Graphics Reference
In-Depth Information
appearance substantially. As with many exploratory techniques, the user should ex-
periment with different settings until he or she feels comfortable enough with the
insight gained.
Sorting and Scaling Issues
6.4.3
Parallel coordinateplots areespecially usefulforvariables whicheither haveanorder
such as time or all share a common scale. In these cases, scaling and sorting issues
are very important for a successful exploration of the data set.
Sorting
Sorting in parallel coordinate plots is crucial for the interpretation of the plots, as
interesting patterns are usually revealed at neighboring variables. In a parallel coor-
dinate plot of k variables, only k
adjacencies can be investigated without reorder-
ing the plot. he default order of a parallel coordinate plot is usually the sequence
in which the variables are passed to the plotting routine, in most cases the sequence
in the data file itself. In many situations this order is more or less arbitrary. Fortu-
nately, one only needs
k +
different orderings to see all adjacencies of k variables
(see Wegman, ).
Whenever all, or at least groups of, variables share the same scale, it is even more
helpful to be able to sort these variables according to some criterion. his can be
statistics of the variables (either all cases or just a selected subgroup) like minimum,
mean, range, or standard deviation, the result of a multivariate procedure, or even
some external information. Sorting axes can reduce the visual clutter of a parallel
coordinate plot substantially.
If data sets are not small, sorting options have to be provided both manually and
automatically.
Scalings
Besides the default scaling, which is to plot all values over the full range of each axis
between the minimum and the maximum of the variable, several other scalings are
useful. he most important scaling option is to either individually scale the axes or
to use a common scale over all axes. Other scaling options define the alignment of
the values, which can be aligned at:
he mean
he median
A specific case
Aspecificvalue
For an aligned display, it isnot obvious what the range of the data should be whenan
individual scale is chosen. For individual scales, a σ scaling is usually a good choice
to map the data onto the plot area.
Alignments do not force a common scale for the variables. Common scaling and
alignments are independent scaling options.
Figure . shows a parallel coordinate plot for the individual stage times of the
cyclistswhofinishedthe TourdeFrancebicyclerace.Intheupperplotwe
Search WWH ::




Custom Search