Graphics Reference
In-Depth Information
lowest production are highlighted in light green across all maps. For acreage,
% of
the states (
out of
) fall into the let column,
% (
out of
) into the middle
column, and
% (
out of
) into the right column. Similarly, for yield,
% of the
states (
out of
) fall into the top row,
% (
out of
) into the middle row, and
% (
out of
) into the bottom row. hus, the four states highlighted in dark green
(i.e.,Indiana,Illinois,Iowa,andMinnesota)inthetoprightmapbelongtothe
%of
states with the highest production,
% of states with the highest acreage, and
%
of states with the highest yield. Wisconsin, one of the spatial outliers identified in
Sect.
.
,is one of the states highlighted in medium green in the top central map and,
thus, belongs to the
% of states with the medium production,
% of states with
the medium acreage, and
% of states with the highest yield.
Most of the remaining numbers by the sliders indicate the class boundaries with
the units communicated in the slider label. he top and bottom sliders have tails on
theright.hisreflectsthepresenceofverylargevaluesrelativetothebodyofthedata.
he lowest row of labels by each slider gives the minimum value, the upper adjacent
value from a box plot calculation (Cleveland,
), and the maximum value. he
slider scale changes more quickly over the tail. his leaves space for greater slider
resolution in the body of the data.
he values in the upper right corner of each map show the weighted means for
production for the states highlighted in those maps. Note that no state is highlighted
in the bottom right map and,thus, no weighted mean isavailable. hefitting of these
means to state values for the states highlighted in the maps leads to the R-squared
valueatthebottomrightofFig.
.
.
he CCmaps sotware has several other features such as enlarged views of indi-
vidual maps and the possibility to obtain specific values when the mouse is moved
over a particular region. he sotware also provides zoom features to focus on a por-
tion of a map and conditioned scatterplot smoothers. It is freely available at http:
//www.galaxy.gmu.edu/~dcarr/ccmaps.
So,finally,howdoLMplotsandCCmapscompare?First,theencodingofvaluesin
the Fig.
.
LM plots retains much more detail. he CCmaps encoding of three vari-
ables each into three classes loses much detail. Second, LM plots can include many
additional statistics. For example, Fig.
.
showstwo values fordifferent time periods
inthefirststatistical panel,rateswitha
%confidenceinterval andareferencevalue
inthesecondstatistical panel,andtheboxplots distributional summaries inthethird
statistical panel.CCmapsconvey justthree values perregion.Carrying LMplotsabit
further, Carr et al. (
) illustrate sorting on one dot plot encoded variable, using
this as the independent variable to obtain a smoothed fit to another variable. A sec-
ond LM plots panel then shows observed and fitted values as well as the smooth in
avertical rank-repositioned smooth. hiscanhelpinseeing afunctional relationship
and large residuals. CCmaps does provide a separate view to show one-way dynam-
ically conditioned smoothers. Since a similar separate view could be added to LM
plots, this is not really a plus for CCmaps.
he CCmaps sotware does have a few merits. First, CCmaps scale better to maps
involving moreregions. FairlycommonCCmapsexamples showover
UScoun-
ties on the screen. he NCI State Cancer Profiles Web version of LM plots discussed
