Graphics Reference
In-Depth Information
between the visit frequency tiles to
. Clearly, the proportion of these patients in-
creases with LOS. In fact, such a “stacked” plot can also be interpreted as a spine plot
with highlighting (Hummel,
), which is particularly useful when analyzing rela-
tionships among categorical data with a binary dependent variable. More variations
on this theme, such as doubledecker plots, are treated in Chapter III-
.
Using colors, even more complementary information can be visualized, either by
adding additional information, or by redundantly coding information already visu-
alized bythe “raw”plottosupportour perceptualsystem. First,weconsider the sieve
plots. hedensity of the gridin the rawversion implicitly gives us an idea of the sizes
of the residuals, butsince the plot doesnot includethe density correspondingtozero
residuals (the null model) for comparison, we cannot easily assess whether there are
more or fewer counts in a cell than expected under the null hypothesis. Using color,
we can add the sign information; for example, we can use blue for positive, red for
negative, and gray for zero residuals.
he simplest mosaicplots are monochrome displays. Friendly (
) introduced
aresidual-basedshadingofthetilestoadditionallyvisualizetheresidualsfromagiven
independencemodelfittedtothetable.heideaistouseacolorcodingforthemosaic
tiles that visualizes the sign and absolute size of each residual r
ij
. Cells correspond-
ing to small residuals (
r
ij
<
) have no color. Cells with medium-sized residuals
(
) are shaded light blue and light red for positive and negative residuals,
respectively. Cells with large residuals (
r
ij
<
) are shaded with fully saturated blue
and red, respectively. he heuristic for choosing the cut-offs
and
is that the Pear-
son residuals are asymptotically standard normal, whichimplies that the highlighted
cells are those with residuals that are individually significant at approximately the
r
ij
Figure
.
.
Mosaic display with Friendly-like color coding of the residuals
