2. Anthropometric data foundations

Belt geometry robustness begins with a precise answer to a deceptively simple question: which body is the belt being routed around? A lap-shoulder belt path that clears the iliac crest of a 50th-percentile male may foul the shoulder of a small female or ride over the abdomen of a heavy occupant. Robustness is therefore not a property of a single manikin but of a family of manikins whose dimensions are drawn from defensible statistical distributions. This section establishes the dimensional vocabulary, the distributional logic, and the public data sources that make such a family reproducible.

Key occupant dimensions and why each matters to belt geometry

Five dimensions dominate belt-path decisions. Stature sets the overall vertical envelope and correlates with seated eye height and reach, but it is a poor standalone predictor of belt fit because torso proportions vary independently of total height. Sitting height (vertex to seated surface) is the more direct driver of shoulder-belt anchor height and D-ring position; two occupants of identical stature can differ by 40–60 mm in sitting height. Hip breadth (typically measured as bi-trochanteric or seated hip breadth) governs the lateral spacing of buckle stalks and the outboard anchor, and it interacts strongly with the lap-belt angle across the pelvis. Shoulder height in the seated posture locates the upper anchor and determines the shoulder-belt angle relative to horizontal, which regulatory and consumer protocols constrain. Torso depth (chest depth at the level of the sternum, or abdominal depth) controls the fore-aft clearance between the belt and the occupant surface, and thus the risk of belt-to-body gap or, conversely, hard contact over the abdomen.

Two cautions apply throughout. First, dimensions are correlated: stature and sitting height share a correlation near 0.8 in most adult populations, and treating them as independent when sampling manikins produces implausible bodies. Second, the relevant dimension is often a seated or posture-specific measure, not the standing one reported in summary tables. Always confirm whether a source reports standing or seated values before using it.

Distributions: percentiles are not people

Anthropometric variables are approximately normal within a homogeneous population segment, which is why percentile tables are meaningful. But a 5th-percentile female is not a person: she is a set of independently selected 5th-percentile values that, combined, describe a body that may not exist. For belt geometry this matters because the binding constraint is frequently a combination — a short sitting height with a large torso depth, for example — that no single percentile row captures.

The practical remedy is to model each dimension as a distribution and sample correlated vectors. A workable approximation uses the mean and standard deviation from the source, a correlation matrix among the key dimensions, and a multivariate normal draw, truncated at plausible physiological bounds. For a design study, 1,000–10,000 sampled occupants per population segment is usually sufficient to expose the extreme combinations that drive belt-path failures. The output is not a single manikin but a family of manikins, each with a full set of dimensions, against which the belt path can be evaluated.

Sources: ANSUR, CAESAR, and NHANES

Three public sources cover most belt-geometry needs, and they are complementary rather than interchangeable.

ANSUR (Anthropometric Survey of US Army Personnel, 1988 and the 2012 ANSUR II) provides dense, well-documented measurements of a working-age adult population, including seated dimensions, with means, standard deviations, and percentile tables by sex. Its strength is measurement quality and the availability of seated posture data directly relevant to belt routing. Its limitation is the population: military personnel, weighted toward younger and fitter adults, not representative of the full civilian range.

CAESAR (Civilian American and European Surface Anthropometry Resource) provides 3D surface scans and a large set of linear dimensions for civilian populations in North America and Europe, ages 18–65. Its strength is civilian representativeness and the availability of surface geometry, which supports torso-depth and hip-breadth extraction at arbitrary landmarks. Its limitation is that derived dimensions depend on landmark definitions, so cross-source comparisons require care.

NHANES (National Health and Nutrition Examination Survey) provides nationally representative US data, including measured stature, weight, and selected circumferences, with sampling weights that support population targets. Its strength is representativeness across age, sex, and ethnicity, and its continuous updating. Its limitation for belt geometry is the sparser set of seated and depth dimensions; it is best used to weight or bound a design population rather than to supply every dimension directly.

From sources to design manikins and population targets

Defining a design manikin family is a four-step process. First, select the population target: for a US passenger-vehicle program, a common target is the civilian adult population aged 18–65, weighted by NHANES sampling weights, with a stated coverage goal such as 90% or 95% of that population. Second, assemble the dimension set: stature, sitting height, hip breadth, seated shoulder height, and torso depth, each with mean, SD, and correlation structure drawn from the most appropriate source — ANSUR II for seated dimensions, CAESAR for depth and breadth, NHANES for weighting. Third, sample correlated occupant vectors as described above. Fourth, reduce to a manageable manikin set by clustering the sampled vectors into, say, 8–12 representative manikins that span the coverage envelope, including the extreme combinations that stress the belt path.

A concrete example makes this tangible. Suppose the target is a 95% coverage envelope for a mid-size passenger car. Using ANSUR II female seated shoulder height (mean ≈ 545 mm, SD ≈ 28 mm as an illustrative figure) and male values (mean ≈ 600 mm, SD ≈ 30 mm), a 95% envelope spans roughly 490–660 mm. Sampling correlated vectors that also vary torso depth from about 200 to 280 mm produces manikins at the corners of the envelope — short-shoulder with deep torso, tall-shoulder with shallow torso — that a percentile-row approach would never generate. These corner manikins are precisely the ones that expose belt-to-body gap at the shoulder and hard contact at the abdomen.

The figure above traces the pipeline from source to manikin family and shows why the corner cases matter: the 95% coverage region is an ellipse in the shoulder-height/torso-depth plane, and the manikins that stress belt geometry sit at its extremes, not at its center. A belt path validated only against the mid-envelope manikin will pass a nominal check and fail the robustness objective.

Practical checks before committing a manikin set

• Dimension definition audit: confirm each dimension's landmark definition matches between sources. Seated shoulder height in ANSUR and a CAESAR-derived equivalent may differ by 10–20 mm purely from landmark convention.
• Correlation sanity check: verify that sampled stature–sitting-height pairs reproduce the source correlation; a sampling bug that drops correlation produces implausibly long-torsoed short occupants.
• Coverage verification: project the sampled vectors onto each dimension pair and confirm the stated percentage falls inside the envelope; a 95% target that yields 88% coverage usually indicates a truncated or mis-weighted distribution.
• Corner-case inclusion: explicitly confirm that at least one manikin represents each extreme combination relevant to the belt path — short shoulder with deep torso, and tall shoulder with shallow torso.

With a defensible manikin family in hand, the next question is how belt geometry is evaluated against each member of that family — the subject of the following section. The foundation established here, however, determines whether that evaluation is meaningful: a belt path is only as robust as the population it was tested against, and that population is only as credible as the dimensions, distributions, and sources behind it.