Page 247 - Six Sigma Demystified

Part 3 s i x s i g m a to o l s 227

of “worse-case scenarios,” where the factor may vary considerably more. Gener-
ally, the wider the difference between the factor levels, the easier the effect will
be to measure.
When you start moving far away from normal operating conditions, you can
enter unknown terrain that even can be hazardous for some processes. In this
case, it might be better to keep the factor levels at reasonable values and, if the
factor is significant, perform additional experiments using the response surface
or evolutionary operation techniques to find optimal factor levels.

Conducting the Experiment
When it comes time to implement the experimental design, some helpful
guidelines include
• Be there! It’s important to be an active participant in the design. The data
collection is a critical part of the learning process, offering opportunities
to experience aspects of the process dynamics that may not have been
discussed in problem solving. For this reason, and to limit potential bias in
the data, process personnel should run the process and collect the data.
• Randomize trials. The order of the experimental runs should be random-
ized across the entire design and within blocks (if blocking factors are
used). This randomization limits any potential bias introduced during the
experiment.

• Independence of runs. When each condition is run, the process should not
be influenced by prior conditions. Some processes will require their setup
conditions to be torn down and reset.

For specific methods of constructing designs, see “Factorial Designs” and

“Central Composite Design” in the Glossary.

Interpretation

Analysis of the experiment includes many tools discussed elsewhere in Part 3.
See “Regression Analysis” as a starting point.
When an experiment is run in blocks, the design isolates the effect of the
blocking factor so that its contribution to an ANOVA may be estimated. As a
result of this design, interaction effects cannot be estimated between the block-
ing factor and any of the main factors. In the regression analysis, if the blocking
factor is significant, its interactions with the significant main factors can be
investigated with additional runs.