Graphics Reference
In-Depth Information
Figure
.
.
Profile, glyphs, and pie glyphs of a subset of data regarding five economic indicators, as
generated with SpiralGlyphics (Ward and Lipchak,
). Features within and between glyphs are
generally easier to compare with profile glyphs
cal attributes), thereisan inherent difference inourability toextract values from
different data dimensions.
Proximity-based bias
In most,if not all, glyphs, relationships between data dimen-
sionsmappedtoadjacent features inaglyphareeasiertoperceiveandremember
than those mappedtononadjacent features. Tothebestof myknowledge,noone
hasperformedexperimentstoquantifythedegreeofthisbias,althoughChernoff
andRizvi(
)reportedasmuchas
%variance inresultsbyrearranging data
mappings within Chernoff faces. It is likely that the amount of bias will depend
as well on the type of glyph used, as comparing lengths of bars with a common
baseline will be easier than comparing the lengths of rays in a star glyph.
Grouping-based bias
Graphical attributes that are not adjacent but may be seman-
tically or perceptually grouped may result in the introduction of bias as well.For
example, if we map two variables to the size of the ears in a face, the relationship
between those variables may be easier to discern than, say, mapping one to the
shape of the eye and the other to the size of the adjacent ear.
Ordering of Data Dimensions/Variables
7.6
Each dimension of a data set will map to a specific graphical attribute. By modify-
ing the order of dimensions while preserving the type of mapping, we will generate
alternate “views” of the data. However, barring symmetries, there are N! different
dimension orderings, and thus distinct views. An important issue in using glyphs is
to ascertain which ordering(s) will be most supportive of the task at hand. In this
section, I will present a number of dimension-ordering strategies that can be used
to generate views that are more likely to provide more information than random
ordering.
