Anthropometry Surveys - sampling.
Sampling is one of the most important issues in
anthropometrics - selecting the individuals to measure, and then
persuading as many as possible of those selected to actually
participate. On the whole, people who are of
'aspirational' size are more likely to volunteer than others, so
every refusal tends to bias the remaining sample towards being taller and
Government health surveys have a great advantage in this - the trust that potential volunteers have in their staff
and in the bona fides of the approach - so most other studies
tend to use the government data to set quotas for their
recruitment. However because data processing takes
typically well over a year, and preparation another year or
more, there is inevitably a time lag
between the quota data and the survey.
Also in the nature of things there are always pressures on
the ground dragging each quota cell towards one end of its
range, so that a quota cell for a range of the shortest people,
for example, tends to have more less-short and fewer
very-shortest people in it. So even though in theory the
quota has been met, the survey ends up with a biased sample.
This chart shows a comparison of
UK adult statures according to two different surveys in 2001,
showing the effects of sampling.
The first survey is SizeUK, a commercially sponsored
survey paid for by clothing retailers.
Although it used health survey data for its quotas,
stature in the end was overestimated by about 20 mm on
average (and the error is probably bigger at the extreme percentiles).
This chart shows the same surveys' results for weight.
The differences are even greater than for stature - UK female adult weight has not really averaged as little
as 65 kg since the 1980's, and is now over 70 kg.
|The effect of the bias is to make the
clothing data representative of an unknown subset of the population,
rather than of the population as a whole.