ADVANCING THE THEORY OF DSS DESIGN FOR USER CALIBRATION - Human-Computer Interaction and Management Information Systems: Foundations - page 64

Table 5.1

Analysis of Variance of User Calibration for Higher (questions 1-4) and Lower Problem
Novelty (questions 5-10)

a. ANOVA Results of User Calibration by Linguistic-Verbal and Spatial-Visual for Higher Problem Novelty
(questions 1-4)

Source d.f. Type III SS F-value p-value

Expressiveness/Visibility 1 .073 5.232 .028*

Error 38 .528

Corrected Total 39 .601

R 2 . 121

*p .05

b. ANOVA Results of User Calibration by Linguistic-Verbal and Spatial-Visual for Lower Problem Novelty
(questions 5-10)

Source d.f. Type III SS F-value p-value

Expressiveness/Visibility 1 .002 .229 .635

Error 38 .405

Corrected Total 39 .407

R 2

. 006

and the independent variable of DSS locus of design, either expressiveness or visibility, for each
subject. This model produced a Wilks Lambda treatment effect of F (2,37) 2.8, p 0.07. Although

not significant at the 0.05 level, this result suggests significant univariate effects. Indeed, in

the case of strong positive correlation between the dependent variables (r

0.0031), and
interaction consistent with that hypothesized in Figure 5.3, the multivariate test is less powerful
than it would be if the data were negatively correlated. In other words, the Wilks Lambda F-value
may be confounded by the nature of the interaction between dependent variables.
To clarify the MANOVA results, analysis of variance (ANOVA) was computed for the higher
and lower groupings of problem novelty separately. The results of these analyses are presented in
Table 5.1.
The first ANOVA, Table 5.1a, shows results for data from the higher category of problem nov-
elty. Subjects using the visibility (V) treatment were much better calibrated (0.078), had calibra-
tion scores closer to zero, than were those assigned to the expressiveness (E) treatment (0.163).
These data show that subjects exposed to visibility produced user calibration that was signif-
icantly better than those subjects exposed to expressiveness (F (2,37)

0.45, p

0.028). The
Bonferroni minimum significant difference of 0.0755 confirms that the difference between 0.163
and 0.078 is significant at the

5.23, p

0.05 level. For this data, H 0 can be rejected. The evidence
shows that for the higher category of problem novelty (i.e., when the problems were the most
novel), the average calibration of subjects using visibility diagrams was significantly better than
it was for those subjects using expressive text.
The results of the ANOVA for the lower category of problem novelty are presented in Table 5.1b.
There seems to be little difference between the average calibration of those exposed to expressiveness
(E) (0.116) and those exposed to visibility (V) (0.100). These data show no significant difference in