Graphics Reference
In-Depth Information
Figure
.
.
hree histograms for the same variable. he two plots on the right have the same frame size,
but different scales. he plot in the top right usesthesamescaleastheplotonthelet. he differences
in the visual impact are only due to the different uses of frame size or scales
histogram definition in Manet specifies five parameters, the lower and upper limit
of the horizontal scale, the bin width, the number of bins, and the maximum height
of a bin. Any two of the first four parameters combined with the fith one are su
-
cient foracomplete specification ofahistogram. Vialinking all these parameterscan
be shared. Moreover, the user has the choice of adjusting the frame size as well. For
a proper comparison, it is essential to also use the same frame size and not only the
same scales. Figure
.
shows three histograms for the same variable. he plot on the
letistheactiveone,theplotonthebottomrightisnotlinkedatall,andtheoneonthe
toprighthasthe samescales astheactive one,butadifferent framesize.hesesimple
examples of linked scales point toward the importance of linking scale information
in general. A very widespread use of linking scales is in the form of sliders. Sliders
are
-D graphical representations of model parameters, which the user can change
dynamically. Moving the slider yields a change of the underlying model parameter
that is then automatically propagated to all plots that display the model. Sliders are
a widely used tool to provide a visual representation for dynamic queries (Shneider-
man,
) to filter and dissect the data into manageable pieces following the visu-
alization mantra: overview first, zoom, and filter, then details on demand (Shneider-
man,
).Another common application for such slidersis interactively controlling
Box-Box transformations or checking various lags for time-series analyses.
he order of the categorization values is of less importance for continuous data
displayed in a histogram. his scale component becomes more important for nomi-
nal categories that do not have a clearly defined natural ordering. Linking this scale
parameter is common and helpful for barcharts and mosaicplots. For the histogram,
the categorization vector C actually belongs to both the observation component and
the scale component. Using the same categorization vector could thus also be seen
as a form of linking the observations of a model. In general, however, it is more ap-
propriate to restrict the concept of linking observations to the variables that are used
in the model and not the categorization vector. In this sense, linking models means
that plots share the same variables. In Fig.
.
all three plots are in principle linked
via the model observations because all three plots represent the same variable. In
this static form,this kind oflinking doesnot really provideaparticular achievement.
However,using this formof linking inaprespecified clusterof graphical displayscan
give a rather complete picture of a dataset. Young et al. (
) have created a system
