SOLVING GEOSPATIAL TASKS IN 2D AND 3D VISUALIZATIONS OF MAPS - Information Visualisation: Techniques, Usability and Evaluation - page 239

null hypothesis (H 0 :r s >=0.0) in the 55 ͼ condition (p=0.0164) and the 65 ͼ
condition (p<0.001), indicating significant negative association in those
conditions. Fig. 12.6 shows scatterplots of accuracy versus RD, as well a
fitted linear model for the three tilt angles studied. Extrapolation using the
coefficients of the linear models indicates that 100% accuracy would be
expected for a level of difficulty RD of roughly 0.7 (RD=0.68 in the 45 ͼ
condition, RD=0.67 in the 55 ͼ condition, and RD=0.69 in the 65 ͼ
condition). In regard to average response times, some positive trends in
terms of increasing times along with increased degrees of difficulty can be
expected in the two mostly slanted maps (see Fig. 12.7). But correlation
tests under the null hypothesis H 0 : r s =0.0 at a significance level of 0.05 do
not allow us to reject H 0 . Thus, no significant positive or negative
associations between response times and the degree of difficulty RD can
be claimed.
Discussion
An interesting finding of this first experiment is that users' performance in
terms of task completion times was not affected by the slant angle and that
accuracy in solving the task was reduced only when the slant angle
exceeded 55 ͼ (i.e., in the 65 ͼ condition). This threshold is clearly higher
than the 35 ͼ that had been reported in similar experiments [12]. In contrast
to that previous study, the experiment presented in this chapter used strong
3D cues which may explain maintained accuracy, although there is
considerable foreshortening of the map at a 55 ͼ slant angle due to the
oblique projection. As is evident in Fig. 12.4 (b), quite some screen estate
is made available at a 55 ͼ slant angle due to the oblique projection of the
map. This available screen space could be used to accommodate other
graphical elements into the visualization with the same window size. The
results of the experiment also confirmed the assumption that for the
correct assessment of arbitrarily oriented distances on maps, their relative
lengths matter. The correctness in determining a shortest distance on a
map depends significantly on how big the difference in length is in relation
to the second shortest distance among all candidates. Based on the
observations in the experiment, and using a linear model, it can be
predicted that the difference in a large number of arbitrarily oriented
distances has to be at least 30% in order to assure reliable identification.