From: email@example.com (Brian Teasley)
Subject: Re: Design of Experiments Workshop in NYC, SF, Miami
Date: 7 Dec 2002 10:38:04 -0800
NNTP-Posting-Date: 7 Dec 2002 18:38:05 GMT
I'll try to be brief on a few different things mentioned in the above
I chose the usenet groups that I did by conducting some keyword
searches. When I found questions pertaining to DOE I noted the groups.
That is why I posted where I did. I see the USENET groups as a
resource - not my personal resource, but a place where researchers and
engineers can go to stay informed on various topics.
In my keyword searching, I found a (not surprising) lack of extensive
knowledge or sources for information about DOE type topics. That
doesn't mean there aren't any, it just means I did not see many within
the USENET groups. So I posted on appropriate groups - and while part
of the purpose is to "advertise" a workshop, it is also a way to put
out on opportunity for people to contact a resource, if they wish,
about a DOE, Taguchi, and other statistical concepts. I see that as a
positive benefit, and a worthwhile contribution to the intent of the
USENET groups. I have already been asked, and answered, a couple of
good questions from appreciative people.
As for the "Why Should Anyone Care About DOE?" and "Don't vary two
things at a time" comment/questions:
You should care and know about DOE, or how to properly design tests if
you are conducting experiments or running processes that have multiple
variables that could influence them.
If, for example, you have a product that could be influenced in the
manufacturing process by different heating temperatures, exposure
times, chemical mix amounts, humidities, presence or lack thereof of
certain components or ingrediants, etc., you definitely should know
how to test using DOE concepts.
Multiple variable tests are actually designed so you CAN test
different combinations of different variables, rather than the "don't
vary two variables" method mentioned by somebody in a previous post.
There are situations when you might have 50 or 100 variables that you
want to examine. In some situations it is impossible to conduct 100
tests on all the combinations and levels of the 100 variables. But if
you set up a proper test design, you can more efficiently and
effectively examine the influence of all the variables. That will
allow you to examine what you need to in order to solve the problem on
which you are working.
Fortunately it is more common to be just dealing with a smaller number
of variables - perhaps four or five. The designs will still help you
set up the most effective and efficient test possible.
In the marketing world it is common to test concepts, offers,
products, pricing, regions, approaches, etc... the same test design
concepts that apply in manufacturing/engineering industries can apply
to marketing. DOE methods are more widely known in the
manufacturing/engineering world, but are becoming more common recently
in marketing. (They have been around, and even are mentioned in
Ogilvy's book. He makes a point of saying he did not understand terms
I will skip my biography, but I have worked in both
manufacturing/engineering and marketing areas. The same statistical
and DOE tools can be applied in all of the industries.
It is interesting to me though: A few years back I was explaining the
concepts to a senior executive at a very large marketing/advertising
company. I explained how some of the testing they were doing for a
very large national U.S. company could be done with half the sample
size they were using, if not less. At about $1 per test subject and
"sample" sizes in the millions (I'll skip the discussion about why
they were 'testing' to that many people), we were discussing to me
what is a lot of money. He simply did not believe that the test
designs would do what they do. I do not think he used the term "bogus"
as somebody here did in one of the posts, but he was very skeptical.
Today the company has a Phd Statistics expert on staff to help them
design the tests.