Graphics Reference
In-Depth Information
Generalizing to Aggregated Views
9.4
he unaggregated approach runs into problems with more than small amounts of
data. If you have tens of thousands of points, oten you want to look at views that ag-
gregate the data, such as barcharts, histograms, and frequency tables. In this chapter,
we use a reasonably sized dataset to examine linking; there are about
play-
ers represented over a total of ca.
player seasons. A quick review of Figs.
.
through
.
shouldconvince thereaderthat thesummaryviewsareeasiertousethan
the unaggregated views for these numbers. herefore, in this section, we consider
techniques for linking aggregated views.
Oneofthefirstcommonlyavailable toolstofeaturelinking wasthestatistical anal-
ysis packageData Desk (Velleman,
). Itwas originally built as ateaching tool but
isnowafull-featured statistical packagethat haslinked views designed in at the core.
Brushing works with aggregated views such as barcharts and histograms as well as
within unaggregated views, and the outputs of analyses suchas regression and corre-
lation analysis canbevisualized andarelinked totheotherviews.Ifauserspotssome
unusual cases inaviewof residuals froman analysis, they can brushthose points, see
if there is anything that might explain it in other variables, modify the model, and
instantly see the residual view update to reflect the new model. By incorporating
model results as linkable views, Data Desk provides a rich set of linking paradigms.
In this section we consider simply the way it performs linking to aggregated views.
Figure
.
shows an example of such linking. his figure is similar to Fig.
.
, except
that the linking is being performed in the reverse direction - from the scatterplot to
the barchart.
As in Fig.
.
,the degree of interest for a bar has been represented by dividing the
bar into two sections; one for the selected subset and one for the nonselected subset.
An alternative method is used by LispStat (Tierney,
), in which each data item
is assigned its own place in the bar and that section of the bar has an appropriate
brightness. his solution is very close to the “one-to-one” relationship method in the
previous section, as under this display system each bar is really a set of stacked rect-
Figure
.
.
Linked views in Data Desk. Points selected in a scatterplot of miles per gallon vs. weight are
highlighted in the country barchart. Selecting the points in the low end of the weight scale shows which
country makes the lightest cars
