Graphics Reference
In-Depth Information
scatterplots of nvariables inatable sothatallthe n(n-
) orderedcombinations ofaxes
are present, the eye can quickly scan a row or column and see how a given variable
depends on every other variable. his useful arrangement technique is enhanced by
the use of a brush as described in Sect.
.
. As in typical later scatterplot brushing
tools, the data points brushed over are painted in adifferent colour, both in the panel
in which the brush is active and in all other panels of the matrix. In our terminology,
thebrushisthemechanism that creates thedegree ofinterest “variable” that links the
scatterplot data views.
One of the reasons this technique is effective is that in each linked view, there is
a one-to-one correspondence between cases of the data matrix and graphical repre-
sentations of these cases, so that in each scatterplot we have complete freedom as to
what colour or glyph to use to represent this data item. Linking scatterplots do not
require considerations of aggregation; each data row maps directly to a single glyph
in the scatterplot. his simplicity can be seen in Figs.
.
and
.
, where we selected
bars and linked to the scatterplot; if we had reversed the direction of interaction, we
would have been let with bars that were partially selected. Deciding how to display
such views requires some thought and will be discussed in Sect.
.
.
Even when restricted to data views that display distinct graphical elements for
each case, linking is a powerful tool. An example of a successful tool in this area is
XGobi (Swayne et al.
).XGobi is an X-Windows-based tool that presents the user
with several glyph-based views (dotplots, scatterplots, rotating plots, grand tour,and
projection pursuit tours) and uses brushing to link the views along the above lines.
he latest incarnation of this sotware is GGobi (Swayne et al.
).
Scatterplots and dotplots are the most obvious examples of unaggregated views.
Raw tables of data can also be considered unaggregated. One useful technique is to
show a table of only those data that have been selected. his is oten termed a “drill-
down” view but is actually a simple example of linked views. Selection in one view
leads to a second view where the “visibility” aesthetic is used to code the degree of
interest - only the selected rows are shown.
Parallel coordinates views are a relatively novel form of view introduced by Insel-
berg (
). hey work well only for relatively small numbers of cases as they show
an n-dimensional point as a line in
-D, taking up a lot of display space for each row.
However, within this limitation, they are an ideal candidate for linked views, as they
areunaggregatedandencouragetheusertospotdifferencesbetween theselectedand
unselected lines.
Figure
.
below shows data for players in the
season. Even restricting our-
selves to a few thousand players makes this chart hard to read, but we can see the
higher values on each scale are in the selected group and the ones at the bottom are
in the unselected group. For this many lines it is hard to distinguish them either by
colour or by dashing style and the result is not illuminating. In Sect.
.
we will show
more compelling uses for parallel coordinates for linking to maps and hierarchical
clustering plots.
One interaction feature that is important for brushing is that of the brush mode or
paint mode. In transient mode, where the user drags the brush over the plot, when
the brush moves away from an item, the item reverts to unselected state. In additive
