Graphics Reference
In-Depth Information
herefore, pivot tables have the same structure as mosaicplots, but allow different
summaries from mosaicplots. A pivot table's structure carries slightly less informa-
tion than a mosaicplot, as the order of the variables in rows and columns is stored
separately. A structure of the form X
, X
′
, X
, X
′
,... isstoredintwoseparatelists:
alistofrowvariablesX
, X
,... andalistofcolumnvariablesX
, X
,... (together
with the order of the variables in each list). his means that a pivot table is unable
to distinguish between the structures X
, X
′
, X
or X
, X
, X
′
or X
′
, X
, X
.Ontop
of the pure cross-tabulation, pivot tables usually provide marginal information: row
sums and column sums are given, as well as totals by variable and grand totals. his
information is incorporated into a mosaic implicitly because of the structure, but is
not readily available; since mosaics are constructed hierarchically, a low-dimensional
mosaicplot is “included”in a high-dimensional mosaicplot. Onlookers must visually
“remove” the last splits performed and recombine tiles to see the low-dimensional
mosaic - whichis only feasible in the standard form of mosaicplot. hiscorresponds
to reading Fig.
.
backwards from let to right. he use of different ways of summa-
rizing the observations corresponding to the combinations of conditional variables
mayleadtonewvariantsofmosaicplot.Eachnewsummarywillmeanachangeinthe
way that the mosaics must be interpreted, as the tile area does not then correspond
to the probability estimate for the joint distribution of the variables.
Implementations
13.5
Implementations of mosaicplots become more and more frequent - almost all sot-
ware packages now have some functionality for creating mosaicplots. Perhaps the
most notable of them is the implementation of mosaicplots in R by John Emerson
(
),as well as the fully interactive implementations in Manet (Unwin et al.,
),
Mondrian (heus,
), and KLIMT (Urbanek,
).
Treemaps were first introduced in the Macintosh sotware TreeViz (Johnson and
Shneiderman,
). hey are now more widely available in a variety of forms; for
example, fully interactive and linked treemaps can be found in the sotware pack-
age KLIMT, the “Map of the Market” (Wattenberg,
)shows a squarified treemap
of stocks available at http://www.smartmoney.com, and the Java sotware TreeMap
(Bederson et al.,
) provides a very much improved treemap version of TreeViz.
Implementations of pivot tables are widespread, although only in disguise. he
name “pivot table” seems to be used exclusively in Excel; other names for them in-
clude cross-tabulations and (multivariate) data summaries. here are various func-
tionsinRthatallowcross-tabulationsinspecificsituations,suchastable, xtabs, table,
....A unified frameworkfor cross-tabulations is provided byHadley Wickhamin his
recast package (Wickham,
).
Both mosaicplots and their extensions can benefit from each other; by accommo-
dating some of the features of the extensions, mosaics should become more applica-
ble and flexible. he extensions, on the other hand, could possibly make use of the
mosaicplot's variants.
