Go to: Open
Ergonomics Home PeopleSize
Zone Safe
Office Practice
Psz Home Page Comparison of the 2 Versions Prices & How to Buy Details PeopleSize FAQ |
Technical
Document: Estimation Methods
|
Source data were used to calculate ratios with stature and weight, separately for each dimension's mean and Standard Deviation, building on the technique described and validated by Pheasant (ref 57). The program calculates each output in real time. The default calculation of each dimension's mean and Standard Deviation was the weighted average of all the known and available sources for each major racial group: each source value was expressed as a ratio to each of the stature and weight means and SD's of its own dataset, then an average weighted by sample size was calculated. The outcomes of these default calculations were inspected, and modified by excluding outliers where appropriate. Internal consistency of the dataset was also sought, some identifiable groups of different dimensions being treated as sets. Very high Standard Deviations were sometimes observed, particularly in datasets from small samples: these were reduced to lie within the ranges of expected Coefficients of Variation (SD/Mean *100) set out in Pheasant (Bodyspace 1986). Data whose measurement points were not clear were not used unless comparison with other data made this information evident. For most dimensions, the variability observed between different studies of the same population was greater than the variability shown between races, even between Japanese and European data. British, German and USA populations were treated as one for proportions to stature, gaining accuracy from the larger collective sample sizes. Differences in proportions between these populations are thus ignored. Given the inconsistent values reported for these differences, and the increasingly variable racial mix in these countries, the error resulting from this assumption is thought to be negligible. Because anthropometry data are so expensive to collect, surveys are conducted only rarely among civilian populations, and are restricted in the number of different dimensions that are measured. The data in PeopleSize are scaled by both stature and weight, to create a single large dataset covering the main consumer populations, brought up to date as necessary from the time of the original surveys. To cater for the steady increase in body weight in Western populations since the 70's, an adjustment factor has been calculated thus:
Some dimensions did not have known correlations, and occasional errors were seen where dimensions have low correlations with both stature and weight. The coefficients were inspected, to fill in missing coefficients and to adjust erroneously high coefficients. Making Fatty Dimensions Accurate Most anthropometry datasets assume that small people are smaller than the average by the same amount as big people are bigger. This is true for bony dimensions, but not true for dimensions that are affected by fat. Because we can get fatter than we can get thin, a heavy person is disproportionally heavier than a light person. The normal way of calculating the various percentile values is to use a statistic called the Standard Deviation, which is a measure of dispersion. It is calculated using a table of coefficients called the Z table, which defines the characteristic 'normal distribution' of human dimensions. The Z table describes arithmetically the way that most people are near the average, and that extreme dimensions are progressively rarer. However Z table is symmetrical, and so is a source of error if applied to fatty dimensions. The PeopleSize 2000 dataset uses both the Z table and a specially developed variant of it, that we have called W table - because it describes the distribution of Weight. W table was generated using the UK Department of Health adult dataset, which is large, recent, and un-weighted (whereas the NHANES data require weighting to re-establish the national ethnic and age proportions). Weight was chosen because no other fatty dimension is sufficiently common among the various survey datasets. W table was derived by this method: Separately for males and females, the real percentile values of weight were calculated (the value below which each given percentage of values fell). The mean was subtracted from each value, and the remainder was divided by the Standard Deviation for weight. The result in each case was a coefficient that would have been the Z value in a perfectly symmetrical Normal distribution, but in fact was smaller below the mean and larger above it. The differences were greater among women than among men, so a separate W table was created for each. This chart shows the distributions for Z and for male W and female W:
For each dimension, the extent to which it uses Z table and W table is determined by wt.coef - an average of the two coefficients weighted by wt.coef, so that if wt.coef is 1 only W table is used, if 0.5 a 50/50 average is used, and so on. This process creates a substantially realistic fatty distribution for dimensions that are highly correlated with weight. Survey data for elderly people were extensively researched. The various datasets were found to be generally of small samples, and hard to combine because of the variety of age groups and nationalities. Much of the data is now quite old. We concluded that we could not meet the needs of designers with published data, but could use it to validate an estimated dataset. Accordingly the age ranges above 64 were calculated as extensions of the dataset for UK 18-64 year-olds. This is a large well-founded dataset containing 290 dimensions and derived from some 85 sources. Net sample sizes are very large and the data has benefited from extensive validation. The elderly data were calculated as proportional variations from these values using the following logic:
Thus the many sources of raw data were adjusted for stature, weight and spinal length to give an extensive up-to-date set of data for the key elderly age groups. The resulting dataset was validated for consistency with raw survey data and for consistency between sex and age groups. This was in addition to the validations that had been made during the initial comparisons and agglomeration of the source data into the core PeopleSize dataset. |
Psz Home Page Comparison of the 2 Versions Prices & How to Buy Details PeopleSize FAQ
|
Open Ergonomics Home Page |