Graphics Reference
In-Depth Information
we clearly see that too many (too few) patients that stay fromtwo to nine (more than
) years get visited regularly than would be expected under the null hypothesis of
independence, and that this pattern is reversed for patients visited less than monthly
or that are never visited.
Summary
12.2.4
Mosaic plots are a toolforvisualizing the observed frequencies of acontingency table
based on recursive conditional splits. If one variable is explanatory, it should be used
first for splitting; the display then shows the conditional distribution of the depen-
dent variable given the explanatory one. Sieve plots basically visualize the table of
expected frequencies, and in addition the deviations from the observed frequencies
by the density of the grid added to each tile. hey complement mosaicplots by de-
tecting dependency patterns for ordinal variables. An alternative way of enhancing
mosaicplots to display deviations from expected frequencies is to use residual-based
shadings (see the next section), which are typically more intelligible than sieve plots,
in particular for nominal variables. Association plots directly visualize Pearson and
raw residuals, i.e., standardized and nonstandardized deviations of observed from
expected frequencies, respectively. hese plots should be used if the diagnostics of
independence models are of primary interest.
Using Colors for Residual-Based Shadings
12.3
Asintroduced intheprevious section forassociation plots, theinvestigation ofresid-
uals from a posited independence model is of major interest when analyzing contin-
gencytables.Inthefollowing,wewilldemonstratehowtheuseofcolorscangreatly
facilitate the detection of interesting patterns. We start with some general remarks
on colors and color palettes.
A Note on Colors and Color Palettes
12.3.1
he plots introduced in the previous section are basically composed of tiles whose
areas represent characteristics derived from the contingency tables - observed and
expected frequencies in the case of mosaic and sieve plots, or residuals as visualized
by association plots. When using color for these tiles, it is imperative to choose the
right color palettes, derived from suitable color spaces. Apart from aesthetic consid-
erations, wrongly chosen colors might seriously affect the analysis. For example,
Using high-chroma colors for large areas tends to produce ater-image effects
which can be distracting (Ihaka,
).
heprintedversionofthearticleismonochrome,buttheelectronicversionusescolors.he
colored versions of the plots are available from the Web page http://statmath.wu-wien.ac.
at/projects/vcd/.
