Graphics Reference
In-Depth Information
value. he task of controlling conversion loss has spawned numerous papers about
proposingmethods fordefining class intervals. Today,guidance isavailable based on
usability studies. Brewer and Pickle ( ) indicated that quintile classes (roughly
%ofregionsineachclass)tendtoperformbetterthanotherclassintervalmethods
when evaluated across three different map-reading tasks. Still, the best of the class
interval selection approaches loses information.
he third key problem is that it is di cult to show more than one variable in
a choropleth map. MacEachren et al. ( , ) were able to clearly communi-
cate values of a second binary variable (and indicator of estimate reliability) by plot-
ting black-and-white stripped texture on regions with uncertain estimates. However,
moregeneral attempts such as using bivariate colors schemes have been less success-
ful(WainerandFrancolini, ).hus,choroplethmapsarenotsuitableforshowing
estimate standard errors and confidence bounds that result from the application of
sound statistical sampling or description. It is possible to use more than one map to
show additional variables. However, Monmonier ( , p. ) observed that when
plottingchoroplethmapssidebysideitcaneasilyhappenthat“similarityamonglarge
areas can distort visual estimates of correlation by masking significant dissimilarity
among small areas.” he change blindness (Palmer, , p. ) that occurs as the
human eyes jump from one map to another makes it di cult to become aware of all
the differences that exist in multiple choropleth maps and hard to mentally integrate
information in a multivariate context.
Carr proposed the use of micromaps and the LM plots design in response to An-
thony R. Olsen's [from the US Environmental Protection Agency (EPA), Corvallis,
OR] challenge to extend traditional statistical graphics called row-labeled plots in
Carr ( ) to include geospatial context. As indicated earlier, traditional statistical
graphicsotenusepositionalongscaleencodingsforcontinuousvaluesandcanread-
ilyshowestimatesandconfidenceintervalstogether.Olsen,Carr,Courbois,andPier-
son unveiled this new design with a
footposterat JSM.hemapregions
were Omernik ecoregions for the continental US. he various ecoregion barplots
and boxplots summarized detailed elevation information and detailed land class in-
formation derived frommultipleadvanced high-resolution radiometer(AVHRR)re-
mote sensing images over time.
he first example in the literature of LM plots (Carr and Pierson, ) showed
state values for the US. his example adapted the state visibility map of Monmonier
toaddressthevisibility problemsforsmallregions.hatpaperpresentedamicromap
plot of unemployment by state with two data columns (unemployment rate with
% confidence interval and total number of unemployed persons). he LM plots
paradigmsupportsthedisplayofdifferentkindsofstatistical panels,suchasdotplots,
barplots, boxplots, binned scatterplots, time series plots, and plots with confidence
bounds. In particular, Carr et al. ( ) presented three micromap plots: (i) CO
emissionsintheOrganizationforEconomicCooperationandDevelopment(OECD)
states with one data column (annual time series), (ii) wheat prices by state with two
data columns (average price and monthly time series), and (iii) an improved version
of the micromap plot from Carr and Pierson ( ). Carr et al. ( a) presented
four micromap plots based on the Omernik ecoregions: (i) three data columns with
Search WWH ::




Custom Search