Graphics Reference
In-Depth Information
returnsomevalueinaknownrangeforeverypossiblesubsetofthedata;thisisneces-
sary when displaying results in aggregated views. An aggregated view is a view where
a single graphic item represents multiple cases of data, as is the case for a histogram
or barchart where a single bar summarizes multiple data rows. In contrast, an unag-
gregated view is one where each row of the data gets a separate visual representation,
as is the case for scatterplots. In this section we use the terms data case and data row
interchangeably, defined as a single observation of data, typically encoded as a row
in a data table. A graphic item is slightly harder to define but should be thought of as
a visually distinct representation that is perceived as a distinguishable unit. A bar in
abarchart,aglyphinascatterplot,anda
-Dsurfaceareallgraphicitems.Asegment
of a path is not; the path is a perceptual whole and is thought of and manipulated as
asingleentity.
In practice, a simplified form of degree of interest can be used where each row of
data (each data “case”) is given a degree of interest value, and the measure of a subset
is defined as a summary function of the degrees of interest of the cases comprising
the subset. hroughout this chapter, we use a degree of interest for the cases in the
range[
,
],where“
”correspondsto“nointerestwhatsoever”and“
”correspondsto
“maximal interest.” For a subset, we will use the mean value summary to summarize
the values in the subset. his remains in the range [
,
] and has the useful property
that subsets of different sizes that have the same distribution of measures of interest
on their cases will have similar measures of interest. However, it should be noted
that other summaries can be useful for certain tasks. For example, the maximum
summary function will highlight subsets that have any selected items in them. his
will be very useful for spotting outliers, as typically outliers are small in number, so
usingthemeansummarywhenoutliersarethesourceofadegree-of-interestmeasure
will result in most subsets having zero or near-zero measure of interest, whereas the
maximum summary statistic will show immediately any subset containing one or
more outliers.
An additional simplification we can make is to state that any view that defines
a degree-of-interest value for each case must define it as either zero or one. In other
words, the process of interacting with a data view to perform linking will result in
a split into selected cases (
s) and unselected cases (
s). his technique is by far the
most commonly implemented technique, but there are cases where the moregeneral
version is useful. In Sect.
.
, we explore distance-based linking, and in Sect.
.
we
show examples of how multiple views can create nonbinary degrees of interest for
cases.
For requirement (
) above, we need a means for the user to indicate what is of
interest to them. here are multiple means of doing so, the most common of which
are:
Brushing. Inthe brushing paradigm, the userhas a fixed shape (typically a rectangle
or circle) that they drag over a view of data. As this brush moves over graphic
itemsintheview,itpaints them.hisisdefinedassetting thedegreeofinterestto
for those items and then typically using colour to code the degree of interest in
all linked views. In Sect.
.
we give more details on scatterplot brushing, which
was one of the earliest implementations of linked views.
