Personal Attribute Judgments - Modeling Users' Experiences with Interactive Systems - page 29

2.6 Analysis Procedure

The analysis procedure consists of three steps. First, a user segmentation map that
expresses the diversity among individuals is derived from the collected dissimilarity
ratings by means of Multi-Dimensional Scaling (MDS) (Martens, 2003). Homo-
geneous groups of users are identified within this map by means of (hierarchical)
clustering. Secondly, attributes are classified into categories based on their seman-
tic content. This semantic classification will then be contrasted to attribute ratings.
Third, perceptual maps are created from the attribute, dissimilarity and preference
ratings to express how homogeneous groups of participants perceive the products
being studied.

2.6.1 Identifying Homogeneous User Groups in the User

Segmentation Map

In this step a user segmentation map that expresses the diversity among users is de-
rived from their dissimilarity ratings by means of Multi-Dimensional Scaling. To
create the user segmentation map, we define the distance D i, j between participants
i and j based on the correlation R i, j between their dissimilarity scores. Derived dis-
tances are then visualized in two or more dimensions using the MDS tool XGms
(Martens, 2003). Figure 2.5 displays a two dimensional configuration of designers
and users. The closer two individuals are in the two-dimensional space, the more
their ratings of overall dissimilarity of products correlate. The dimensionality of the
configuration may be judged by the experimenter using the Stress Value which is an
index of goodness of fit of the model. In this case, the two dimensional visualization
was judged as adequate (stress value S=0.18) (Clarke, 1993).

R i , j D i , j = 1 − (2.2)

∑ k D i (

k ) D j k

R i , j = D i 2 (2.3)

D j 2

( k ) ( k )

Hierarchical clustering performed on the reduced two-dimensional space (with min-
imum variance) reveals two main clusters that can be further subdivided into five
more or less homogeneous participant groups. Groups 3 and 4 consist entirely of
end users, while groups 1, 2 and 5 consist mostly of designers. Identification of
the designers reveals that group 1 consists mostly of technically-oriented designers,
while group 2 consists mostly of user-oriented designers.

2.6.2 Classifying Attributes for Interpersonal Analysis

In this step, attributes are submitted to a content analysis (Hsieh and Shannon, 2005)
where key concepts, i.e. distinct categories of attributes, emerge from the data and
attributes are subsequently classified into one of the elicited categories. A total of