Graphics Reference
In-Depth Information
icon with two cyclical arrows. A variable is selected and the min/max values on that
axes are inverted. Diverging lines (as obtained for positive correlation) nowintersect
in Fig.
.
, making it easier to visually spot the crossing and hence the correlation.
Actually, it is worth working with the
A
query, experimenting with various angle
ranges and using the inversion to check out or confirm special clusters.
vary one of the variables watching for interesting variations in the other vari-
ables.
Doing this on the
Yen
axis (see Fig.
.
), we strike another gold connection. he
(rough) intersection of a bunch of lines joining
Yen
to the
D
mark corresponds, due
to the duality, to their rate of exchange. When the rate of exchange changes, so does
the intersection and the price of gold! In other words, movements in currency ex-
change rates and the price range of gold go together. Are there any indications that
are associated with the highrange ofgold? hetoppricerange isselected(Fig.
.
),
and prompted by the results fromthe previous query, we check out the exchange rate
between
S
terling and
D
mark (or
Yen
). he results are stunning: a perfect straight
line. he slope is the rate of exchange, which is constant when
G
old tops out. he
relation between
S
terling and
D
mark is then checked for different price ranges of
G
old (Fig.
.
), and the only regularity found is the one straight line above. Aside
from the trading guideline it establishes, we could say that this suggests behind-the-
scenesmanipulationofthegoldmarket...butwewon't.Webanishanysuchthought
and proceed with the boolean complement (Fig.
.
) of an
I
(or any other) query.
Not finding anything, we select a narrow but dense range on the
Yen
(Fig.
.
) and
notice an interesting relation between
D
mark, interest rates and
G
old.
here is an exploratory step akin to “multidimensional contouring,” which we
fondly call the zebra; it is activated by the last icon button on the right, with the ap-
propriateskin-color.Avariableaxisisselected(the
SP
axis in Fig.
.
), divided
into a number (user-specified) of intervals (four here), and colored differently. his
shows the connections (influence) of the intervals forthe remaining variables, which
arerichlystructuredhere,especially forthehighestrange.Sowhatdoesittake forthe
SP
to rise? his is a good question and helps introduce Parallax's classifier. he
result,showninFig.
.
,confirms theinvestment community's experience that low
M
TB and
G
old predict high
SP
.A comparison with the results obtained on this
dataset using other visualization tools would be instructive, although unfortunately
they are not available. Still, let us consider how such an analysis would be performed
using a scatterplot matrix. here are ten variables (axes), which require
pairwise
scatterplots each; even with a large monitor screen, these could be no larger than
about
.
.
cm
. Varying one or more variables in tandem and observing the ef-
fects simultaneously over all of the variables in the
squares is possible but quite
a challenging task. By contrast, the effects of varying
D
mark, for stable
Y
en, on the
two interest rates,
G
old as well as the remaining variables are easily seen in just one
plotusing
-coords:Fig.
.
.hisexample illustrates the di
culties associated with
high representational complexity (see Sect.
.
.
, item
), which is O
N
for the
(
)
scatterplot matrix but O
N
for
-coords, and made even clearer with the next data-
(
)
set.
