|
THE COMPETITIVE SEMICONDUCTOR MANUFACTURING HUMAN
RESOURCES PROJECT:
Second Interim Report
CSM-32
Clair Brown, Editor
3.0 Results
Our modeling methodology (Bowen, 1992) is based on the premise that
performance in a given work system is determined by a combination
of factors occurring at four different levels; individual, group,
organizational and external-enviromental.
The functional form of the model is
P=C1-C2X-2,
where
P is the Performance of the work group,
C1, is a constant representing the system upper limit
to performance,
C2 is a constant representing the costs incurred as
a result of imperfect KSA, and X is the level of KSA of the work
group.
For the model, we make the following
assumptions:
-the work system is attempting to produce as much as possible, i.e.,
that production is not purposefully restricted due to inadequate
demand.
- the organizational structure is group-based
rather than individual-,based.
- the group works within a work system
and that there is some autonomy within the group, but there is a
group boundary and limits to the autonomy of the group.
-the amount of formal training is a reasonable
estimate or proxy for the amount of knowledge, skills and abilities
of the group-members.
For the work group under investigation, a linear regression model
using performance as the dependent variable and the inverse square
of KSA as the independent variable produces an equation with the
following coefficients.
Moves-per-week = 13,900 - 4.22 x 108
(KSA)-2 or
P = 13,900 - 4.22 x 108 (X)-2 (Equation
1)
This model of performance results in
a coefficient of determination, R2, of 0.78 and F(l,
22) = 78.2, which is significant at the P<0.0000001 level. The
standard error of the model is 740 moves-per week. The model and
data are depicted graphically in Figures 3A
and 3B. Figure 3A is restricted to the range of existing data,
while Figure 3B shows the performance predicted by the model over
a wider range of group KSA.
Because this study examines a single group, and in order to be unobtrusive
in collecting data, little group process data were collected. For
example, data concerning group cohesion or the emergence of informal
leaders were not collected.
In spite of these limitations, there are still some factors that
we can examine that may have an impact on group performance. To
accomplish this, we first view a plot of the performance data versus
group KSA, and then identify the data points by factors suspected
of influencing performance. We then observe whether these points
consistently fall above or below the performance predicted by the
KSA only model.'
By this method, we find that the data contains interesting patterns
concerning performance differences regarding addition of members
to the group and between three and four day work weeks. We now present
these findings.
The change in KSA over time is shown in Figure
1.
Performance is measured in 'moves', which is the successful processing
of a wafer on a piece of equipment and its transport to the next
processing step or to an appropriate inventory--holding area. The
group's facilitator calculated the weekly average moves-per-shift
as a routine data collection activity. The performance data used
in this report encompasses the time interval from the be-inning
of the introduction of the group structure, to the last complete
month before the site visits occurred, a period of 23 weeks. This
performance information is shown in Figure
2. Time for both Figure 1 and 2 is in calendar year weeks. Both
performance and KSA show increasing trends, with performance fluctuating
week to week and group KSA monotonically increasing.
By this method, we find that the data contains interesting patterns
concerning performance differences regarding addition of members
to the group and between three and four day work weeks. We now present
these findings.
3.1 ADDING MEMMBERS TO THE GROUP
Four individuals were added to the work group during the period
under study, in the 2nd, 4th 16th and 22nd weeks. From
Figure 4, we see that the four weeks in which new members joined
the group are all weeks that lie above the KSA versus performance
curve predicted by the model. This result is interesting in that
it is counter intuitive. One could naturally expect the addition
of a new person would disrupt normal operations and consume the
time of the facilitator and other group members required to orient
and the newcomer, resulting in a reduction in performance which
would last until the new employee 'comes up to speed.' Based on
the data, this does not appear to be the case.
An alternative explanation is that the addition of a person to the
group means an addition of KSA which is not captured by the model,as
this 'innate' KSA is not the result of any of the formal training
that is captured in the training database.. Because the KSA is not
captured in the training data, while the increase in moves-per-shift
is captured in the performance data, the addition of a person to
the group could appear as a jump m group performance. This explanation
must also be rejected because the jumps in performance are not sustained,
i.e., they are always followed by a reduction in the following week's
performance which is not explained by this 'uncaptured KSA' theory.
While the data is sufficient to discredit the above propositions,
it is not sufficient to support or refute others. Further investigation
is indicated, and relationships that could be explored include:
-increase or reduction in cleanroom presence of facilitator
-performance motivated by trying to impress
newcomer or by mistrust of member whose allegiances have not been
ascertained by the group,
-genuine excitement over the prospect
of a fresh face in the group
-increased production on the fastest/easiest
equipment as a newcomer is trained, which inflates the move-per-shift
performance numbers, and requires later balancing when the harder/more
time consuming- processes are required to catch-up.
3.2 THREE VERSUS FOUR DAY WORK WEEKS
The group alternates between working three and four day weeks. When
the performance data is identified by three and four-day work weeks
in a graph which shows the line predicted by the model (see
Figure 5), it is seen that the three day work weeks are generally
above the predicted performance while the four-day work weeks are
generally below that predicted by the model. Performance occurring
in the four day work weeks shows a steady rate of improvement over
time. Performance in three day work weeks generally exhibits more
variation from what the model predicts From the level of data analyzed
for this study, it is not possible to determine the cause(s) of
this difference between three and four-day work weeks. The data
shows that people were added to the work group only on three day
work weeks. Though this is likely a coincidence' rather than a conscious
policy, it is reasonable to conclude that whatever is driving the
higher than expected performance in the weeks when people are added
to the group is also contributing to the disparity in performance
between three and four day work weeks in the data set.
Some of the variation is likely due to day to day 'random' variation
which is likely to appear more random when days are grouped together
and averaged in smaller sets. However if the variation were due
only to such random causes, the data should reflect this by being
unbiased, i.e. performance in 3-day work weeks would be equally
likely to be above or below the regression line, as would the performance
in 4-day work weeks. For the data to appear as it does due to random
variation is unlikely (p<0.005 by a sign test analysis), leading
to the conclusion that some systematic causes produce the observed
patterns of variation in work group performance.
Factors that could affect the performance differences between three
and four-day work weeks include maintenance or training practices.
For example, if maintenance or vendor training occurs only during
four-day work weeks, this could affect performance. Similarly, if
deliveries are always promised on the first and fifteenth of the
month, then motivation could increase dramatically as deadlines
approach and the likelihood of missing a shipment increases.
Some other potential determinants warranting consideration are;
'communication and hand-off differences' (especially WTP levels
and equipment status) from the other shifts, 'group norms' as the
extra day in a four day work week may lead to a more relaxed pace
of production, and 'fatigue' since by the fourth consecutive day
of 12 hour shifts group-members could be fatigued to the point of
being unable to continue production at the higher pace.
Go to Next Section
of this Chapter
Go to Table of Contents for the
CSM-HR Interim Report
|
