Graphics Reference
In-Depth Information
Figure . . Two-dimensional mosaicplot of the Letter Recognition data. Due to the same-bin-size
representation, the pattern for the highlighted cases is clearly visible
ier to compare the highlighted heights of the upper and lower rows; we only need to
compare the positions on identical, although unaligned, scales in the same-bin-size
display.
Figure . contains a mosaicplot of the Letters Recognition data (Freyand Slate,
). Cases with an x bar value of at least are highlighted. he mosaicplot shows
all of the letters in the alphabet fromlet to right, classified according to five different
classesofwidth.hetopfourrowsallshowthesamepatternforthehighlightedcases;
theconditional probabilities areincreased forthefive lettersD,H,M,NandU.Balso
shows increased proportions of highlighting in two of the rows.
he second application of the same-bin-size variation does not deal with addi-
tional information, but rather with missing information. If we have a sparse data
cube, we can check for cells devoid of data. he goal now is to provide methods that
provide a quick overview of the empty cells in the data set, and thus to deduce the
empty-bin pattern within a dataset, which can help us to answer questions like:
. How many empty bins are there?
. Where do they occur?
. Isthereanyrecognizablepatterntotheiroccurrence,ordotheyoccurcompletely
at random?
he order of variables is important, especially when looking for patterns and groups
of combinations. Reordering the variables and collapsing over empty combinations
can help.
Mondrian Displays
By default, mosaicplots are space-filling, since only thin spaces are present between
the tiles. If these spaces are omitted we obtain another variant of a mosaicplot: the
Mondrian display. While Mondrian displays are even more space-e cient than the
Search WWH ::




Custom Search