BEST USE OF IN-HOUSE STATISTICIANS
When management wants the benefits of statistical Design of Experiments (DOE), Multi
Variable Testing (MVT) and Response Surface Methodology (RSM), three types of staff are
needed on the team:
- In-house statisticians trained to the post-graduate level in DOE, MVT and RSM.
- Scientists and engineers trained 4 days in DOE, MVT and RSM so that they can do many of
their own experiments and understand the help they get from the statisticians.
- Managers trained to monitor the benefits of DOE, MVT and RSM.
DUPONT'S MANAGEMENT OF DOE
Statisticians are in short supply. When I was at duPont in 1956, we had three
statisticians for a company with 90,000 employees; our group increased to 15 statisticians
by 1971 when I left. For best use of duPont's R&D dollars we should have been involved
in every experiment. We didn't have enough in-house statisticians to do that. We came up
with a way to use our statisticians where they were needed most and also use statistical
design of experiments throughout duPont:
We prepared a training course (in the 60s) to teach scientists and engineers how to do
the simpler experiments without the help of a statistician. We called it The Black Box
Course (chemical processes are "black boxes" to be mastered by tweaking the
process settings in a statistically designed series of experiments).
In preparing this course at duPont, we found out how expensive it is to develop a
course which will teach the basics rapidly and effectively. We used small computers
(analog at the time) to simulate real processes with random noise to make the examples
intensely real. Since then duPont has been recovering their initial investment in
course-writing by teaching people outside duPont for a fee.
Lessons Learned at Dupont using the Black Box DOE course:
- It works: Scientists and engineers can learn in a few days to design and analyze
many of their own experiments in a huge variety of applications.
- It's expensive to write a good course. A large market is needed to recover the
cost. An R&D company should be cautious about taking on the writing of an in-house
course.
- After scientists and engineers have been taught the basics, they will more
often seek the help of in-house statisticians to apply what they understand. They will
go beyond what they learned with the help of the professional statistician because they
understand the basics.
Experience in Managing DOE at Other Companies
Phil Bantz brought our Basic Experiment Strategies course (DOE) into Monsanto. He saw
that his statisticians were being used more often after the engineers had been trained in
DOE; Bob Easterling at Sandia National Labs in Albuquerque, NM had this same experience.
Advice to Managers
Teach people the Basic Experiment Strategies to start them using DOE. Then they will
make better use of professional in-house statisticians. Scientists and engineers will be
more likely to use DOE if they have a basic understanding of the concepts. They will get
better results in less time.
Teach managers how to monitor the effectiveness of statistical experiment design and
you will see better results at a lower cost, in less time.
Don't Reinvent the Wheel
- Use an existing course: Basic Experiment Strategies, from The Experiment Strategies
Foundation
- because it has been proven successful;
- because it costs less than writing your own course.
- Then use your in-house statisticians to help scientists and engineers apply the Basic
Experiment Strategies they have learned.
CONCLUSIONS
- Don't use in-house statisticians to write, update, and teach courses in statistical
design of experiments. That's not the best use of their time. Statisticians are needed to
help scientists and engineers with specific applications.
- Use a proven training course; have DOE taught by specialists who concentrate on
delivering excellent teaching. The Experiment Strategies Foundation's course, Basic
Experiment Strategies has been taught since 1979 with excellent results.
- Use our Manager's DOE Evaluation Kit to determine:
- "Who among my scientists and/or engineers already know how to use Design of
Experiments (DOE) and are they using what they know?"
- And: "How much time and money will I save by using DOE?"?