Average body size shifts across regions and generations, so old anthropometric tables go stale

What it is

Anthropometric data is the product of a specific population measured at a specific time, not a universal constant. Body dimensions show a secular trend — average stature and related dimensions in the same region shift systematically over decades as nutrition and health conditions change — and they also vary by region and ethnicity, since different countries and populations have genuinely different body proportions. Data collected decades ago, or from a different country's population, carries a systematic bias — not random error — when applied directly to today's users or users in another region.

Why it happens

The bias is systematic rather than random because it has a clear direction. Secular trends mostly move one way (stature has trended upward in most populations over the past century), so a decades-old dataset will generally underestimate the size of the same population today, and the underestimate grows as the dataset ages — not noise that can be ignored. Regional differences are not a single direction but a structural difference in proportion — the same stature can correspond to different trunk-to-limb ratios across populations, so one dataset cannot be "converted" to another population with a single overall scaling factor. More troublingly, the most widely circulated, most readily available classic anthropometric datasets often originate from a specific country's military physical examination survey from a particular decade, sampling physically screened young adult male soldiers — a group that represented neither the general population of that country at the time, nor can it be applied to civilians, women, older adults, or minors without correction.

Studying it

Checking whether a dataset is applicable means going back to its sampling metadata: the country and year of collection, and the composition of the sample (military vs. civilian, sex and age coverage), then comparing that against the actual target user population. Comparing datasets published by the same body at different points in time (two anthropometric surveys from the same country decades apart) makes the magnitude of the secular shift directly observable.

Where it stops holding

Even a dataset with a large, carefully collected sample carries a bias if the sampled population itself was systematically screened (healthy males of conscription age only) — that bias does not shrink just because the sample is large. A large sample only reduces random error; it cannot correct for groups the sampling frame excluded outright. Once a product's actual user base extends beyond the original sampling frame — spanning the full age range, including a large share of women, or serving a region other than the dataset's country of origin — a bias uncovered by the original sample must be assumed.

Applying it

• Before using any anthropometric dataset, check its country, year, and sample composition (military/civilian, sex and age range) against the product's actual target population item by item, rather than judging by the dataset's size or reputation alone.
• When the target market doesn't match the dataset's country of origin, or the dataset is too old (a global product relying on one country's decades-old military data), first look for a survey closer to the contemporary, local population; if none is available, explicitly widen the design margin and document that assumption, rather than silently carrying the old data over.
• Verification: identify the most extreme body types actually present in the real target market and age range, and fit-test or measure against them directly, rather than only checking whether the design meets the extreme percentile values quoted in the referenced dataset — the dataset's own extremes may no longer represent the true extremes of today's target population.

Related

• Same group: A11.06.1 Percentile selection and design coverage · A11.06.4 Static dimensions vs. dynamic working dimensions · A11.06.6 The design-to-the-average fallacy
• Adjacent: A11.09 Cultural differences in interface comprehension
• Search terms: secular trend · anthropometric survey · population representativeness