Analyzing Personal Attribute Judgments - Modeling Users' Experiences with Interactive Systems - page 51

simpler (i.e., average) model and the AIC of the more complicated one (i.e., con-
sisting of three sub-models). Δ i ≤ 2 provides significant evidence for the simpler

model, 4 ≤ Δ i ≤ 7 provides weak support for the heterogeneous model, while Δ i

≥

10 , providing significant evidence that the diverse models, despite the increase in
the model complexity, perform better than the average model.
Figure 3.4 illustrates the insights gained by the average and the three diverse
models that are derived from the views corresponding to clusters 1 through 3. A
significant overlap exists between models 1 and 3 (five common attributes), while
model 2 provides a completely different view.

10 provides strong support for the heterogeneous model. In our case Δ i = 354

Average Model Model 1

Aachen

Munchen

Aachen

Graphical layout

Munchen Fast access to

information

Fast access
to information

Mainz

Supports Search

Mannheim

Supports search

Frankfurt

Heidelberg

Darmstadt

Karlsruhe

Darmstadt

Karlsruhe

Heidelberg

Mainz
Frankfurt

Mannheim

Refers to student life

Model 2 Model 3

Graphical layout

Munchen

Emphasis on
achievement

Aachen
Karlsruhe

Fast access
to information

Heidelberg

Munchen

Frankfurt

Darmstadt

Mainz

Frankfurt

Aachen

Supports search

Mannheim Heidelberg

Adequate information

Karlsruhe

Mainz

Darmstadt Mannheim

Fig. 3.4 The average model and the three diverse models corresponding to clusters 1 to 3 in
figure 3.3.