strategies, they often examine only the funds currently active in the market. Funds that perform poorly, incur heavy losses, or suffer major reputational damage are routinely closed, liquidated, or quietly merged into larger, more stable funds. As a result, poor historical performances are removed from the database.
When an analyst measures the historical average return of existing funds over ten or twenty years, the resulting figure is artificially high. The sample contains only the winners that survived the market's attrition, while the failed funds that dragged down the true average have been erased. An investor reviewing this curated history receives a misleadingly optimistic picture of the likelihood of sustained high returns, mistaking the exceptional endurance of a few surviving funds for an industry-wide norm.
The Playbook of the Victorious
The modern business world is saturated with advice derived from survivorship bias. Popular business literature frequently analyzes mega-successful companies or visionary founders, compiling lists of their habits, management philosophies, and personal quirks. Readers are advised to drop out of university, embrace extreme risk, sleep four hours a night, or maintain unyielding confidence, because these traits are associated with famous billionaires.
What these narratives routinely ignore are the vast numbers of entrepreneurs who dropped out of school, took reckless gambles, and worked exhausting hours only to watch their companies collapse into bankruptcy. Because failed ventures do not write memoirs, deliver keynote speeches, or produce case studies, their experiences remain invisible. Mimicking the habits of successful survivors does not guarantee success if those exact same habits were also shared by thousands of forgotten failures.
High Falls and Vanishing Patients
A classic medical demonstration of survivorship bias arose in veterinary studies analyzing cats that fell from multi-story apartment buildings. Researchers noticed an unusual pattern in veterinary hospital records: cats that fell from two to six stories suffered severe injuries that worsened with height, but cats that fell from higher floors, such as seven to thirty-two stories, appeared to have less severe injuries. Some theorized that after reaching terminal velocity, cats relaxed, spread their bodies, and absorbed the landing impact more efficiently.
The simpler and more probable explanation lay in who brought the cats to the clinic. Cats that fell from extreme heights and died on impact were far less likely to be taken to a veterinary hospital by their owners. The hospital records only documented the surviving cats that required medical attention, creating an artificial statistical dip in injury severity at higher falls. A similar dynamic occurs in clinical drug trials when patients suffering from severe adverse side effects drop out of a study early; if researchers only measure the health metrics of the patients who completed the full trial, the medication can appear far safer and more effective than it actually is.
The Golden Age Fallacy in Culture
Survivorship bias also distorts our perception of history, architecture, and art. It is common to hear claims that ancient builders, classical composers, or mid-century manufacturers possessed superior craftsmanship compared to modern creators, because ancient cathedrals, centuries-old symphonies, and vintage appliances still endure today. People look at the Pantheon or listen to Mozart and conclude that previous eras produced universally superior work.
In reality, time acts as a massive selective filter. The vast majority of buildings constructed centuries ago were shoddy, unstable, and quickly crumbled or burned down; only the most exceptionally well-engineered and continuously maintained structures survived into the modern era. Similarly, countless mediocre musical compositions, dreadful plays, and poorly written books from past centuries were abandoned and forgotten. We compare the curated masterpiece survivors of the past against the unselected, raw totality of the present, creating an illusion of cultural decline.
Key takeaways
•Survivorship bias happens when conclusions are drawn exclusively from visible survivors, ignoring the invisible data of those that failed.
•Abraham Wald identified this error by showing that military aircraft armor should be placed where returning planes lacked damage, because hits there had caused unobserved crashes.
•In finance and business, tracking only existing funds or successful founders creates an artificially inflated picture of performance and misidentifies the causes of success.
•Physical structures, art, and medical studies are frequently misinterpreted when the poorly made, forgotten, or deceased subjects are omitted from the sample.