Survivorship Bias

The classic example

WWII: aircraft returning from missions had bullet holes mostly in wings and fuselage, not the engine or cockpit. The naive conclusion: armour the wings. Abraham Wald inverted: aircraft hit in the engine or cockpit didn’t return; armour those. Survivorship bias misled the visible data.

Where it matters in longevity

• Centenarian studies: people who reach 100 are a tiny survivor pool. Their lifestyles look "longevity-protective" partly because comparable people doing the same things died younger. Lifestyle attributions from centenarian populations should be tempered.
• Drug-trial completers: analysing only those who finished a trial ignores those who dropped out due to side effects. Per-protocol vs. intention-to-treat analyses can diverge sharply.
• "Super-agers": studying cognitively-intact 80-year-olds isolates protective factors but ignores survival selection (others died first).
• Self-experimenter cohorts (e.g. Bryan Johnson’s public protocol): a single person who feels great after a regimen says nothing about its average effect.

Why this matters operationally

When evaluating any longevity intervention based on:

• "People who do X live longer" — ask: how did the comparison group die first?
• "Look at this centenarian’s habits" — ask: how many people did the same habits and died younger?
• "We followed [exceptional individual] for years and they thrived" — ask: what about all the others who tried the same thing?

Antidotes

• Randomised controlled trials.
• Intention-to-treat analysis.
• Mendelian randomization for causal questions.
• Cohort studies that follow people from younger ages.

Related entries

Confounding, Mendelian randomization, Centenarians, Bradford-Hill criteria.