Graphics Reference
In-Depth Information
bounds and should be layered within the foreground. Grid lines always belong in
the background. here are many factors related to statistical graphics quality, start-
ingwithcontent andcontext, andextendingtoadditional visual considerations, such
as perceptual grouping and sorting. While specific instances of LM plots may pro-
videopportunities forimprovement, theparadigmcanincorporatemuchknowledge
about quality statistical graphics design.
LM plots oten provide a good alternative to displaying statistical information us-
ingchoroplethmaps.Choroplethmapsusethecolororshadingofregionsinamapto
represent region values. Choropleth maps have proved very popular but have many
problems and limitations as indicated by writers such as Robinson et al. (
), Dent
(
),and Harris (
).Reviewing these problems helps to indicate why LM plots
are a good alternative.
here are two kinds of choropleth maps, called unclassed and classed. Unclassed
maps use a continuous color scale to encode continuous values (statistics). his is
problematicbecause perceptionofcolorisrelative toneighboringcolorsandbecause
color has poor perceptual accuracy of extraction in a continuous context. Classed
choropleth maps ameliorate this problem and dominate in the literature.
Classed choropleth maps use class intervals to convert continuous estimates into
an ordered variable with a few values that can be represented using a few colors.
Whenafewcolors areeasily discriminated and regions aresu
ciently large forcolor
perception, color identification problems are minimal. he color scheme also needs
to convey the class ordering based on values. Brewer (
) and Brewer et al. (
)
provided results evaluating different color schemes in a mapping context. he Web
site http://colorbrewer.org (see Leslie,
, for a short description) contains guid-
ance on ordered color schemes and additional issues such as suitable schemes for
people with color vision deficiencies and for different media. Perfect examples on
how colors should be used in choropleth maps can be found in the
“Atlas of
United States Mortality” (Pickle et al.,
).
Even with a good color scheme,three keyproblems remain forclassed choropleth
maps. he first problem relates to region area. As suggested above, some map re-
gions can be too small to effectively show color. Examples include Washington, DC,
onamapoftheUnited States (US)andLuxembourg onaEuropean map.Mapcarica-
tures, such as Monmonier's state visibility map (Monmonier,
), can address this
problem, by enlarging small regions in a way that maintains region identifiability
and shows each region touching the actual neighboring regions. Another facet of the
area problem is that large areas have a strong visual impact while in many situations,
such as in the mapping of mortality rates, the interpretation should be weighted by
the region population. Dorling (
) addressed this problem by constructing car-
tograms that changed region shapes to make areas proportional to population. Is-
sues related to this approach are region identifiability, and map instability over time
as their shapes change with changing populations. Area related problems persist in
choropleth maps.
Asecondkeyproblemisthat converting acontinuous variable intoavariable with
a few ordered values results in an immediate loss of information. his loss includes
the relative ranks of regions whose distinct values become encoded with the same
