THE COMPETITIVE SEMICONDUCTOR MANUFACTURING HUMAN
Second Interim Report
Clair Brown, Editor
14. Statistical Tools for Industry Data
Pursuing Interesting Leads
After viewing the two original "overall picture" graphs,
some of the more interesting aspects may be pursued:
1. What are the differences between
the high PIRK and low PIRK fabs?
2. What are the differences between the high performing fabs and
the low performing fabs?
3. Are there detectable differences in any of the above that are
Five hundred factors are still too many
variables to go through without some statistical help. Several methods
are proposed to help pinpoint delineate patterns.
Percent Alike Graphs
One would like to find the interesting
areas of an organization that reflect the differences found in the
fabs overall. For example, if one is interested in finding the organizational
areas of fabs in group A that differ from fabs in group B, these
areas should have the following characteristics:
The similarity measure is used, but instead
of using it on all the organizational factors, each area of an organization
is measured separately. The organizational factors are grouped into
the following 11 categories:
1. Com: Communications. Factors involving
meetings, problem notification, work instructions and shift hand-offs.
2. DT-A: Direct Teams - Approach. How teams involving direct labor
(operators and technicians) are organized and set up.
3. DT-F: Direct Teams - Function. How teams involving direct labor
(operators and technicians) function - responsibilities, training,
meetings and member descriptions.
4. Hind: Hindrances/Desired Changes. Problems employees may have
or changes they would like to see.
5. Hist: Historical Changes. Changes that occurred in the past
such as work force reductions, new hours introduced, demand dropping,
6. IT-A: Indirect Teams - Approach. How teams involving indirect
labor (engineers and management) are organized and set up.
7. IT-F: Indirect Teams - Function. How teams involving indirect
labor (engineers and management) function - responsibilities,
training, meetings and member descriptions.
8. Knowledge & Skills. Education requirements and training.
9. Org: Organization. Shifts worked, work organization, job grades
10. Perf: Performance. Perceived job priorities and performance
11. Task: Tasks Performed. Job category tasks, leadership roles
and how time is spent.
The groups analyzed are:
To create the "Percent Alike Graphs"
the data of similarity scores need to be acquired. The techniques
used to obtain these scores are outlined in the appendix. The following
sections display and explain the graphs.
High PIRK versus Low PIRK
The similarity measures of high PIRK
fabs with each other (H) are in the left bars of the bar pairs.
The similarity measures of high PIRK fabs with low PIRK fabs (L)
are in the right bars. It is interesting here to note that over
half of the high PIRK fabs are also Japanese fabs (thus separating
the strictly Japanese characteristics with the PIRK characteristics).
In this figure, the interesting areas include:
K&S: Knowledge & Skills
and, possibly, with less of a contrast:
DT-A: Direct Teams - Approach
DT-F: Direct Teams - Function
IT-F: Indirect Teams - Function
The top three interesting areas are not too surprising since these
areas relate directly to one or more of Power, Information, Rewards,
and/or Knowledge. The possible interesting areas related to direct
teams probably display more of a Japanese/non-Asian difference than
a High/Low PIRK difference, since the minimum is so low in these
measures. It is especially interesting to see the high similarity
in communication with high PIRK fabs since this does not seem to
be a Japanese similarity. The factors will be explored in greater
detail with the Common Value Analysis.
High Performance versus Low Performance
In Figure 14-11
it is shown how similar high performance fabs are with each other
in the left bar and how similar high performance fabs are with low
performance fabs in the right bar. In this figure there are two
areas of interest:
As discussed previously, performance
is not solely a result of organizational characteristics, and Figure
14-11 certainly seems to reflect this. However, the graph reflects
only the interesting areas of an organization where all the high
performing fabs do very similar things. If each high performing
fab was on its own "road to success," the areas may be
too dissimilar to be reflected in the graph. Within the CSM study,
high performing fabs did tend to stress the importance of the direct
employee. High performance fabs tended to have their operators and
technicians do more complicated/self-managed tasks (DT-F), thus
acquiring (or requiring) a more advanced set of skills (K&S).
It may be of interest to see if there is anything the low performing
fabs have in common. Is there, perhaps, one road to mediocrity?
To study this, another Percent Alike Graph (Figure
14-12) was created which displays the similarity measures the
low performing fabs have with each other versus the similarity measures
with the high performing fabs.
There are few striking contrasts. The biggest differences seem to
be in the areas of:
Even these areas do not reflect the extreme differences found in
the other analyses. However, employees of low performing fabs in
the CSM study did tend to view their job priorities in terms of
"whatever my boss tells me to do" (Perf). They also had
hindrances (Hind) which reflected a negative view of management
rather than, for example, problems with equipment. Training (K&S)
was not emphasized as much in low performing fabs. These differences
will be explored further in the next section.
Common Value Analysis
Description of Technique
To explore the individual components,
a method of Common Value Analysis is proposed. This analysis finds
the individual factors of a "Group A" that greatly differ
from those in "Group B." The technique used is described
in the appendix.
The Common Value Analysis was used on the two interesting group
High PIRK versus Low PIRK
In the Common Value Analysis of high
PIRK versus low PIRK fabs, many factors were found that were reflected
in only the Japanese companies. This is expected because of the
high correlation between the two. What is interesting to find are
factors or themes in high PIRK fabs that are not necessarily "Japanese"
qualities. These are highlighted these as follows:
Meetings are more common in high
PIRK fabs, especially employee-supervisor meetings and with
peers in shift hand-offs.
There is a theme in high PIRK fabs
of the employees knowing how performance is. This can be through
formal evaluations, supervisory meetings, or simply a good knowledge
of SPC to know when a machine is not performing correctly.
On-the-job training is a PIRK theme.
As opposed to classroom training, employees learn by doing,
usually at the hands of their supervisors or team leaders. Extensive
training in Japanese companies is well documented in the literature
(, and  for example) and is beginning to show itself
in high participative companies outside Japan.
It may be possible that a fab could become
a high involvement fab by focusing on one of the PIRK aspects. For
example, one high PIRK fab concentrates its efforts on information.
Monitors are everywhere in the fab, giving detailed information
on each piece of equipment. Communication is encouraged between
all employees, further spreading information. By placing a high
emphasis on information to the lowest levels, management may be
implicitly stating that operators are important; thereby eventually
giving more power, knowledge and rewards to this lowest level.
High Performance versus Low Performance
As discovered earlier, a theme in high
performing versus low performing fabs is difficult to detect when
so many other factors are involved in high performance. The list
of interesting factors in the Common Value Analysis is short, but
a few points can be mentioned:
Pre-employment screening for operators
and technicians tends to be common in high performing fabs.
Low performing fabs seem to have
disgruntled employees who mention production problems, inadequate
training, and communication problems as some of their job hindrances.
Using multivariate analysis and exploratory
methods some interesting and intriguing possibilities were discovered.
Using the tools outlined in this chapter, the following industry
characteristics were brought to light:
Mapping organizational practices
in fabs relative to each other resembles a map of the world.
The progression from Asia to Europe is easily distinguished
in this graph (Figure 14-1).
Significant biases in the methodology
were found, mostly in the type of data that were coded rather
than in the coding itself. It was also discovered that much
of the bias can be removed by eliminating variables that are
coded only in a particular year.
Once biases were removed, interesting
patterns appeared in the data. PIRK and fab performance increased
as "self-management" increased and "uncertainty"
The differences between high and
low PIRK fabs seemed to lie in the specific areas of communication,
knowledge and skills, and performance.
There are many roads to high performance
(and low performance as well). High performing fabs are similar
in the way their direct teams function and requiring (or allowing
employees to acquire) more advanced knowledge and skills.
The tools developed in this chapter help
a researcher delve into a large data set to underscore interesting
characteristics for further research. The observations highlighted
in this chapter may be used as focal points for concentrated research
efforts and/or theories to test with new data.
End of Chapter 14
Go to Appendix
of this Chapter
Go to Table of Contents for this
Go to Table of Contents for the
CSM-HR Interim Report