A friend of mine in industry introduced me to "Design of Experiment", and I
was wondering if any of you had heard of and/or worked with this branch of
study. The basic aim of DOE is to obtain accurate, statistically
significant answers to experimental questions from a minimal number of
trials. The principles are applied regularly in industrial settings (at
least by smart companies ;-) to improve products and processes, especially
where there are a large number of variables and the effects of each
variable can't be easily predicted in advance. This also applies in many
biological settings.
Over the past year or so, I have been picking up bits of DOE and find it
enlightening. Given how central experimentation is to physics, I'm amazed
that more ideas from DOE aren't taught to physicists. Guides to
experimentation in physics tend to concentrate on specific techniques, but
not on general questions like dealing with multiple variables or picking an
effective set of trials to obtain the desired information. Other than
particle physicists, most experimentalist are happy that they don't have to
stop to using statistics and are content to use inefficient (or even
ineffective) plans to get to the answer they want.
Some of the DOE techniques require statistical sophistication to fully
understand, but you don't need to know all the details to apply the
techniques (just like you don't need to know how a least squares fit is
done to recognize its usefulness). There are many designs that have been
developed for various types of questions - curve fitting, optimizing a
result, optimizing consistency, or determining which subset of parameters
has the greatest effect on the results.
A few results that may or may not surprise you:
* DON'T vary one parameter at a time
* perform trials in random order, rather that stepping through in order
* standard deviation isn't a good measure of accuracy (but you probably
knew that one)
* optimizing consistency is often more important that optimizing magnitude
At one level I equate this with the current discussion of Geometric Algebra
- both provide powerful improvements over current tools, but few people
even recognize that improvement is warranted.
Most university libraries have books on the subject. A few guides are
available on line, but I haven't seen any that really jump out at me. One
way to learn is to just download a trial version of DOE software (e.g. http://www.umetrics.com/methodtech_doe.asp?section=methods or http://www.statease.com/dx6descr.html) and play around. Many statndard
statistical applications can do DOE as well (I know Minitab does)
Tim Folkerts
Department of Physics
Fort Hays State University
Hays, KS 67601
785-628-4501
This posting is the position of the writer, not that of SUNY-BSC, NAU or the AAPT.