Page 246 - Six Sigma Demystified

226 Six SigMa DemystifieD

If you have a strong suspicion that the blocking factor would interact with
another factor, you might be able to include it as a main factor. In the cookie
baking example, you could treat each oven cycle as a single run of the experi-
ment and vary the temperature, time, and ingredients for each oven cycle. In
this way, you could estimate interactions among temperature, time, and
ingredients.
There are other factors, sometimes called casual factors, that may have an
impact on your experimental response, such as temperature, humidity, time of
day, and so on. If you think that these factors are truly important, you should
make them controllable factors for the experiment. If you can’t, or you choose
not to because it would increase the size or cost of the experiment, you should
at least measure them. You then can estimate if they are correlated with the
response, which would suggest the need for additional experimental runs to
analyze their effect.

Defining Factor Levels
For each factor, you must define specific levels at which to run the experimen-
tal conditions.
Factors may be either quantitative (measured) or qualitative (categorical).
The qualitative factor categories are converted to coded units (such as –1 and
+1) for regression analysis.
Qualitative factor levels may be inherent to the process or product under
investigation. For example, you may have three product configurations, or you
may be interested in the variation among four machining centers.
For quantitative factors, if you expect the response to be nonlinear with
respect to the factor, you need at least three levels for that factor. Nonlinear

effects are not addressed in initial experiments but instead are left until you can
optimize in the improve stage. Earlier experiments will be used to screen out
insignificant factors, which can be done with only two levels per factor. Bear in
mind that more levels lead to more experimental runs. In addition, software
used for generating designs may limit the design choices when there are mixed
levels for the factors.
When you define the levels for each factor, you want to span the region of
interest. It’s helpful to think of the expected variation you are likely to see for
the factor during normal operations, but sometimes this results in factor levels
being too close to measure an effect. For example, if you think that temperature
typically only varies from 70 to 80 degrees, you may not see much of an effect
owing to temperature over that 10-degree difference. It’s usually better to think