Graphics Reference
In-Depth Information
Figure
.
.
[his figure also appears in the color insert.] Parallel coordinate plot of the financial ratios
with skewed distributions. he seven outliers selected in Fig.
.
have been removed. he plot's (red)
border is a sign that not all data are displayed
exceedingacertainsizemusthavesomeliabilities.hedistortingeffectoftheextreme
values is demonstrated by the fact that the zoomed plot, which covers a region
−
times the size of the default plot, contains
% of the data. α-Blendinghas been used
to make the distributional structure more visible. α-Blending weights the shading of
each object by a specified fraction. he darkness of a pixel is that fraction multiplied
by either the number of objects overlapping the pixel or
, whichever is smaller.
Setting aside the seven worst outliers, Fig.
.
rescales to Fig.
.
. Some of the
variables becomepotentially moreinformative, whilethescales of someof theothers
are dominated by newly visible outliers.
Outliers can be dealt with in several ways. A transformation might help (but it is
notalways appropriate; inthis case,several ratios havesomenegative orzerovalues).
heoutlierscouldbetrimmedordiscarded(dependingonwhatkindsofoutliersthey
are).With ratio variables, suchasthose usedhere,thecomponent variables of the ra-
tioscanbeexamined.Figure
.
showsascatterplotofthevariable Salesagainst Total
Assets with the same seven outliers still highlighted. All turn out to be small compa-
nies in terms of both Sales and Total Assets, and small companies should probably be
treated separately anyway. he scatterplot also reveals that there are some bivariate
outliers. Querying shows that the six cases with high Sales but low Total Assets (the
upper let group) are all data for the same big retail company over several succes-
sive years. he seven cases with low Sales but very high Total Assets (the lower right
group) represent three companies from the information and communication sector.
hese two groups make up a tiny proportion of the dataset, even if they all have very
large Sales or Total Assets.
For exploratory work it is distributional structure that is of interest, not precise
values. he minima and maxima of the variables are used to determine the limits of
the axes. he lower limit for Sales in Fig.
.
shows that the lowest Sales values are
actually negative and that plenty of them are zero or almost zero too. he negative
