TL;DR: In the late 1940s, the U.S. Air Force faced a practical design problem. Pilots were struggling with control and comfort issues, leading to accidents and reduced performance. The root cause was the assumption that a cockpit designed for the “average” pilot would fit most pilots. The data showed the opposite: when many body measurements matter at once, almost nobody is average on all of them.
The Myth of the “Average” Pilot
In the 1920s, the U.S. Air Force began collecting anthropometric data on its pilots to improve cockpit designs. Engineers focused on 10 key body dimensions—height, chest circumference, arm length, leg length, etc.—to define an “average” pilot. The assumption was simple: if a cockpit suited the average pilot, it would suit most pilots.
However, by the late 1940s, mounting evidence suggested this assumption was flawed. Pilots frequently reported discomfort, difficulty operating controls, and even injuries during emergencies. The Air Force recognized the need for a deeper investigation.
The Statistical Groundwork: Why Averages Fail
From a statistical perspective, each body measurement follows a normal distribution, especially with large sample sizes. The Air Force’s approach defined “average” as the middle 30% of values (between the 35th and 65th percentiles) for a single dimension. For example, if a pilot’s height fell within this range, they were considered “average” for height.
But what happens when multiple dimensions are considered simultaneously?
For independent variables:
\[ P(\text{within range for all dimensions}) = P(\text{within range for one dimension})^k \]
Here:
- \(P\) is the probability of being within the middle range for a single dimension (30%, or 0.30).
- \(k\) is the number of dimensions (10 in this case).
Thus: \[ P = 0.30^{10} \approx 0.0000059 \quad \text{or } 0.00059\% \]
This means the likelihood of a pilot being “average” across all 10 dimensions is virtually zero! Daniels confirmed this in his dataset: not a single pilot was “average” across all 10 dimensions.
Even when considering just three dimensions, fewer than 5% of pilots were “average” in all.
Breaking the Myth: Implications for Cockpit Design
Daniels’ findings exposed the flaws of designing for the “average.” Averages oversimplify human variability and exclude a majority of the population. Designing for the average, it turns out, effectively designs for no one.
Armed with this knowledge, the Air Force adopted a new approach: designing for diversity. This included:
- Adjustable seats to fit pilots of different heights.
- Customizable controls for varied arm lengths and reach.
- Flexible straps and harnesses to accommodate different chest circumferences.
These changes improved safety, performance, and pilot comfort.
The Curse of Dimensionality
The Curse of Dimensionality, a term introduced by mathematician Richard Bellman, refers to a suite of phenomena that arise when data is analyzed in spaces with many dimensions, phenomena that simply don’t occur in our everyday three-dimensional world. As Wikipedia notes, “it refers to various phenomena that arise when analyzing and organizing data in high-dimensional spaces that do not occur in low-dimensional settings.”
In the cockpit design example, considering multiple body measurements (height, arm length, etc.) creates a high-dimensional space where the notion of an “average” pilot quickly collapses. Even if a large percentage of individuals are “average” on one or two dimensions, that fraction becomes tiny when you look at all dimensions together. This explains why a cockpit based on average measurements serves almost no one. More broadly, whenever we add more variables—whether in product design, machine learning, or any data-driven field—the space expands exponentially, making intuitive ideas like “typical” or “close together” far less meaningful.
For more information, see the Wikipedia article on the Curse of Dimensionality.
Statistical Considerations: Multivariate Normality
While Daniels’ analysis assumed independence between dimensions, body measurements are often correlated. For example, taller individuals generally have longer arms and legs. Accounting for these correlations requires evaluating probabilities within a multivariate normal distribution, which involves the joint probability density function:
\[ P = \int \cdots \int_{R} f(x_1, x_2, \ldots, x_k) dx_1 dx_2 \ldots dx_k \]
Where \(f(x_1, x_2, \ldots, x_k)\) is the joint probability density function and \(R\) is the range of interest. Although complex, the conclusion remains the same: the likelihood of being “average” across multiple dimensions is extremely small.
Hands-On: Exploring the Concept with R
For those interested in a practical demonstration, here’s how to explore this concept using R and the webr package in Quarto:
This simulation assumes three correlated dimensions: height, chest circumference, and arm length. Each dimension is represented by a mean and variance, with the covariance matrix defining the relationships between dimensions.
Simulating Multivariate Normal Data
This plot demonstrates the correlation between height and chest circumference, showing a clustering effect that reflects real-world data patterns.
Calculating “Average” Probabilities
This code demonstrates how the probability of being “average” diminishes rapidly as more dimensions are considered. Despite accounting for correlation, the proportion of pilots who are “average” across all three dimensions remains small, typically around 3–5%. This is slightly higher than the independent case ( 0.3 ^ 3 * 100% = 2.7%) but still demonstrates the challenge of designing for the “average” person.
Conclusion
The U.S. Air Force’s experience with cockpit design reveals a fundamental truth: designing for an “average” that doesn’t exist serves no one. Whether in aviation, technology, or daily life, we must design for diversity and inclusivity. Daniels’ groundbreaking research remains a powerful reminder that true innovation begins when we embrace the full range of human variability.