Graphics Reference
In-Depth Information
he Fig.
.
approach of accumulating foreground regions from the ends toward
the center is one of several micromap variations tried over the years. Accumulating
regions from one end to the other has been popular and is easier to explain. One
drawback is that micromaps look more complicated when more than half of the re-
gions appear in the foreground. A second drawback is that the chances increase of
a background region being completely surrounded by outlined states. his will cause
the background region to appear in the foreground. he region color fill will clarify
but may not be noticed. Usability studies could help in ranking the various options.
Figure
.
illustrates different kinds of statistical panels. he first statistical panel
shows summary values for the
-year period
to
as filled dots. It shows
the values for the
-year period from
to
as arrow tips. Both encodings
are position along scale encodings. Arrow length encodes the difference in rates for
the two time intervals. Length is a good encoding in terms of perceptual accuracy of
extraction. he huge increase in mortality rates for all US states is obvious.
he second statistical panel in Fig.
.
shows rate estimates and
% confidence
intervals for the
-yearperiod from
to
.Typically, the confidence intervals
are considered secondary information and the rates are plotted on top. In Fig.
.
,
this totally obscures the confidence intervals except for Alaska, Hawaii, and Wash-
ington, DC. here are various remedies, such as showing
% confidence intervals.
Our purpose in Fig.
.
is to call attention to how well the state rates are known and
how little this conveys about the geospatial variation within the states.
he third and rightmost statistical panel in Fig.
.
shows boxplots of the rate esti-
mates for the
-year period from
to
for the counties of each US state. he
outliers appear as open circles. he geospatial variation based on
years of data is
substantial. Note that the scale has changed from the panel with state rates and con-
fidence intervals. Using the county scale for both columns would better convey the
county variability. Ofcourse,the
-yearstate rate estimates are alsohiding variation
over time.
In recent years, US federal agencies have placed increasing emphasis onconfiden-
tiality. he suppression of data is increasingly common. One approach toward mak-
ingsomethingavailabletothepublichasreliedonaggregationtoobscuredetails.his
leads to aggregation over important factors such as time, geospatial regions, race,
and sex. here is currently a serious consideration for suppressing all county-level
mortality-rate estimates. Suppression of data is a problem for the public concerned
about human health. here are additional issues related to data that is not collected.
For example, data are not collected on cigarette smoking in terms of packs perday at
the county level.
Micromaps via nViZn
1.5.2
nViZn (Wilkinson et al.,
) (read en vision) is a Java-based SDK, developed and
distributed by SPSS (http://www.spss.com/visualization/services/). nViZn was in-
spired by the idea of building on the BLS Graphics Production Library (GPL), de-
scribed in Carr et al. (
), with a formal grammar for the specification of statisti-
cal graphics (Wilkinson,
). nViZn was created as a distinct product whose wide
