Page 153 - Six Sigma Demystified

134 Six SigMa DemystifieD

However, on the 330 rubber band size setting, there is a very large change in
distance from the target moving from a short to a long drawback distance (left
to right along the line). This implies that drawback distance is a significant
contributor to the change in distance from the target when a large rubber band
is used.
The estimate of the drawback distance effect changes depending on whether
the effect is measured for small or large rubber bands. This implies that there
is an interaction between the drawback distance and rubber band size. This
interaction is revealed by the nonparallel lines on the interaction plot, such as
shown in Figure 6.6.
When interactions are ignored, improvement efforts may achieve haphazard
results:
• Significant factors may appear unimportant when other factors are not
manipulated at the same time, as shown in the one-factor-at-a-time ex-
ample above.
• Process improvement may be realized only when other factors remain
constant. The improvement may seem to “disappear” as the process re-
turns to a prior level of performance for unknown reasons.
• The possibility of reducing the effect of a factor by minimizing variation
of another will not be realized. This Taguchi approach to robust design of
processes can be seen in the above example. If, for example, variation in
rubber band size is costly to control, using a short drawback distance will
dampen the impact of variations in rubber band size.

Historical data often are analyzed with ANOVA and multiple regression

techniques to investigate patterns or significance of process factors. This
so- called data mining has some usefulness, but it also lacks many of the key
properties of designed experiments.
Designed experiments estimate parameter effects with fewer data than data
mining by using an orthogonal array of data. An orthogonal array is a minimum
set of data conditions designed to independently estimate specific factors and
their interactions. The data are collected over a relatively short period of time,
allowing the experimenters to control the conditions under which the data are
collected. Casual factors such as environmental conditions and personnel are
observed or controlled, and anomalies are recorded.
Historical, or happenstance, data often are incapable of detecting interac-
tions. The effects of interactions can be estimated only when the data include