Graphics Reference
In-Depth Information
beensetuptohavememory,sotheuserwasabletoselectthecritical yearrangemore
closely, and then each bar was successively selected in the position bar chart. When
the catcher position was selected, the intersection with the year histogram identified
the cluster in the scatterplot. It appears that catchers had a fundamentally different
method of fielding prior to
.
he advantage of a system with memory is that it allows the user to make a query
involving a number of variables with little effort, and is forgiving; if you make an
error in adjusting one selection, it is easy to change it. he advantage of a memo-
ryless system is in power and adaptability. It is hard to generalize a memory-based
system while retaining an intuitive interface (imagine the di
culties in coordinating
selections from a map, a scatterplot and a network view, for example), and it makes
choosing different selection operations di
cult as the method of combination of the
different selections istypically fixed.Fromobservation ofexisting systems, it appears
that the memory-based approach is particularly good for directed queries and an-
swering the question “Which objects satisfy this description?” and the memoryless
approach is better for discovering structure in the data and answering the question
“Are there unusual features in these data?”
Figure
.
isthe mostcomplex figureinthis chapter.Each barchart keeps track of
itsownselectionanddisplaysthatselectionusingahashpattern.Wearealsodefining
a nonbinary degree of interest, in this case with three levels. As well as showing the
black cluster that was the focus of the investigation, we can also see an interesting
gap in the linear feature along the bottom of the scatterplot. A putout ratio of about
eight putouts per game was rarely observed prior to
and was not observed for
catchers ever. he use of a nonbinary degree of interest, while more complex to use,
shows additional information that would have been missed otherwise.
Linking to Domain-speciic Views
9.7
Oneoftheattractions ofthelinkedviewsparadigmisthatitmakesiteasytointegrate
aview that is useful only for a particular form of data into a general system. All that is
needed is that a view be able to fulfill requirements (
) and/or (
) of Sect.
.
and the
viewcan beadded directly into the general system. Forspatial data, Unwin and Wills
built a system that combined a number of statistical views with geographical views.
REGARD (Haslett et al.
) allowed the user to manipulate maps of geographical
information, containing layers of data representing sets of geographical information
(towns, rivers, countries, etc.). hese different layers contain entities with statistical
data, so that users can create data views on one or more of these variables and use
the linking system to tie the views together.
Figure
.
shows the simplest and most intuitive way to add a geographic view
to a linked views system. We simply code the selection by brightness to give the well-
known choropleth map.hismethod will workfornonbinary degrees of interest and
isanintuitiveview.Inthisfigure,wecanseethatthereisastrongrelationshipbetween
marriage patterns and spatial location. It is let as an exercise to the reader to guess
