## Synthesizing anthropometry – overview

Anthropometric data specific to the population for which an artifact, task, or environment is being designed are seldom available. Without detailed information on the body size and shape of the shape of the target users, their spatial requirements can be difficult to consider appropriately. When data have not been collected from the specific population, they can be estimated or *synthesized*.

In practice there are two methods that are typically used to estimate relevant anthropometry. The first involves downsampling or reweighting published data. For example, the ANSUR data are actually a downsampling (to ~4000 individuals) from data collected from nearly 9,000 individuals. This approach ensures, in particular, that the tails of the distributions are well-represented. Methodologies for weighting survey data are well developed and are applied in many domains, from crash statistics to health surveys.

A second method uses relationships within the detailed databases to estimate anthropometry for populations for which only summary data (e.g., stature and BMI) are known. Proportionality constants, for example, describe the average ratios of body segment lengths (e.g., arm or leg length) to stature. They are calculated by taking a large sample of anthropometric data and determining either the mean or median ratio of the length of each measure to stature. Proportionality constants are used due to their simplicity and the ready availability of the necessary input: stature. The upper and lower limits of the anthropometric variables of interest are estimated by multiplying the nth percentile stature by the appropriate proportionality constant. My OPEN Design Lab has a web-based tool for using proportionality constants.

Regression-based approaches relate basic body dimensions (e.g., stature and BMI) to those measures that most directly influence the design. This method exploits the same correlation between stature and body dimensions as proportionality constants, but adds an intercept term and the ability to use multiple measures (besides stature) to obtain better predictions. For example, using BMI improves the prediction of measures of breadth which typically are not well-correlated with stature. This is a critical capability given the rapid increase in the prevalence of obesity in the U.S. population.

Both proportionality constants and regression techniques assume that humans are similarly proportioned across ethnicities. However, body proportions are known to vary both within and across populations. They also assume that two people of a given stature and BMI will be comprised of segments of the same length. Proportionality constants assume that an nth percentile, by stature, individual is comprised of nth percentile segments, which is also not true. While both of these approaches can provide good estimates of mean values, they do not work well in the tails of distributions which is where most of design decisions are made (i.e., designers should look at the 5th percentile value, not the 50th). The proportionality constant approach is the one most often taught in undergraduate engineering courses.