Page 120 - Six Sigma Demystified

Chapter 5 m e a s u r e s tag e 101

come and go. A persistent cause, usually referred to as a common cause, tends to
influence all data in a relatively uniform fashion. Common causes of variation
are inherent to the process itself, by design. A given reason (the actual cause)
often can’t be assigned to a specific amount of variation, but the total level of
variation is considered typical for the process owing to the influence of these
(unknown) common causes. This amount of variation should be expected in
the process.
Common causes of variation reflect our ignorance of process dynamics. So
long as the amount of variation is acceptable, the ignorance can be tolerated.
Since the variation is persistent, the process can be considered stable and can
be predicted with the appropriate statistical tools.
In contrast, special causes of variation are sporadic in nature: They come and
go, generally in an unpredictable fashion. This sporadic occurrence causes the
process to behave differently in their presence, resulting in process instability.
In the presence of special causes, the process outcome cannot be predicted
accurately because the process is not stable.
While logic, reason, or scientific and engineering principles instinctively may
be used to differentiate between common and special causes of variation in a
process, this is a mistake: Only a properly designed statistical control chart can
distinguish correctly between common causes and special causes of variation.
A statistical process control (SPC) chart, developed by Walter Shewhart in the
1920s, provides an operational definition of a special cause.
SPC uses the element of time as one of its principal axes. Samples are col-
lected over a short period of time. This sample is referred to as a subgroup. Each
subgroup indicates two things about the process: its current location and its
current amount of variation.

Once enough subgroups have been collected over a period of time, the
short- term estimates (i.e., the subgroups) can be used to predict where the
process will be (its location) and how much the process is expected to vary over
a longer time period.
Figure 5.3 displays an SPC chart calculated from the same data used to cal-
culate the confidence interval seen earlier. The control chart reveals something
hidden in the distributional curve in Figure 5.2—the element of time. There is
a predictable variation inherent to the process evident for the first 40 or so
samples. During this period, the process variation was relatively stable.
The most recent sample at 3.8 percent looks different from the earlier 40
samples but is much like those seven months immediately prior to it. These last
eight samples apparently represent an unknown shift in the process. The “prior