Graphics Reference
In-Depth Information
Figure
.
.
Multiscale visualization methods of minimum temperature data. Darker regions of the
mode forest represent higher-confidence “bumps,” while light, medium,anddark regions on the sizer
plot represent, respectively, significantly positive, nonsignificant, and significantly negative slopes in
“scale space”
though thereis suggestion ofa morecomplicated structureas well.hefull complex-
ity of the mode behavior may be seen in the basic mode tree, while the filtered mode
tree emphasizes the two key modes as found in its underlying mixture model. he
dark regions of the mode forest perform implicit inference as they indicate bumps
found in many or most of the subsamples of the full data, while the SiZer plot makes
the inference explicit, with alternating light and dark regions strongly suggesting the
presence of underlying modes.
Bivariate Density Estimates
5.2
he basic tool for graphically exploring the bivariate distribution between pairs of
measurements is the well-known scatterplot. A scatterplot is constructed by simply
plotting the pairs of points in the coordinate plane. Generally, some functional rela-
tionship between the variable represented by the y-axis and the variable represented
by the x-axis is not assumed. However, it is precisely the power of the scatterplot to
visualize such relationships that makes it so useful a tool for data analysis.
Friendly and Dennis (
) present an excellent treatise on the scatterplot and
note that early authors recognized the need to augment the scatterplot toaccount for
repeated values in the data, such as glyphs and different plot characters to indicate
a multiplicity of points with the same observed value. Cleveland and McGill (
),
Scott (
), and others have also noted this and many of the modifications and en-
hancements that have been proposed to convey additional information in scatter-
