Information Technology Reference
In-Depth Information
specific page indexed by its
y
-axis value. A propositionally larger dot represents the
number of duplicate cross-references. The references within a unit is then typified
by the short distance between their indexes due to the same leading string in their
names. The associated cross references would be represented by points close to the
diagonal.
We examined the characteristics of high-level operations, and found that many
items are loosely connected. This can be seen by the sparse points plotted in
Fig. 9
,
particularly after we group some of the related pages together. Therefore, indepen-
dent and individual probabilities used in Musa OPs are appropriate for high-level
statistical web testing. For example, most of the pages with the highest access
frequencies are index pages for individual academic units within SMU/SEAS, not
tightly linked with each other and with little cross-references. Further analysis of ex-
panded list of frequently visited pages follow similar patterns. In addition, the usage
frequencies as well as the cross-reference frequencies are very unevenly distributed,
as shown by the uneven distribution of points and masses in
Fig. 9
, thus justifying
our use of statistical testing to focus on high-risk/high-leverage areas.
The exception to the loose connection above among the top access index pages
is the hierarchical structure reflected, where there are numerous “up” and “down”
references between the index page for SMU/SEAS and those for its individual de-
partments and other academic units. In fact, there is a cluster of cross-references
representing the common links followed from the site index page and other related
pages to individual sub-sites or units, as represented by the cluster of points that form
a vertical band to the right of
Fig. 9
. This can be tested using a high-level UMM to
test the interaction between SMU/SEAS and its academic units.
While investigating the transition from top-level Musa OPs to middle-level
UMMs, we noticed some natural clusters, which would be handled by individual
UMMs. There is a close link and high cross-reference frequencies within a cluster but
low visit frequencies across pages from different clusters. For example, within each
academic department or unit, there are numerous cross-references but few across
boundaries, as illustrated by the dominance of diagonal clusters in
Fig. 9
. Conse-
quently, a UMM can be associated with each academic unit in statistical testing of
this web site.
On the other hand, completely (or nearly completely) isolated operations can be
tested by the top-level Musa OP alone, or if necessary, can bypass the middle level
and go directly to the bottom-level model for further testing. This latter finding would
require minor adjustments to our initially proposed approach and make our three-
tiered testing strategy more flexible. In effect, we have modified our model to provide
a bypass from top-level Musa OP to bottom-level structural testing.
