Yesterday I had the opportunity to attend a seminar by George Box where
he discussed some of the ideas that will be incorporated in the second
edition of Box, Hunter, and Hunter "Statistics for Experimenters" due
out in a few months.
At the end of the presentation he distributed a list of quotes from the
book and I felt that many of these would be appealing to members of this
mailing list.
I refer those who want R-related content in messages to this group to
the quote "Seek computer programs that allow you to do the thinking."
My thanks to Professor Box for giving me permission to forward these.
QUAQUAVERSAL QUOTES
The following list of quotations may be used for a number of purposes.
You may wish to be reminded of some of the ideas in this book.
Your boss, who may not have time to read the whole book, can employ
them to understand the philosophy of what you are doing.
If you use the book to teach a course some quotes can be used as
topics for short essays.
Among the factors to be considered there will usually be a vital few
and a trivial many. (J.M. Juran)
A process should be routinely operated in an evolutionary mode so as
to produce not only product but information on how to improve the
product.
Sometimes the only thing you can do with a poorly designed experiment
is to try to find out what it died of. (R.A. Fisher)
The experimenter who believes that only one factor at a time should
be varied is amply provided for by using a factorial experiment.
If there were a probability of only p = 0.04 of finding a crock of
gold behind the next tree, wouldn't you go and look?
The democratization of Scientific method.
Designing an experiment is like gambling with the devil: only a
random strategy can defeat all his betting systems. (R.A. Fisher)
Seek computer programs that allow you to do the thinking.
When the ratio of the largest to smallest observation is large you
should question whether the data are being analyzed in the right
metric (transformation) .
Original data should be presented in a way that will preserve the
evidence in the original data. (W. A. Shewhart)
You can see a lot by just looking. (Yogi Berra)
A computer should make both calculations and graphs. Both sorts of
output should be studied; each will contribute to understanding.
(F.J. Anscombe)
Murphy works hard to ensure that anything that can go wrong will go
wrong. With an adequate system of process monitoring, therefore, more
and more of the things that can go wrong will be corrected and more
and more of Murphy's tricks can be permanently stymied.
A useful type of time series model is a recipe for transforming
serial data into white noise.
When you see the credits roll at the end of a successful movie you
realize there are many more things that must be attended to in
addition to choosing a good script. Similarly in running a
successful experiment there are many more things that must be
attended to in addition to choosing a good experimental design.
Iterative inductive-deductive problem solving is geared to the
structure of the human brain and is part of every day experience.
What does what to what? How, with a minimum of effort, can you
discover which factors do what to which responses?
Only in exceptional circumstances, do you need to try to answer all
questions with one experiment.
Actions called for as a result of an experiment are of two kinds:
1) "Cashing in" on new knowledge
2) Using the new knowledge to look for further possibilities of
improvement
The business of life, is to endeavor to find out what you don't know
from what you do; that's what I called "guessing what was at the
other side of the hill". (Duke of Wellington)*
The best time to plan an experiment is after you've done
it. (R.A. Fisher)
Every model is an approximation.
It is the data that are real (they actually happened!)
The model is a hypothetical conjecture that might or might not
summarize and/or explain important features of the data.
All models are wrong; some models are useful.
Don't fall in love with a model.
It is a capital mistake to theorize before one has data. Sherlock
Homes in "Scandal in Bohemia" (Conan Doyle)
It is not unusual for a well-designed experiment to analyze itself.
Correlation may have nothing to do with causation: beware the lurking
variables(s)!
The idea of a process in a perfect state of control contravenes the
second law of thermodynamics: thus a state of control is an
unrealizable and must be regarded as a purely theoretical concept.
The design of experiments was invented by R.A. Fisher to make it
possible to conduct valid experiments in an environment (agricultural
trials) that was never in a state of control.
To find out what happens when you change something it is necessary to
change it.
It's better to solve the right problem approximately than the wrong
problem exactly. (J.W. Tukey)
Experiment and you'll see!
Perfection is not possible it's always an approximation.
Most often an experiment does not allow us to make a final decision
but to see what's worth trying.
"Block what you can and randomize what you can't" can
approximately
justify an analysis "as if" standard assumptions were true.
The largest member of any group is large - but is it exceptionally
large?
Where there are three or four machines, one will be substantially
better or worse than the others. (Ellis Ott)
That conclusions reached in one environment (say from lab
experiments) will apply in a different environment (say the full
scale process) is based not on statistical reasoning but on what
Deming called "a leap of faith". Statistical methods can reduce but
not eliminate the necessary leap.
Discovering the unexpected is more important than confirming the
known.
One must learn by doing the thing; for though you think you know it,
you have no certainty until you try. (Sophocles)
We should not be afraid of discovering something.
When running an experiment the safest assumption is that unless
extraordinary precautions are taken it will be run incorrectly.
Knowledge is power (Francis Bacon)
Show me the data!
With sequential assembly, designs can be build up so that the
complexity of the design matches that of the problem.
At any given stage, the current model helps us appreciate not only
what is known but what else it may be important to find out.
