Graphics Reference
In-Depth Information
One of the biggest challenges in data visualization is to find general representations
of data that can display the multivariate structure of more than two variables. Sev-
eral graphic types like mosaicplots, parallel coordinate plots, trellis displays, and the
grand tour have been developed over the course of the last three decades. Each of
these plots is introduced in a specific chapter of this handbook.
his chapter will concentrate on investigating the strengths and weaknesses of
these plots and techniques and contrast them in the light of data analysis problems.
One very important issue is the aspect of interactivity. Except for trellis displays,
all the above plots need interactive features to rise to their full power. Some, like the
grand tour, are only defined by using dynamic graphics.
Introduction
6.1
It is sometimes hard to resist the problem that is captured in the phrase “if all you
have is a hammer, every problem looks like a nail.” his obviously also holds true for
theuseofgraphics. Agrandtourexpertwillmostlikely includeacategorical variable
in the high-dimensional scatterplot, whereas an expert on mosaicplots probably will
try to fit a data problem as far as possible into a categorical framework.
hischapterwillfocusontheappropriateuseofthedifferentplotsforhigh-dimen-
sional data analysis problems and contrast them by emphasizing their strengths and
weaknesses.
Data visualization can roughly be categorized into two applications:
.
Exploration
In the exploration phase, the data analyst will use many graphics that are mostly
unsuitable for presentation purposes yet may reveal very interesting and impor-
tant features. he amount of interaction needed during exploration is very high.
Plotsmustbecreated fastand modifications like sorting or rescaling should hap-
pen instantaneously so as not to interrupt the line of thought of the analyst.
.
Presentation
Once the key findings in a data set have been explored, these findings must be
presented to a broader audience interested in the data set. hese graphics oten
cannot be interactive but must be suitable for printed reproduction. Further-
more, some of the graphics for high-dimensional data are all but trivial to read
without prior training, and thus probably not well suited for presentation pur-
poses - especially if the audience is not well trained in statistics.
Obviously, the amount of interactivity used is the major dimension to discriminate
between exploratory graphics and presentation graphics.
Interactive linked highlighting, as described by Wills (
, Chapter II.
same
volume), is one of the keys to the right use of graphics for high-dimensional data.
Linking across different graphs can increase the dimensionality beyond the number
of dimensions captured in asingle multivariate graphic. hus, the analyst can choose
