If a design team sets each of four seat dimensions to the 5th to 95th percentile range, each dimension accommodates 90 percent of users. What share of users meets all four at once? Many design teams assume the answer is close to 90 percent. It can be much lower, and the only way to know is to check the same people against all four criteria.
The question is older than most design standards. As Todd Rose recounts in The End of Average, Gilbert S. Daniels examined it for the US Air Force in a study published in 1952. He defined “average” as the middle 30 percent of the range on each of ten body dimensions relevant to cockpit design. Of the 4,063 U.S. Air Force personnel in Daniels’ sample, none fell within the study’s average range on all ten measurements. For a cockpit designer, that raises a practical question: can the same pilot reach the controls, see over the panel, and sit with enough clearance?
An earlier article on percentile models and 3D personas explains which body models to choose. This article covers the next step: how far a team can go in drawing conclusions about accommodation from those models.
What Percentiles Actually Describe
A percentile describes the position of a value within the distribution of a single measurement. A 95th percentile stature means that 95 percent of the population is shorter. A 5th percentile hand length means that 5 percent of the population has shorter hands.
Percentiles are useful and unambiguous for single dimensions. Problems arise when they are treated as descriptions of people. There is no such thing as a “95th percentile person.” Someone with a 95th percentile stature may have a 70th percentile sitting height, an 85th percentile arm length and a 50th percentile hip breadth. Body dimensions are correlated, but far from perfectly.
A Simple Illustration
Consider a seat designed to accommodate users from the 5th to the 95th percentile on four dimensions: sitting height, buttock–popliteal length, hip breadth and popliteal height. On each dimension, 10 percent of users fall outside the range.
If each of four fit criteria accommodates 90% of users and the criteria are independent, about 66% will meet all four (0.9 raised to the fourth power). In practice, the result depends on how much the excluded groups overlap. In a sample of 100 people, if the same 10 people fail all four criteria, 90 remain. If each criterion excludes a different 10 people, only 60 remain.
Real body dimensions are correlated to different degrees. Stature and sitting height are strongly related; hip breadth and leg length much less so. Correlation changes the result, but it does not tell a team where the result will land. To estimate accommodation, check all four criteria for each person in the target sample.
This is the core reason multivariate methods matter. They replace an assumption about accommodation with a measurement of it.
The Percentile Fallacy
The percentile fallacy is the assumption that combining percentile values for several dimensions produces a realistic body, or that designing from the 5th to the 95th percentile on each dimension accommodates 90 percent of users. Both assumptions fail for the same reason: dimensions do not line up perfectly in individuals.
| Common assumption | What actually happens |
|---|---|
| A “5th percentile female” model has 5th percentile values for every dimension | Few, if any, real people share this combination; the model can misrepresent the proportions of real small women |
| Designing from the 5th to 95th percentile on one dimension accommodates 90% of users | True only if that dimension is the only one that matters |
| Designing from the 5th to 95th percentile on several dimensions accommodates 90% of users | Accommodation decreases only when an added criterion excludes someone who met all the previous criteria |
| Percentile models represent the extremes of the population | They represent extremes of individual dimensions, not of the combinations that create fit problems |
Each additional dimension used as a design constraint can exclude people who passed the others. The share accommodated drops only when an added criterion excludes people who met all the others. Researchers Kathleen Robinette and John McConville demonstrated this problem in their 1981 paper on alternatives to percentile models, which became a foundational reference for multivariate accommodation.
Why the Fallacy Persists
If percentile models are so limited, why are they still widely used? Several reasons stand out:
- Simplicity. A table of percentile values is easy to use and easy to communicate.
- Legacy tools and standards. Many design guidelines and digital human tools were built around percentile manikins.
- Data availability. Summary statistics are easier to find than the individual-level data needed for multivariate analysis.
- Habit. “Design for the 5th to 95th percentile” is a memorable rule, even when it is misapplied.
None of these reasons justifies the cost of poor fit, especially when better methods and data are now widely available.
Better Alternatives
Modern human-centered design replaces percentile stacking with approaches that respect how dimensions combine in real people:
- Multivariate accommodation. Define the design problem by the dimensions that matter together, then evaluate what share of the population is accommodated by the combination, using individual-level data.
- Boundary and representative cases. Use statistical methods, such as principal component analysis, to select a small set of realistic body cases that span the multivariate distribution, including combinations at its edges.
- 3D body models from real people. Use 3D body scans or statistical shape models derived from them, so every model reflects a body that could actually exist.
- Accommodation reporting. Report results as the share of the target population a design accommodates, and identify which people fall outside and why.
These methods do not require abandoning simplicity. A well-chosen set of representative 3D bodies can be as easy to use in a design review as a pair of percentile manikins, while revealing problems that stacked percentile values can hide.
Representative personas help a team identify where a design fails and compare alternatives. To estimate how many users a design accommodates, apply the same fit criteria to individual records from the target population, with weighting where appropriate.
Comfo Labs builds representative 3D personas and digital human models from measured population data for exactly this reason. In a hypothetical seat review, a team selects a SIZE LAB persona with a relatively tall torso and short thighs, then uses Doodll to compare two seat pan depths. The comparison shows under which conditions the front edge of the seat pan meets the back of the knee. That finding tells the team which design to revise. It does not, on its own, show what share of the market has that body shape.
The same question applies across the vehicle interior. Companion articles apply it to seat position and seat track range, pedal and steering wheel reach, driver field of view, seat belt fit and child car seat fit.
Percentile Use Checklist
- Single or multiple dimensions: Does the design problem depend on one dimension or several working together?
- Model realism: Are the body models used in evaluation based on real people or stacked percentile values?
- Accommodation claim: Is the claimed accommodation calculated on the combination of relevant dimensions?
- Edge cases: Have unusual but realistic proportions been included, such as long torso with short legs?
- Data source: Is individual-level or 3D body data available to support multivariate analysis?
- Reporting: Can the team say who is excluded by the design and which dimension excludes them?
Daniels’ sample contained no one who was average on all ten dimensions. For a design team, the useful question is not whether each dimension covers the 5th to 95th percentile, but this: who falls outside the design, and which fit criterion rules them out?
Frequently Asked Questions
What is the percentile fallacy in product design?
The percentile fallacy is the mistaken belief that combining percentile values across several body dimensions creates a realistic person, or that designing from the 5th to the 95th percentile on each dimension accommodates 90 percent of users. Because body dimensions are not perfectly correlated, both assumptions fail and real accommodation is usually lower than expected.
Why does designing for the 5th to 95th percentile not accommodate 90% of users?
A 90% range for each dimension does not guarantee 90% coverage across all dimensions. If each of four fit criteria accommodates 90% of users and the criteria are independent, about 66% will meet all four. In practice, the result depends on how much the excluded groups overlap, so accommodation should be estimated by checking all four criteria for each person in the target sample.
Does the average person exist?
Not in Daniels’ sample. In a study published in 1952, Gilbert S. Daniels found that none of 4,063 U.S. Air Force personnel fell within the study’s average range on all ten body measurements. A design based on average values may still fit some users, depending on its adjustment range and tolerances, but its fit has to be checked against real combinations of dimensions.
What is multivariate accommodation in ergonomics?
Multivariate accommodation is a method that evaluates what percentage of a target population a design fits when several relevant body dimensions are considered together. It uses individual-level or 3D body data to identify exactly who is excluded and by which combination of dimensions, rather than checking each dimension separately.
When should designers use percentiles?
Percentiles work well when a design depends mainly on one body dimension, such as door height relative to stature or grip span for a simple handle. When several dimensions interact, as in seats, vehicle interiors, workstations, helmets and wearables, multivariate methods and representative body cases give more reliable results.
What is the best alternative to percentile manikins?
The most reliable alternative is a set of representative body cases selected from real population data, often using principal component analysis, combined with 3D body models built from scans of real people. These cases reflect body proportions that actually exist and help teams find where a design fails and compare alternatives. To report the share of users a design accommodates, apply the same fit criteria to individual records from the target population, with weighting where appropriate.
References
- Daniels, G. S. (1952). The “average man”? (Technical Note WCRD 53-7). Wright Air Development Center, Wright-Patterson Air Force Base. https://apps.dtic.mil/sti/tr/pdf/AD0010203.pdf
- Robinette, K. M., & McConville, J. T. (1981). An alternative to percentile models (SAE Technical Paper 810217). SAE International. https://saemobilus.sae.org/papers/alternative-percentile-models-810217
- Rose, T. (2016). The end of average: How we succeed in a world that values sameness. HarperOne. https://search.worldcat.org/title/926105728
