Database Reference
In-Depth Information
lines in code' and 'number of methods in a class'. It is interesting to note that this property
is neither satisfi ed by the suite of cohesion metrics that Chidamber and Kemerer (1994)
themselves propose.
Nonequivalence of Interaction
It is easy to see that the fi fth property, nonequivalence of interaction, will be satisfi ed in
most practical cases. We only have to consider two different activities with equal cohesion
and an operation that is related to one or more activities within the fi rst, but to none of the
activities in the other activity. In adding the operation to both activities, the cohesion may
respectively increase in the fi rst case and will
always
decrease in the second.
Interaction Increases Complexity
Finally, the sixth property, interaction increases complexity, can be satisfi ed by choosing
any two operations from an operations structure that share an input or output. Two separate
activities with only one of these operations will have a cohesion that equals zero; one activity
that includes both operations will have a positive cohesion, so that the property is satisfi ed.
Although it will be not very common that µ((( ) + µ(
Q
) < µ((( +
Q
) in a practical case, our
criterion is specifi cally intended to identify cases where µ((( ) < µ((( +
Q
) and µ(
Q
) < µ((( +
Q
).
After all, in this situation we would prefer to combine
P
and
Q
(see example 2).
Discussion
On the basis of the above evaluation, the cohesion metric we presented could be said to
satisfy all relevant theoretical requirements of Chidamber and Kemerer (1994). Obviously,
there are many reservations to be made, for example the absence of more context-specifi c
theoretical requirements. As this is really a best effort, it is interesting to take a look at an
empirical evaluation of the heuristic as well.
EMPIRICAL PERSPECTIVE
Web-Based Survey
The approach we followed for the empirical evaluation is to test the heuristic by ap-
plying it to a set of design dilemmas and compare its outcomes to the judgment of human
experts. For this purpose, we used a digital web-based survey, which contained ten design
dilemmas in the same spirit as the examples of Figures 1 and 2. A respondent must choose
for each of the dilemmas on a three-point Likert scale whether he or she:
•
prefers to combine the operations in one large activity,
•
has no preference for combining or splitting up these activities, or
•
supports the split-up of the same operations in the two
given
activities.
The respondent is instructed to follow his intuition whether the operations as depicted
seem to “belong together” or not. The only thing that is explained to the respondent is the
meaning of the used symbols in each fi gure and the context of workfl ow design. A screen-
shot of one of the presented dilemmas in the web-based survey is given in Figure 3.
