Basic Mapping Principles for Visualizing Cancer Data

Using Geographic Information Systems (GIS)

Cynthia A. Brewer, PhD

Abstract: Maps and other data graphics may play a role in generating ideas and hypotheses at the
beginning of a project. They are useful as part of analyses for evaluating model results and
then at the end of a project when researchers present their results and conclusions to
varied audiences, such as their local research group, decision makers, or a concerned
public. Cancer researchers are gaining skill with geographic information system (GIS)
mapping as one of their many tools and are broadening the symbolization approaches they
use for investigating and illustrating their data. A single map is one of many possible
representations of the data, so making multiple maps is often part of a complete mapping
effort. Symbol types, color choices, and data classing each affect the information revealed
by a map and are best tailored to the specific characteristics of data. Related data can be
examined in series with coordinated classing and can also be compared using multivariate
symbols that build on the basic rules of symbol design. Informative legend wording and
setting suitable map projections are also basic to skilled mapmaking.
(Am J Prev Med 2006;30(2S):S25–S36) © 2006 American Journal of Preventive Medicine

Introduction The available years of prostate cancer mortality data are

mapped (aggregations for 1950 –1994), drawing from

A
geographic information system (GIS) allows
the counts, rates per 100,000 person-years (age ad-
epidemiologists and cancer researchers to in-
justed using 1970 populations), and upper and lower
vestigate spatial patterns within their data and
bounds of 95% conference interval (CI) offered on the
understand relationships between cancer and other
site for black and white races. The map area is cropped
health, socioeconomic, and environmental variables.
to produce compact demonstration figures that can be
High-quality maps also allow researchers to present a
compared in series. The data are freely available
compelling case to others who are interested in their
through the NCI website to other mapmakers who
work. GIS is an additional tool in the exploration,
would like to work with the methods described.
analysis, and communication of cancer data, and
The basic overview of thematic mapping offered in
knowledge of the basic principles for representing data
this article has wide application in cancer and epidemi-
can help cancer researchers make the most of GIS and
the opportunities for insight it offers. ologic mapping. Other tools in GIS are also of use to
This article is structured in three sections: mapping epidemiologists, such as address geocoding and net-
methods, mapping multiple variables, and map finish- work analysis. The links between spatial statistics soft-
ing. Two common symbol types, choropleth mapping ware tools and GIS are also improving.2 The focus of
and proportional symbols, are featured, and decisions this short article, however, is limited to symbolizing
involved in making effective use of these symbols are statistical data, which is a common use of GIS. Basic
summarized. Supporting figures present maps of pros- criteria for choosing symbols to map derived values,
tate cancer data to correspond with the topic of this significance levels, model results, and smoothed rates
special issue of the American Journal of Preventive Medi- are the same as for simpler measures such as crude
cine. These maps were produced in ArcGIS (ESRI, rates. Likewise, multivariate maps that combine or
Redlands CA, version 9) with no further augmentation overlay model results with original data or related
in illustration software. The data and geography used variables can illuminate relationships between them by
for these maps are from the National Cancer Institute combining symbolization approaches.
(NCI) Cancer Mortality Maps and Graphs Website.1 Cartographers use visual tools, and epidemiologists
use statistical tools to investigate their data. This is an
From the Department of Geography, Pennsylvania State University, oversimplification, to be sure, but it seems to be a core
University Park, Pennsylvania difference in approach between the two fields and each
Address correspondence and reprint requests to: Cynthia A. could be enhanced by further use of the other’s meth-
Brewer, PhD, Department of Geography, Pennsylvania State Univer-
sity, 302 Walker Building, University Park PA, 16802-5011. E-mail: ods. The tools cartographers use to improve their visual
[email protected]. representations of data can complement epidemiolo-

Am J Prev Med 2006;30(2S) 0749-3797/06/$–see front matter S25

© 2006 American Journal of Preventive Medicine • Published by Elsevier Inc. doi:10.1016/j.amepre.2005.09.007
gists’ sophisticated adjustments for potentially spurious Symbols that show categorical differences well are
rates and small numbers. For example, cartographers color hue and symbol shape. Symbols well suited to
adjust class breaks when mapping a given data set. representing quantities are color lightness and symbol
Epidemiologists adjust the data while mapping with a size. There are a variety of other symbol characteristics
given classing algorithm without adjustment, such as (such as pattern spacing) that may be used for data
quantiles. Epidemiologists question the data; cartogra- representation, and these are organized as “visual vari-
phers question the symbolization. This contrast is exag- ables” in the cartographic literature.3 The workhorse
gerated in the hope that this brief introduction will visual variables are hue, shape, lightness, and size for
encourage epidemiologists to expand their insights the types of data common in cancer mapping, and
from data by expanding their approaches to data common symbolization methods have names: For ex-
representation. ample, choropleth maps use lightness to represent
quantitative areal data, and proportional symbols use
size to represent quantitative data at points or for areas.
Mapping Methods
A basic characteristic of cancer data that guides choice Choropleth Mapping
of a map symbol is whether categories or quantities are
recorded. Categorical differences in cancer may be Choropleth maps present areal enumeration units—
case/control or benign/malignant. They may code such as states, counties, ZIP codes, and census tracts—
race differences (Figure 1) as well as many other filled with colors that symbolize ranges in the data
socioeconomic categories. Quantitative data may be (Figure 2). In addition to choosing which type of
counts, ranks, or derived values such as rates and enumeration unit best suits your mapping goals,4 two
percentages (Figure 2). basic decisions for choropleth mapping are color selec-
tion and data classing. The decision criteria for color
and classing choices can also be applied to point
symbols, but they are presented in this article with
choropleth examples.

Color Symbols
The main goal in choosing colors for choropleth maps
is to order lightness so it parallels ordering in the data.
The simple case is light-to-dark color for low-to-high
values with a constant hue (blue is used in Figure 2a).
Adding hue variation can help make it easier to see
differences between color symbols. A lightness se-
quence combined with a progression through adjacent
hues produces some of the best sequential choropleth
color schemes (for example, yellow– green– blue are
adjacent in the ordering of hues through the spectrum;
Figure 2b). These hue and lightness sequences are
more challenging to design, and useful series of ready-
made sequential schemes are offered online through
ColorBrewer.org (Figure 3) to assist mapmakers who
are not experienced with color specification.5,6
Many map readers find spectral (rainbow) schemes
appealing (Figure 2c). These color schemes are not
well suited to sequential data because lightness varies
through the spectrum (yellow and cyan are often
lighter than other hues). Spectral schemes can be
adjusted to better order lightness, and the intrinsically
light yellow hue can also be used to emphasize critical
values within a data range.7,8 For these diverging
schemes (versus sequential schemes), lightness diverges
from a mid-range critical value toward two contrasting
Figure 1. Hue symbolizes categorical difference in counts for
two race groups. Pie chart symbols are scaled to a constant hues. Figure 4 shows a modified spectral diverging
size and show relative proportions of mortality for two popu- scheme (Figure 4a) and other diverging examples that
lations: black and white males. use fewer hues (Figure 4b,c).

S26 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net

Figure 2. Three color schemes are shown for the same data set. (a) Sequential, single hue scheme (blue). (b) Sequential scheme
with hue transition (yellow-green-blue). (c) Spectral scheme. The spectral scheme is used as a diverging scheme with the lightest
colors marking the overall U.S. rate.

Diverging data may have an obvious structure, such scheme, and looking at distributions using a variety of
as positive and negative values diverging from zero representations may offer the most insight (compare
(Figure 4c). Dark red to white to dark blue is an Figure 2b to Figure 2c).
example color scheme that parallels this diverging Color blindness in map readers becomes an issue
structure. Data may also be presented as diverging from when using multiple hues.9 About 8% of men and ⬍1%
a calculated value such as a national rate, threshold of women have one of the varied forms of red– green
value in disease incidence, or median. These data color vision deficiency. Color blind people do see many
might be equally well represented using a sequential hues but there are predictable groupings of hues that
will be confused with each other. The extent of color
confusions depends on the severity of a person’s color
vision deficiency. The range of hues from red through
orange, brown, yellow, and green may all look the same
or similar if they are also similar in lightness. This set of
color confusions means that some popular color
schemes, such as spectral and “stop light” (red–yellow–
green) schemes, produce maps that are difficult to read
for a substantial number of people.
Red– green combinations are not the only hues that
are confused by people with common color vision
deficiencies. Other example sets of hues that can be
confusing are magenta– gray– cyan and blue–purple.
Example hue pairs that work well as the anchors in
diverging schemes for color blind readers are: red–
blue, red–purple, orange– blue, orange–purple,
brown– blue, brown–purple, yellow– blue, yellow–
purple, yellow– gray, and blue– gray.10 The colorfulness
Figure 3. An example screen from ColorBrewer.org, an on-
line tool offering color specifications for each color in of spectral schemes can also be taken advantage of
schemes suited for thematic maps. Color schemes are while still accommodating most readers’ vision impair-
grouped into sequential, diverging, and qualitative sets. ments by using a spectral scheme that skips the greens:

February 2006 Am J Prev Med 2006;30(2S) S27

Figure 4. Example diverging schemes. (a) Spectral scheme modified to accommodate color blind map readers by skipping green
hues. (b) Two hues (green and purple) diverging from a central light class at the U.S. rate. (c) Change in rates between two time
periods with diverging reds (increasing rates) and blues (decreasing).

dark red, orange, yellow, light blue, blue–purple, dark units each are shown in Figure 5a. Equal interval
purple (Figure 4a).8 ColorBrewer includes a variety of classing breaks the data range into equal segments for
these diverging schemes with full color specifications. predictable and equal class ranges (unlike the variation
The tools at Vischeck.com are also useful for correcting in quantile ranges, as seen in Figure 5a, where the first
the appearance of graphics to accommodate people class has a range of 19.2 deaths and the second has a
with color vision impairments. range of 2.3). The number of counties in each class
varies with equal intervals (Figure 5b). Jenks methods
Classing (called natural breaks in ArcGIS; Figure 5c) minimize
Data classing is another basic decision made when variation within classes and maximize variation between
creating choropleth maps of data. For example, in classes. With this approach, enumeration units that
Figure 5a, counties with data values between 19.23 and share a color are statistically more similar to each other
21.54 are grouped into one class and represented by a than to units in other color classes.3
green color. There are numerous methods for classing Cartographers most commonly choose a Jenks
data11 and most GIS and mapping programs offer a method for their first look at data. In contrast, quantile
selection that often includes quantiles (Figure 5a), classing is the more common choice of epidemiologists,
equal intervals (Figure 5b), and a Jenks optimized perhaps because variation in calculated values pro-
method (Figure 5c). Other choices include classing by duced by different types of standardization and age
standard deviations and minimizing differences across adjustment means death rates may be usefully seen as
boundaries. There is no one correct way to class a data ranked values.11 Cartographers recommend looking at
set, and different methods will produce different map a histogram (Figure 6) or other aspatial graph of the
patterns, especially if data are skewed or include ex- data to assist in choosing classes.3 Generally, a sound
treme outliers. approach is to start with a standard classification and
Quantile classing assigns the same number of enu- adjust breaks to improve the map based on knowledge
meration units to each class (it is a generalized form of of the data and the audience. For example, a useful
percentiles). Four quantiles (quartiles) allocate one adjustment is to group extreme outliers into their own
quarter of the data values to each class with the median class and then class the rest of the data range using a
at the middle break. For example, four classes of 391 standard method. In Figure 7a, for example, rates

S28 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net

Figure 5. Three classifications of the same data set showing different patterns resulting from different classing methods. (a)
Quantile. (b) Equal interval. (c) Jenks optimized classification (natural breaks). The number of counties in each class is shown
to the right of each legend.

⬍12.71 and ⬎30.64 are in separate classes and equal assign colors that readers can tell apart with too many
intervals are applied to the remainder of the data range classes (the extreme being an n-class map). Seven
(compare Figure 7a to Figure 5b). Likewise, when there classes is often the most you will want to use on a
are many zero values in a data set, it works well to choropleth map, and an optimal number of classes can
separate them into their own class and then class the be calculated by examining diminishing reductions in
remainder of the data set. Another adjustment strategy variance with increasing numbers of classes.3 A quick
is to apply Jenks for good statistical breaks and then look at a rough proportional symbol map can also
adjust classes to include the national rate and round provide an alternative understanding of the data distri-
data values to assist map reading by a general audience bution that helps you judge how well the classed view
(Figure 7b).12 represents the data.
Watch map patterns while changing methods and
adjusting breaks to check the sensitivity of the distribu-
tion. The more classes used, the less changeable the Proportional Symbols
map pattern will be with different classing methods and Another way to represent quantitative data, for either
adjustments. There are diminishing returns with in- points or areas, is with symbols that vary by size in
creasing numbers of classes, and it becomes difficult to proportion to data values. Symbols such as circles and
squares are usually scaled by the software in proportion
to the square root of each data value so that symbol
areas visually represent the data values. Sizes of linearly
scaled symbols, such as bars, are more accurately inter-
preted by map readers, but they soon become imprac-
tical with large data ranges. A symbol scaled by area,
such as a square, is more compact and easier to
associate with the location for which it represents data.
Proportional symbols may be placed directly at data
points, such as cities or address locations, or they may
be centered in areas. The order in which symbols are
drawn, so that smaller symbols appear above larger
ones, aids map reading.
Figure 6. Example histogram display in the classify window of Use of proportional symbols for enumeration areas is
ArcMap (ESRI, Redlands CA, v9). particularly useful for count data (total number of

February 2006 Am J Prev Med 2006;30(2S) S29

 C
O
L
C O
O R
L
O
R

Figure 8. Example of a proportional symbol map with legend

showing example symbol sizes and data values they represent.
Each map symbol is scaled to an individual county value.
C
O incidences for example). On choropleth maps, the size
L
O of an enumeration unit has a big effect on the amount
R of color shown on the map, but unit area may have little
relationship, or even an inverse relationship, to base
populations and related counts. United States counties
are a good example of this inversion with dense popu-
lations in small eastern counties and sparse populations
in large western counties that then have an overwhelm-
ing impact on the look of the data distribution. This is
a common failing of choropleth maps that is improved
on by using proportioned symbols.
Symbols that vary by size may also be used to repre-
sent data ranges, as with choropleth classing. The same
decisions in choosing classes are required for this

Figure 7. Classification examples using more customization

to suit a particular dataset. (a) High and low extreme values
grouped into separate classes with the remaining range
classed using equal intervals. (b) Rounded Jenks classes for
improved map reading (data are re-classed but minimal
change to map pattern results; compare to Figure 5c).

S30 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net

symbol form, and a selection of readily differentiated flow lines or networks, and “spider” maps offer a
symbol sizes are usually assigned to the classes. One version of linear symbol sometimes used for epidemio-
name for this symbol type is graduated symbols. They logic data, such as connecting locations to service
are sometimes used when data ranges are too great to points.14
practically represent the full range on a small map.
Another solution to proportioning symbols to extreme
Mapping Multiple Variables
data ranges is to assign all values below a threshold
(⬍100 in Figure 8) to a single small symbol before Improved representations of data and relationships
proportioning. Thus, Figure 8 combines a graduated between cancer variables may be revealed by overlaying
symbol (for the range 0 to 99) and proportioned symbols for one variable onto those of another; map-
symbols (for counts of 100 and over). Another option ping related variables as a series; mapping differences,
for extreme data ranges is to switch to a volumetric modeled indices, residuals, or other derived values; and
symbol and cube-root scaling. mapping with symbols that combine variables. Exam-
ples of these approaches are described in this section.
Other Symbols
Map Series
Many other symbol types are common in mapping
although less useful in cancer mapping. Cartograms are Comparisons among maps of related data are facili-
used much like proportioned symbols, with the forms tated by using the same class breaks on all maps in the
of the enumeration units warped or resized to produce series.2,11 This often means that manual breaks are
areas proportioned to data values. Cartograms were applied to each map and that some maps in the series
popular during and after the 2004 U.S. presidential will not include classes from the entire data range.
elections for interpreting voting data,13 so epidemiolo- Comparisons among map patterns are also aided by
gists may find that they are now a more accepted arranging them as small multiples,15 with many small
symbolization form. Dot density representations that maps on a page or screen. Figures 9 and 10 show two
vary the number of dots in enumeration areas in series of three maps. In the first series (Figure 9), each
proportion to data values (for example, one dot repre- map is classed using quantiles. The second series (Fig-
sents 100 people) are well suited to sparse or discrete ure 10) uses a set of breaks shared among all three
phenomena. Continuous surfaces, such as air quality, maps. The shared breaks are a mix of rounded values
are represented with isolines, filled isolines, or smooth and the U.S. rate for each 5-year aggregation. These
gradations of color based on values across a continuous two series provide quite different views of the data. The
surface. Linear phenomena may be represented with quantiles show how the relative locations of the highest

Figure 9. Example of a map series with each map classed separately using quantile classing. The maps are a time series.

February 2006 Am J Prev Med 2006;30(2S) S31

Figure 10. The same map series seen in Figure 9 with all maps sharing the same set of classes to aid map comparison within the
time series. Class breaks based on the U.S. rate for each time period are included on all maps; the U.S. rate for the 5-year period
mapped is highlighted in each maps’ legend.

and lowest rates change, with the highest rates shifting smoothing and representation of a more generalized
to the east by 1990 –1994. In contrast, the shared breaks surface. Likewise, aggregating to larger enumeration
of Figure 10 also make the overall increase in rates units, through longer spans of time, and across related
through time more obvious (Figure 9 requires careful cancer types may improve the meaningfulness of maps
study of the legends for this information). of cancer data.4

Overlay and Special Classes Multivariate Symbols

Standard mapping methods are often improved with Maps may combine variables by including them within
augmentations such as overlaying reliability informa- one representation. Symbols that combine two or more
tion or other data. For example, a cancer distribution variables include proportioned pie charts and two-
may produce a revealing pattern of rates across coun- variable choropleth maps.3 More generally, multivari-
ties with reasonably high populations, but the extreme ate symbols can be grouped as category/category, cat-
highs and lows associated with small numbers may egory/quantity, and quantity/quantity combinations. A
interfere with the evaluation of this pattern. Masking or category/category symbol may use shape for one vari-
hatching counties with low populations (or using an- able and hue for another. An example category/quan-
other relevant variable such as significance) helps bring tity representation is size for the quantitative variable
the more stable patterns to the fore.16,17 and hue for the category variable. Quantity/quantity
Figure 11 shows data on low numbers of deaths as symbols may be separable, such as size and lightness
area symbols beneath death rates symbolized with point (Figure 13), or more integral, such as the bivariate
symbols to signal sparse and unreliable data. This choropleth shown in Figure 14. Captions for Figures 13
added information allows readers to focus their atten- and 14 provide tips on how to read the two-dimensional
tion on counties with high rates where data are not legends on these bivariate maps.
sparse. Figure 12 includes overlays that mask areas with
no deaths, mask areas where rates are not significantly
Map Finishing
different than the U.S. rate, and hatch areas where data
are sparse. A variety of special categories appear in the Guidelines for mapmakers often do not discuss the
legend of Figure 12 to indicate these exceptions to the seemingly trivial issues of map titles and legend titles or
regular choropleth classes. Small-number problems issues of file export. The look and credibility of final
may also be handled by statistical modeling or Bayesian maps are also affected by selecting a map projection

S32 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net

suited to the area mapped. These are keys to making a
map that can be understood, be presented in multiple
media, and be distributed to a wide audience.

Wording
Completed maps may be missing critical information
about the calculations behind the symbols they present
or, conversely, they may have such laborious titles that
the main issue presented by the map is obscured. A
map title should present the basic topic of the map and
invite the reader to investigate further. The legend title
should provide details of the map calculations (i.e., it
should not be labeled “legend” or something so terse as
“%”). If the calculation is complex, then the clarifica-
tion is best continued in a note in small type on the
map or in associated text. Map sources, data sources,
and authorship are also in small type on the map. This
format varies with media. For example, a journal pub-

Figure 12. Special classes improve a map: zero deaths are

separated to a class and significance and sparse data symbols
are overlayed on a diverging choropleth representation. This
map also shows more complete wording for a thematic map,
with general information in the map title, and specific
information about the calculation and data mapped in the
legend title and note.

lication may move the title and note information to a

figure caption, but the legend title within a map figure
should still be complete.
Figure 12 shows a map title, legend title, and note
suited to publication of a single map. The wording that
comes up automatically for the corresponding map on
the NCI Cancer Mortality website1 (www.cancer.gov/
atlasplus) lists the data parameters chosen by the user
in the map title: Cancer mortality rates by state eco-
nomic areas (age-adjusted 1970 U.S. population), Pros-
tate Gland: black males, 1970 –1994, all ages. The
corresponding legend title online is: Rates per 100,000
person-years, 1970 –1994. The lengthy and complete
online title is reasonable because the user creating the
map selected the parameters and is already involved
Figure 11. Two variables are shown on the map. Propor- with exploring the data. The wording in Figure 12 is
tioned point symbols for rates overlay choropleth symbols for
number of deaths. Counts of zero, ⬍6, and ⬍12 are used to better suited to presenting the map to a wider audience
indicate sparse populations and suggest caution in judging who need to be invited to read the map, to understand
rates (especially extreme rates) in these counties. its primary topic, and to then learn about the particular

February 2006 Am J Prev Med 2006;30(2S) S33

subset of data presented by the map. The title is shorter
and details are in the legend title and note in Figure 12.

File Export
When preparing a GIS map, prepare in advance to
export it for distribution to others. It is difficult to share
a map project (such as an .MXD from ESRI ArcGIS)
directly with graphics production people, and often it

Figure 14. A bivariate choropleth map offers a visual combi-

nation of two variables, making visible their covariation.
Breaks between classes for white male death rates separate
columns. Breaks for black male death rates separate rows.
Overall U.S. rates for black and white groups are used as class
breaks for both races. The lightest color (light green) shows
the lowest values for both groups: both are below the U.S. rate
for white males. The darkest color (top right in legend)
represents counties where each group is above their U.S. rate.
Blacks’ rates (rows) are above 47.2 and whites (columns) are
above 22.0 in counties filled with dark purple.

does not even work to share it with others who do have

the GIS software because paths to associated geography
files and data tables are difficult to preserve.
If the map is going to press, the publishers will likely
Figure 13. A bivariate map showing both number of deaths
and death rates. Size is used for the count variable (rows; request that it to be exported in an Adobe Illustrator
larger symbols for more deaths) and hue and lightness are format (.AI) or a bitmap format (such as .TIFF or
used for the rate data (columns; light yellow to dark red). For .JPEG). Many mapping packages export a version of the
example, the large dark square represents counties with many .AI format that can be read by other graphics software,
deaths and high rates (upper right in legend). The combina- not just Illustrator. The high resolutions and high-
tion of small size and color reduces the visual prominence of
counties with few deaths and thus less reliable rates. Size and quality compression required for bitmap files to pre-
color are separable visual variables. The 2-D legend reads serve small lettering and fine lines will produce large
much like a cross-tabulation table. files that may make file transfer more challenging for

S34 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net

 some authors. Likewise, very detailed geographic data-
bases, such as county shapes, may also produce large .AI
files. If you want to post map products on the web,
exports to .PDF and .JPEG will be good choices. All of
these export formats will cause some problems so be
careful with interesting fonts (that you are not licensed
to share); special type effects (such as halos); complex
symbols (that may be derived from fonts that are not on
others’ computers); and patterns (that may produce
very large files). These choices may not export and
transfer as you intend—the basic rule of thumb is to test
exported files before you do much custom work.

Map Projection
Another basic mapmaking issue is map projection.3 A
projection transforms the base geography from a spher-
ical model of the Earth to the flat page or screen. GIS
implements map projections by applying a series of
equations to geographic coordinates. Projections that
preserve area are suitable for most epidemiologic maps.
These are called equal area projections. The Albers
Equal Area projection is commonly used for the U.S.
C (seen in all figures in this article). In addition, custom-
O ized projections can be created to suit any map scale
L
O and world area.
R If the mapmaker does not attend to map projection,
software defaults usually present the mapped area
underpinned with a regular grid of latitude and longi-
tude lines, producing an inappropriate east–west
stretch and north–south compression at U.S. latitudes.
These distortions interfere with readers’ judgments of
densities and relative areas of cancer rates, which are
crucial for much epidemiologic map interpretation.
Slab-like default projections also mark a map as the
product of an amateur, calling into question compe-
tence of data handing and other GIS decisions.
Figure 15 shows a portion of the southeast U.S. on
three maps at the same scale with a graticule (latitude
and longitude lines) overlay. Figure 15a has no projec-
tion set, the graticule is square, and the counties and
states are distorted by being stretched east to west.
Figure 15b is projected using an Albers Equal Area
projection with settings for the entire U.S., causing this
eastern portion to be tilted, as seen in the angled
latitude lines. The third example, Figure 15c, is ad-
justed by changing the central meridian so north is up

Figure 15. Example of effects of different projections of the

same map area in the southeast United States. (a) No
projection is set so latitude and longitude remain in the
square-grid default arrangement. (b) Albers Equal Area pro-
jection for the U.S. (c) Albers modified by adjusting the
central meridian to the center of the mapped area to position
north as up (this rotation may be made with the Data Frame
toolbar in ArcMap). The gray-filled counties had more than
five prostate cancer deaths for black males, 1970 –1994.

February 2006 Am J Prev Med 2006;30(2S) S35

in the center of the mapped area. This third example 3. Slocum TA, McMaster RB, Kessler FC, Howard HH. Thematic cartography
and geographic visualization. 2nd ed. Upper Saddle River, NJ: Pearson
has suitable projection settings: it is equal area, and the Education, 2005.
central meridian runs through the middle of the area 4. Boscoe FP, Pickle LW. Choosing geographic units for choropleth rate
mapped. maps, with an emphasis on public health applications. Cartogr Geogr Inf
Sci 2003;30:237– 48.
5. Brewer CA, Hatchard GW, Harrower MA. ColorBrewer in print: A catalog
of color schemes for maps. Cartogr Geogr Inf Sci 2003;30:5–32.
Conclusion 6. Harrower MA, Brewer CA. ColorBrewer.org: an online tool for selecting
colour schemes for maps. Br Cartogr Soc 2003;40:27–37.
Maps and other data graphics can play a role in 7. Brewer CA, MacEachren AM, Pickle LW, Herrmann D. Mapping mortality:
generating ideas and hypotheses at the beginning of a evaluating color schemes for choropleth maps. Ann Assoc Am Geogr
project. They are useful as part of analyses for evaluat- 1997;87:411–38.
8. Brewer CA. Spectral schemes: Controversial color use on maps. Cartogr
ing model results and then at the end of a project when Geogr Inf Sci 1997;24:203–20.
researchers present their results and conclusions to 9. Olson JM, Brewer CA. An evaluation of color selections to accommodate
varied audiences, such as their local research group, map users with color-vision impairments. Ann Assoc Am Geogr
1997;87:103–34.
decision makers, or a concerned public. Cancer re- 10. Brewer CA. Guidelines for selecting colors for diverging schemes on maps.
searchers are gaining skill with GIS mapping as one of Br Cartogr Soc 1996;33:79 – 86.
their many tools and are broadening the symbolization 11. Brewer CA, Pickle LW. Evaluation of methods for classifying epidemiolog-
ical data on choropleth maps in series. Ann Assoc Am Geogr
approaches they use for investigating and illustrating 2002;92:662– 81.
their data. A single map is one of many possible 12. Brewer CA. Reflections on mapping Census 2000. Cartogr Geogr Inf Sci
representations of the data, so making multiple maps is 2001;28:213–35.
13. Gastner MT, Shalizi CR, Newman MEJ. Maps and cartograms of the 2004
often part of a complete mapping effort. Symbol types, U.S. presidential election results. 2004. Available at: www-personal.
color choices, and data classing each affect the infor- umich.edu/⬃mejn/election. Accessed August 1, 2005.
mation revealed by a map and are best tailored to the 14. Armstrong MP, Densham PJ, Lolonis P, Rushton G. Cartographic displays
to support locational decision making. Cartogr Geogr Inf Sys
specific characteristics of data. Related data can be 1992;19:154 –164.
examined in series with coordinated classing and can 15. Tufte ER. Envisioning information. Cheshire, CT: Graphics Press, 1990.
also be compared using multivariate symbols that build 16. MacEachren AM, Brewer CA, Pickle LW. Visualizing georeferenced data:
representing reliability of health statistics. Environ Plan A 1998;
on the basic rules of symbol design. Cartography 30:1547– 61.
texts,3,18 –20 more advanced reference books,21,22 and 17. Pickle LW, Mungiole M, Jones GK, White AA. Atlas of United States
handbooks oriented to wider audiences23–26 include mortality. Hyattsville, MD: National Center for Health Statistics, 1996.
18. Dent BD. Cartography: thematic map design. 5th ed. Boston: McGraw-Hill,
guidance on mapping choices implemented in GIS that 1999.
epidemiologists may find useful. 19. Robinson AH, Morrison JL, Muehrcke PC, Kimerling AJ, Guptill SC.
Elements of cartography. 6th ed. New York: Wiley, 1995.
20. Kraak MJ, Ormeling F. Cartography: visualization of geospatial data. 2nd
No financial conflict of interest was reported by the authors of ed. Upper Saddle River, NJ: Pearson Education, 2003.
this paper. 21. MacEachren AM. How maps work: representation, visualization, and de-
sign. New York: Guilford Press, 1995.
22. Fairchild MD. Color appearance models. 2nd ed. New York: John Wiley &
Sons, 2005.
23. Brewer CA. Designing better maps: a guide for GIS users. Redlands, CA:
References ESRI Press, 2005.
1. National Cancer Institute. Cancer mortality maps and graphs. 1999. 24. Brown A, Feringa W. Colour basics for GIS users. Upper Saddle River, NJ:
Available at: www.cancer.gov/atlasplus. Accessed August 1, 2005. Pearson Education, 2003.
2. Wiggins L, Using geographic information systems technology in the 25. Krygier J, Wood D. Making maps: a visual guide to map design for GIS. New
collection, analysis, and presentation of cancer registry data: a handbook of York: Guilford, 2005.
basic practices. Springfield, IL: North American Association of Central 26. Monmonier, M. Mapping it out: expository cartography for the humanities
Cancer Registries, 2002. and social sciences. Chicago, IL: University of Chicago, 1993.

S36 American Journal of Preventive Medicine, Volume 30, Number 2S www.ajpm-online.net