Graphics Reference
In-Depth Information
Figure . . Examples of multivariate glyphs (from Ward, )
helistabove isample evidence that asignificant number ofpossiblemappings exist,
many of which have yet to be proposed or evaluated. he question then becomes
determining which mapping will best suit the purposeof the task, the characteristics
of the data, and the knowledge and perceptual abilities of the user. hese issues are
described in the sections below.
Biases in Glyph Mappings
7.5
One of the most common criticisms of data glyphs is that there is an implicit bias
in most mappings, i.e., some attributes or relationships between attributes are easier
to perceive than others. For example, in profile or star glyphs, relationships between
adjacent dimensions are much easier to measure than those that are more separated,
andinChernofffaces,attributessuchasthelengthofthemouthornoseareperceived
more accurately than graphical attributes such as curvature or radius.
In this section I attempt to isolate and categorize some of these biases, using both
results from prior studies on graphical perception as well as our own empirical stud-
ies. It is clear, however, that much more substantial work is needed in measuring and
correcting for these biases when designing and utilizing glyphs in data analysis.
Perception-based bias Certain graphical attributes are easier to measure and com-
pare visually than others. For example, Cleveland ( )reports on experiments
that show length along a common axis can be gauged more accurately than, say,
angle, orientation, size, or color. Figure . shows the same data with three differ-
ent mappings. Relationships are easier to see with the profile glyphs (length on
a commonbase),followed bythe star glyphs (length with different orientations).
he pie glyph fares the worst, as the user is required to compare angles. hus in
mappings that are not many-to-one (i.e., those that employ a mixture of graphi-
Search WWH ::




Custom Search