Graphics Reference
In-Depth Information
variables. Lookingatthevariables Water Sotness and Temperature wewillfindsome-
thing that is to be expected: harder water needs warmer temperatures for the same
washing result and a fixed amount of detergent.
Mosaic plots allow the inspection of the interaction of M-user and Preference con-
ditioned for each combination of Water Sotness and Temperature, resulting in a plot
thatincludesthevariablesintheorderinwhichtheyarelistedabove.Figure
.
shows
the increasing interaction of M-user and Preference for harder water and higher tem-
peratures.
Several recommendations can be given for the construction of high-dimensional
classical mosaicplots:
he first two and the last two variables in a mosaicplot can be investigated most
e
ciently regarding their association. hus the interaction of interest should be
putintothelasttwopositions oftheplot.Variables thatcondition aneffectshould
be the first in the plot.
To avoid unnecessary clutter in a mosaicplot of equally important variables, put
variables with only a few categories first.
If combinations of cells are empty (this is quite common for high-dimensional
data due to the curse of dimensionality), seek variables that create empty cells at
high levels in the plot to reduce the number of cells to be plotted (empty cells at
ahigherlevelarenotdivided any further,thus gathering many potential cellsinto
one).
If the last variable in the plot is a binary factor, one can reduce the number of
cells by linking the last variable via highlighting. his is the usual way to handle
categorical response models.
Figure
.
.
he interaction of M-user and Preference increases for harder water and higher
temperatures
