Graphics Reference
In-Depth Information
Figure
.
.
Effect of h on kernel density estimates of a subset of the minimum temperature data
better estimates, the results can be unstable and should not replace the examination
of multiple estimates by the skilled analyst.
One advantage kernel estimates have over simpler density estimates such as his-
togramsistheeasewithwhichtheycanbecomparedacrossdifferentdatasets. Kernel
density estimates are continuous and hence are well suited to overplotting with dif-
ferentcolorsorlinetypes,whichmakes visual comparisons between different groups
simple and effective.
An example is shown in Fig.
.
. he black curve represents a kernel density es-
timate of the temperature midpoint from the PRISM data for grid cells in Colorado
using a normal kernel and normal reference bandwidth. he grey curve represents
asimilarlyconstructedkernelestimateoffthetemperaturemidpointforthegridcells
in the neighboring state of Kansas. he bimodality in the kernel estimate for Col-
orado is attributable to the differences between the eastern plains and the mountain
Figure
.
.
Univariate kernel density estimates of average temperature midpoint based on the PRISM
data. Black line: grid points in Colorado; grey line:Kansas
