Problems of using anthropometric data

As stated in Section 3.1, percentiles is an easy concept for presenting human dimension distribution for one body dimension for a certain population, but the problems arise when it is assumed that using the data is equally easy. This is particularly true in the common situation where more than one body dimension affects the design (Porter and Porter, 1998; Robinette, 1998a).

Figure 5.5. Illustration of how percentile values are not additive (Robinette and 1 2 3 4 5 6 7 8

Variable Measurement 5 %-ile 95 %-ile

1 Shoulder to vertex 270,5 327,9 2 Bust to shoulder 107,9 177,7 3 Waist to bust 134,2 217,8 4 Buttock to waist 137,8 216,7 5 Crotch to buttock 47,8 104,2 6 Ankle to crotch 578,4 710,9 7 Ankle height 92,3 132,9 TOTAL (mm) 1368,9 1888,1 8 Stature 1525,0 1730,6

One aspect of the difficulty of using percentile values is that they are not additive, except the 50th percentile values (Robinette and McConville, 1981; Annis and McConville, 1990). This consequence is illustrated in Figure 5.5. This phenomenon is caused by the one-dimensional and non-linear nature of percentiles, and the relationships between statistical variables not being preserved when calculating percentiles (Speyer, 1996).

Percentiles do not represent individuals but rather probability distribution data for certain body dimensions within a certain population (e.g. Sitting height, British male, 18- 64 years old). According to Robinette (1998a), the source of the problem is that

percentiles are univariate (one-dimensional) statistics applied to multivariate (many dimensional) situations. Robinette argues that it is common, e.g. for designers and

engineers, to make assumptions about the relationships between the variables that are not true. Since many anthropometric databases present data for male and female as 5th percentile, 50th percentile and 95th percentile values, it is reasonable for a non-specialist to assume that such 'constant percentile people' exist, and that by designing from the 5th percentile female to 95th percentile male, the product would accommodate 95% of the population, due to the overlap of the two distributions (Haslegrave, 1986). This may be true for design based on one dimension, e.g. defining proper headroom in a doorway, but will not be true for multivariate problems, such as vehicle occupant accommodation or workstation design (Roebuck et al., 1975; Porter et al., 2004). Besides, the assumption that the dimension of the 95th percentile male always will be larger than the 95th

percentile female will in some circumstances not be true, e.g. hip breath and chest depth (Annis and McConville, 1990; Smith et al., 2000).

The anthropometric software PeopleSize distinguishes between 'Dimension

percentiles' and 'People percentiles' (PeopleSize, 2004). Dimension percentiles refer to 'common' percentiles as discussed previously, i.e. probability distribution data for one certain body dimension within a certain population. People percentiles indicate which Dimension percentiles would be required to meet the required level of accommodation. This functionality may be of great assistance for a designer. The People percentile is calculated using advanced (Monte Carlo) statistics, which will be further discussed in later sections. The People percentile function can also be used to calculate the actual accommodation based on a set of predetermined Dimension percentiles, similar to the case in Figure 5.4.

Annis and McConville (1990) highlight the fact that gender, race/ethnicity, age and occupation are sources of anthropometric variation, which can have significant effects on

anthropometric data. They give following four examples: 1) Women can with some reliability be rendered as scaled down males for height and weight dimensions, whereas for many other dimensions, particularly those involving body tissue and dimensions of the hands, feet and head, this is not possible. For some dimensions, women are rather scaled up versions of men, such as for buttock circumference and hip breadth

dimensions. 2) White people on the average have shorter legs and arms than blacks, and longer legs and arms than Asians. 3) Stature starts to decline at an accelerating rate after the age of approximately 65. 4) Certain occupations, e.g. airline stewardesses, are typically taller than average women.

Marras and Kim (1993) found significant differences in weight and abdominal dimensions between industry and the US Army populations, the latter often used as the basis for anthropometric surveys. Abeysekera and Shahnavaz (1989) found large differences in body sizes, in almost every part of the body, of people living in different countries. As an effect of this, a product designed to accommodate 90% of the British population (according to stature) would fit only 35% of Sri Lankans and 13% of

Vietnamese (Abeysekera and Shahnavaz, 1989). Income also affects anthropometric data and people with high income are typically taller than people with low income

(PeopleSize, 2004). Also, secular (historical) trends affect the relevance of the

anthropometric data, i.e. changes in body size and rate of growth over time (Peebles and Norris, 2000). The average secular increase in height in Europe and North America is thought to be around 1 cm per decade (Peebles and Norris, 2000). More information about human diversity, related to gender differences, ethnic differences, growth and development, secular trends, social class and occupation as well as aging, is available in Pheasant (1986).

Consequently, when using anthropometric data, it is important to make sure that the data is valid for the design issue at hand, and to know for example from what population the sample was drawn, how big the sample was and how old the study is.

One method that has been used to approximate a percentile person (e.g. to construct 'small' and 'large' crash test dummies) while avoiding some of the pitfalls noted with percentiles, is regression analysis (Zehner et al., 1993). This approach begins with one or two 'key dimensions' such as Stature and Weight, and predicts values for a number of other measurements statistically. Zehner et al. highlight that, while the use of regression predictions provides additive values that can be assembled into a person, the results may not be as uniformly extreme as are usually desired when the intention is to look at the

ends of the body size distributions. Furthermore, in practical applications the regression approach does not take into account the fact that humans show considerable variation in the combinations of dimensions, e.g. that there are numbers of individuals who combine short torsos with long limbs or tall heavy bodies with small heads (Zehner et al., 1993).

Bubb (2004) describes how recent body scanning technologies, enabling three dimensional descriptions of the body surface, are really new forms of anthropometric measurement and application, and that these scanning methods will gain in importance in the future. Advantages of the methods are the fast collection of entire body geometry descriptions and the compilation of measurements for individuals rather than as disintegrated body segment measurements as with the one-dimensional percentile approach.