Information Technology Reference
In-Depth Information
in 3D, which is why one should rely on the lines that visualize the clusters (cluster-
ing was performed in 3D). Participant 7 and 8 are excluded, because no individual
model could be fitted. In case of two fitting models per participant (1,2,4,6,10) the
first model is denoted as a, the second as b.
Fig. 3.3 A 2-D perspective of the 3-D visualization of distances between individual's differ-
ent views.
Three clusters of models emerged. Cluster 1 summarizing 6 of the 13 single mod-
els (1a, 2b, 4a, 6b, 9, 10b), cluster 2 summarizing 4 models (1b, 2a, 4b, 10a) and
cluster 3 summarizing the remaining 3 models (6a, 5, 3). The complementary mod-
els (a & b) for these five participants appear to be quite dissimilar as illustrated in
figure 3.3 by the fact that they belong to different clusters. These clusters represent
homogenous views, which can subsequently be mapped out.
3.3.5
Grouping the Homogeneous Views
In the last phase we establish a final set of configurations that represent the major di-
verse views across all participants and all attributes, on the set of stimuli. Views that
belong in the same cluster are analyzed together and a shared MDS configuration is
sought. Attributes that are not adequately predicted by the model are eliminated with
the same criteria as in phase 1. The resulting 'averaged' views are then used for mod-
eling the attributes from all participants. Attributes are allowed to exist in more than
one configuration if they are adequately explained by all of them. When attributes
in the same semantic category are not significantly different (which can be deduced
from the fact that they have overlapping confidence ellipses in the K-dimensional
Search WWH ::




Custom Search