Quality Progress - July 2013 - 52
BY LYnne B. hare
Wanderlust and Memory
How the EWMA model improves on Shewhart charts
by accounting for process memory
PoP quiz. Question one: How do you
specifications. At best, it will operate within
ed. If data are autocorrelated, their standard
derive Shewhart (X-bar and R, or X-bar and
its inherent, intrinsic capability.
deviation is not the same as that of indepen-
S) control chart limits?
A. Control chart limits are the same as
But more than likely, there will be ran-
dent samples taken from the same process.
dom shocks and interventions of assignable
And when you make decisions on the basis
cause variation that disturb the process and
of control charts, you are—knowingly or
cause the end-product
not—assuming independent samples. More
moved toward the target
variation to exceed
on that later.
from the specifications
limits to allow for sam-
are, or should be, based
of the above.” Control chart limits should
on product perfor-
be based on the estimate of the capability
C. Run the process and wait
mance from an end-user
standard deviation derived from a carefully
until it achieves stability.
perspective, not process
conducted process capability study.1
Take samples of size four
performance from a
or five—or whatever is
tive. All, or almost all,
You get partial credit if you answered either
B. Control chart limits are
So the answer to question one is “none
minutes. Calculate the average range of
product should be within specification
A or B. It is not that these reasons were
the sets of these periodic samples and
part of Walter Shewhart’s intent in develop-
multiply it by A2 from the control chart
If you answered C, you may have been
ing control charts, but it is true that many
tables. Add the result to and subtract it
influenced by some texts on statistical pro-
organizations rely on control chart data as
from the target to derive the limits.
cess control or some training that recom-
a legal record and, when needed, to show
mends 10 easy steps to process perfection
regulatory officials that they are on the ball.
D. Take 30 consecutive samples. Calculate
their standard deviation. Divide it by the
that you saw on commercials listing 800
square root of the usual sample size. Mul-
numbers on late-night TV.
tiply that by three, and add and subtract
the result from the target.
E. None of the above.
One clue to C being the wrong answer is
Grudgingly, I’ll also give partial credit
for C, but only because triggering process
adjustment has become a common use for
that if you wait until the process achieves
control charts. “Wait a minute,” you say.
stability, you could wait forever. Besides, if
“I thought that’s what control charts are
you actually waited for the achievement of
for.” Not so. Shewhart’s original intent was
Shewhart control chart?
stability before you took data, how would you
answer D: to detect the entry of assignable
A. To please regulatory officials.
ever know the process is stable? And how do
cause variation in the process. If you want
B. To maintain a process record for legal
you measure stability in the first place?
to signal the need for process adjustment,
Question two: What is the purpose of a
Answer D has two major flaws. One is
C. To tell when to adjust the process.
that those 30 consecutive samples may or
D. To identify sources of assignable cause
may not be representative of the process
short-term variation. Would you get the
there are better devices than Shewhart
The Shewhart model is:
y = f(x) + ε,
same result if you took a different set of
meaning that some observed process value,
30 samples? Another flaw is that samples
y, is a function of process settings, x, plus
taken adjacently from most processes are
noise, ε. The process mean is fixed, and
highly correlated. That is, when the value of
the variation is experienced around that
If you answered A or B, I’m sorry, but you
one sample is high (low), it is likely that its
fixed mean. This is a useful way to look
got it wrong. Your process doesn’t know
neighbor sample is also high (low).
at processes, and as the late, amazingly
E. None of the above.
That’s it. Only two questions.
about specifications; it doesn’t care about
52 QP • www.qualityprogress.com
The results are considered autocorrelat-
great George E.P. Box said, “All models
Table of Contents for the Digital Edition of Quality Progress - July 2013
Mr. Pareto Head
Measure for Measure
Quality in the First Person
ASQ's Continuing Education and Professional Development Directory
One Good Idea
Quality Progress - July 2013