Graphics Reference
In-Depth Information
Evaluation based on the scalability of techniques in terms of number of records
and dimensions.
A large number of evaluation studies on glyphs and other forms of multivariate data
analysis have been carried out over the years. Some of these have been rather ad hoc,
or based just on the opinions and observations of the authors, while others have in-
volved detailed and carefully orchestrated user studies.
Cluff et al. (
) evaluated and categorized
methods of multivariate data pre-
sentation, including several forms of glyphs. heir evaluation criteria fell into three
groups:objectives, information levelanddimensioncapacity,andglobalcriteria.Un-
dertheobjectivesgrouptheyconsideredaccuracy,simplicity,clarity,appearance,and
design.Intheinformationlevelanddimensionalcapacitycriterion,visualizationsare
classified as towhat extent they retain the level of information present in the data. At
the lowest level (elementary) is the ability to convey the individual data values, while
at the intermediate level, relationships among subsets of the data can be seen. he
highest level (overall) provides linkages between multiple relationships and allows
users to understand the data su
ciently to solve real tasks. he global criteria group
includes flexibility, interpretability, visual impact, time to mastery, and computa-
tional tractability. For each visualization method, each of these criteria was classified
(subjectively bythe authors) as
=does not su
ciently meet objective,
=meets ob-
jective satisfactorily, and
=meetsobjective inanexcellentmanner.Whiletheresults
of the classifications lack statistical significance due to the small sample size, the cat-
egorization of evaluation criteria was far more extensive than in most other studies.
Leeetal.(
)analyzed theeffectiveness oftwoglyphtechniques(Chernofffaces
and star glyphs) and two spatial mappings, where each data record was simply rep-
resented by a marker whose position was based on similarity to other data records.
Binary data was used, and
subjects were asked a range of questions regarding re-
lationships between records (both local and global). Results showed that the sub-
jects answered many of the questions more quickly and accurately, and with more
confidence, using the spatial mappings. his confirmed the hypothesis of many re-
searchers, which is that glyph interpretation can be quite slow for tasks that involve
a significant scanning and comparison. However, questions regarding the values of
particular data features couldnotreadily beanswered with the spatial mappings. he
implication is that, given a known set of questions, it may be possible to assign posi-
tions of simple points to facilitate a task. For general tasks, however, a combination
involving positioning of glyphsbased on data relationships, as suggestedin theglyph
layout section of this paper, would likely be most effective.
As mentioned earlier, Borg and Staufenbiel (
) compared snowflake and star
glyphs with factorial suns, with the angles of the lines conveying relationships be-
tweendimensions. Intheirexperiment,theyusedtheclassification of
prototypical
psychiatric patients across
attributes into
categories, asdetermined by
experts.
hen,usingeachofthe
glyphtypes,theygenerateddrawingsofeachofthe
cases.
hirty beginning psychology students were asked to group the drawings into
cate-
gories based on shape similarity. hey then studied the frequency with which draw-
ings of the same category (according to the experts) were grouped together by the
