Graphics Reference
In-Depth Information
with lower
hic
values. Since there is a fair number of such vehicles in Node
, this
could also account for the low sample mean response.
Finally, a third confounding variable is evident in Fig.
.
, which shows barplots
of the proportions of barrier shape type (
barshp
) within each leaf node of the tree.
Node
, whose bars are colored green, stands out in that barrier shapes EOB, GRL,
IAT, MBR, and SGN practically never appear in the other two nodes. For some rea-
son, the testers seem to prefer these barrier shapes for high-speed crashes. hus bar-
rier shape is yet another possible explanation for the low mean response value in the
node.
Despite these di
culties, it is clear from the plots that three vehicle models stand
out as outliers:
Honda Accord,
Nissan Altima, and
Subaru Legacy.
All areforeignimports.he
Subaru Legacyappearsasanoutlier intwoseparate
tests, one at moderate speed and one at high speed.
Figure
.
shows the corresponding tree model for the front passenger data. Now
airbagandbarrierrigidityappearassplitvariablesaterthetop-levelspliton
vehspd
.
he plots of the data in the leaf nodes are presented in Fig.
.
. Everything seems to
make sense: injury is less severe when a vehicle is equipped with airbags and when
it is crashed into a deformable barrier, and also if impact occurs on the driver side
(Node
). It is interesting to note that in Node
, where
vehspd
.
and the
vehicles are equipped with airbags, rigid barriers are used for the higher speeds and
deformable barriers for the lower speeds. his may exaggerate the effect of
vehspd
in this node. he outliers for these data turn out to be all domestic models:
Figure
.
.
Proportions of different barrier shapes within the three leaf nodes of the tree model in
Fig.
.
. he lengths of the bars sum to one for each color
