Graphics Reference
In-Depth Information
plots as panel plots, which is only a technical detail and not relevant for data analy-
sis purposes). In principle the panel plot can be any arbitrary statistical graphic, but
usually nothing more complex than a scatterplot is chosen. All panel plots share the
same scale. Up to three categorical variables can be used as conditioning variables to
form rows, columns, and pages of the trellis display. To annotate the conditioning
categories of each panel plot, the so-called strip labels are plotted atop each panel
plot, listing the corresponding category names. he two remaining variables - the
so-called adjunct variables - can be coded using different glyphs and colors (if the
panel plot is a glyph-based plot).
Trellisdisplaysintroducetheconceptofshingles. Shingling is the process of di-
viding a continuous variable into (possibly overlapping) intervals in order to convert
this continuous variable into a discrete variable. Shingling is quite different to con-
ditioning with categorical variables. Overlapping shingles/intervals leads to multiple
representations of data within a trellis display, which is not the case for categorical
variables. Furthermore, it is hard to judge which intervals/cases have been chosen to
build a shingle. Trellis displays show the interval of a shingle using an interval of the
strip label. his is a solution which does not waste plotting space, but the informa-
tion on the intervals is hard to read from the strip label. Nonetheless, there is a valid
motivation for shingling, which is illustrated in Sect.
.
.
.
In Fig.
.
wefindone conditioning variable (Car Type)andoneaxisvariable(Gas
Mileage).he panel plot is a boxplot. Strip labels have been omitted as the categories
can be annotated traditionally.
An example of a more complex trellis display can be found in Fig.
.
. For the
samecarsdatasetasinFig.
.
,thescatterplotofMPG vs. Weight is plotted.hus the
panel plot is a scatterplot. he axis variables are MPG and Weight.hegridissetup
bythetwoconditioning variables Car Type along x and Drivealong y.Afithvariable
is included as adjunct variable. he Number of Cylinders is included by coloring the
points ofthe scatterplots. heupperstrip label showsthecategory of Drive,thelower
strip label that of Car Type.InFig.
.
wefindacommonproblemoftrellisdisplays.
Although the data set has almost
observations,
of the
panels are empty, and
panels have fewer than
observations.
Trellis Display vs. Mosaic Plots
6.3.2
Trellis displaysand mosaicplots donothave very muchin common.his can beseen
when comparing Figs.
.
and
.
. Obviously the panel plot is not a
-D mosaicplot,
which makes the comparison a bit di
cult. On the other hand, the current imple-
mentations of trellis displays in R do not offer mosaicplots as panel plots, either.
In Fig.
.
the interaction structure is far harder to perceive than in the original
mosaicplot. In a mosaicplot the presence of independence can be seen by a straight
crossing of the dividing gaps of the categories (in Fig.
.
the user preference and
the prior usage of product M can be regarded as independent for sot water and low
temperatures; see lower right panel in the figure). But what does independence look
like in the conditioned barchart representation of the trellis display of Fig.
.
? Two
variables in the panel plot are independent iff the ratios of all corresponding pairs of
