ROBUST DESIGN FOR SOFTWARE DEVELOPMENT - Software Design for Six-Sigma: A Roadmap for Excellence - page 489

TABLE 18.7 Main Effect

Factor Main Effect

A 0.00

B 0.44

C 0.11

D 0.11

E 0.03

F 003

G 0.44

H 1.77

A summary of all main effects is shown in Table 18.7.
Variation of AB interaction with 2 degrees of freedom:

S Ax B = S AB − S A − S B

= 0 . 44 − 0 . 00 − 0 . 44

= 0 . 00

In the next step, they divided the combinational effect, S AB , and interaction effect,
S A × B , by each corresponding degree of freedom:

S AB
5

Combination effect = = 0 . 09

S A × B
5

Interaction effect = = 0 . 00

Because these results are computed from the approximate two-way tables, they
considered such results to be a clue for debugging in particular if the occurrence of
bugs is infrequent. When there are more bugs or when a large-scale orthogonal array
is used, they used these values for finding bug locations.
The authors succeeded in finding bugs by taking advantage of each combination
of factors (Table 18.8). As is shown, using the method as described, the bugs can be
found from an observation of specific combinations.
Following are the differences between our current debugging process and the
method using an orthogonal array:

1. Efficiency of finding bugs
a. Current process: What can be found through numerous tests are mainly
independent bugs. To find bugs caused by a combination of factors, many
repeated tests need to be performed.
b. Orthogonal array: Through a few experiments, they can find independent
bugs and bugs generated by a combination of factors. However, for a
multiple-level factor, they need to conduct one-factor tests later on.