Page 160 - Six Sigma Demystified

Chapter 6 a n a ly z e S tag e 141

For example, how many design replicates are needed to estimate an effect
l
with a magnitude 20 for a two- evel factorial design with 5 factors, 8 corner
points, and no center points when sigma is 17 (using α = β = 0.05)?
Using Minitab’s Stat/Power & Sample Size/2 Level Factorial Design function
(alpha is entered using the “Options” button), five replicates, or 40 runs, are
needed.
The various parameters can be manipulated in the Minitab dialog box to
estimate their impact. For example, a larger effect, such as 25, would require
only four replicates. If the power of the experiment is reduced to a 75 percent
chance of detecting an effect with magnitude 25, then only three replicates are
needed.
The purpose of this discussion is to emphasize the limitations of an experi-
ment, which can be overcome by replicating the results before proceeding to
implementation in the improve stage.

PRojeCt exAmPle: Determine­Process­Drivers

The team proposed an experimental design to determine the significance of the
proposed process factors (from the brainstorming exercise discussed previously).
The response variable selected is the process time, calculated as the difference
between the recorded start time and the recorded end time for each order less
any wait time experienced during the process order. actual orders will be se-
lected from previous orders to match the experimental conditions (further de-
fined below). The process under study will be conducted just as in current

operations until the pass- off to accounting (and limited to nonreseller orders).
The personnel entering each order will be removed from any operational respon-
sibilities during the experiment so as to prevent any uncontrolled interference
with the experiment.
The team established factor level settings that would represent the best and
worst cases, assuming that the factor was significant:

• Product family. The baseline data indicated that Product A had the highest
cycle time and Product C the lowest. (Note that this doesn’t mandate that
Product A be the low level for that factor, because the relative impact on the
response is assumed to be unknown, so it does not determine the factor
levels.)