#143 1950 · US Air Force (Lt. Gilbert S. Daniels) · Aviation / ergonomics
The Air Force found zero pilots were actually 'average,' and stopped designing cockpits for one
the problem
Cockpits built to fit 'the average pilot' fit almost no one
background
US military cockpits were designed around the physical measurements of the 'average pilot,' a standard last updated from a 1926 survey and used ever since to fix seat position, stick reach, pedal distance and instrument placement. Through the late 1940s, accident rates were high enough — over 20 destroyed aircraft per 100,000 flying hours — that poor physical fit between pilot and cockpit was suspected as a real contributing factor alongside pilot error and mechanical failure, but no one had actually checked whether the pilots flying these planes matched the average the cockpits were built for.
The standard response to a fit complaint was to adjust the average — remeasure the pilot population and shift the target dimensions closer to whatever the new numbers showed. That approach assumed an average pilot existed somewhere close to the middle of the distribution on every dimension at once; nobody had tested whether that assumption itself was true before building an entire generation of cockpits around it.
what everyone would do
Remeasure the pilot population and update the average the cockpit is built around — the standard response to any fit complaint, since it treats the problem as a stale or slightly wrong number rather than questioning whether an 'average pilot' who fits the cockpit design on every dimension at once exists at all. It keeps producing a single fixed target to design around, just a recalculated one, leaving the underlying one-size design philosophy completely intact.
what they saw
Daniels saw that the real question wasn't whether the average was measured correctly, it was whether any real pilot actually matched the average across all ten relevant dimensions simultaneously, rather than just on any one dimension in isolation. A population can have a perfectly well-defined average height, chest size, and arm length while containing almost nobody who is average-height AND average-chested AND average-armed at the same time — which meant the entire premise of designing a single cockpit around 'the average pilot' was built on a body that didn't exist.
the move
In 1950, Lieutenant Gilbert S. Daniels measured 4,063 Air Force pilots across the ten body dimensions most relevant to cockpit fit — height, chest circumference, arm length and similar measures — and checked how many pilots fell within a normal range of the average on all ten simultaneously, rather than just on any single dimension. He found essentially none did.
why it works
Measuring 4,063 pilots across ten dimensions simultaneously, rather than checking each dimension separately, revealed that the population's variation compounds: even if most pilots are close to average on any single measurement, the odds of any one pilot being close to average on all ten at once shrink toward zero as more dimensions are added, because each additional dimension is another chance to fall outside the acceptable range. This reframed the fix from 'find a better average' to 'stop assuming a representative individual exists' — the only design that fits a population with no representative member is one that adjusts to whoever is in the seat, which is why mandating adjustable seats, pedals and straps closed the fit problem in a way no amount of average-recalculating ever could, and mishap rates fell from roughly 23.6 to 4.3 per 100,000 flying hours over the following two decades.
the payoff
The finding that no pilot was actually 'average' across the dimensions that mattered discredited the practice of designing a cockpit around a single average body, and the Air Force mandated that manufacturers build adjustable seats, pedals and harness straps so the cockpit could be fitted to each individual pilot rather than the reverse. Aircraft mishap rates fell sharply over the following two decades, from roughly 23.6 to 4.3 per 100,000 flying hours by the late 1960s, alongside broader safety and training improvements.
where it breaks
The method only produces a decisive result when the thing being designed depends on multiple independent dimensions at once — a system with only one relevant dimension really can have a well-fitting average, so checking for a nonexistent 'multidimensional average pilot' problem where none exists just adds cost without benefit. It also depends on adjustability actually being feasible to engineer into the product; a cockpit could be re-tooled with movable seats and pedals, but not every system can be made adjustable at reasonable cost, in which case designing for a distribution instead of an average may require offering discrete size classes rather than continuous adjustment. And the underlying statistical trap can resurface anywhere a team quietly assumes a 'typical user' exists without ever checking whether any real user actually matches that profile across every dimension the design depends on.
what came after
Daniels's study is a foundational case in ergonomics and human-factors design, routinely cited as the origin of adjustable seating, straps and controls now standard in vehicles, aircraft and equipment of all kinds, and it remains a standard teaching example — well beyond aviation — of the statistical trap in designing around an average rather than a distribution.
references
- [1]How the US Air Force Ditched the "Average" and Saved LivesMann Howie, 2022mannhowie.com
- [2]Lt. Gilbert S. Daniels and the myth of averagesLessWrong, 2023lesswrong.com