In article <uon58.2302$[EMAIL PROTECTED]>,
Donald Macnaughton <[EMAIL PROTECTED]> wrote:
At great length, and with many quotes, on a very interesting topic. I
fear I may have missed the original postings on this thread, though.
There is however one area that seems not to have been addressed. This is
the field of statistical design of experiments. The word "design"
appears just once in the article, in connections with the t test.
I am really disappointed that there was not some emphasis on the value of
correctly designed experiments at all levels in the sciences, both "hard"
and "soft".
As a non-statistician, non-mathematician and non-academic (merely a
practical chemist who spent his entire working life in industry) I
introduced myself to statistics via the experiment design route using
Brownlee's "Industrial Experimentation" in 1956.
The elegance of simple ANOVA became apparent even to me, but the
introduction to the ideas of "design" were even more exciting. Many
practical scientists at "bench" level can I feel readily appreciate many
of the concepts of "design" and thus the notion of constructing a model
which their experimental work will address and hence "prove" or "fail to
support" the underlying hypotheses.
This I feel is the way to get otherwise sceptical scientists and
engineers into the way of considering their practical real-life problems
as ones that require an "holistic" approach. Few industrial
investigations are single variable problems.
My belief and experience is that too much emphasis on the formal
mathematical exposition of statistical ideas - however relevant they are
to statistics majors - serve only to distance the experimental scientist
from the huge advantages to be gained from making use of designed
experiments in a complex world.
Quite simple examples can serve to generate acceptance and even
enthusiasm for what we might regard as a rational approach but which
might otherwise be discouraging for the newcomer to statistical design of
experiments. I've proved this to myself time and again in an industrial
context.
--
Robin Edwards ZFC Ta Serious Statistical Software
REAL Statistics with Graphics for RISC OS machines
Please email [EMAIL PROTECTED] for details of our loan software.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at
http://jse.stat.ncsu.edu/
=================================================================
