Graphics Reference
In-Depth Information
Figure
.
.
Two-dimensional mosaicplot in fluctuation mode. Highlighting shows pupils who intend
to go to college. hree trends are recognizable visually
Another area in which fluctuation diagrams are applied in matrix visualization.
Figure
.
shows an example of a
correlation matrix for three years of
monthly temperature averages fromasetof locations. Large cellsindicate highly cor-
related variables, while small cells correspond to correlation values that are close to
zero. Negative correlations don't occur in this example, but would be displayed as
theirabsolute values.hedominant feature inFig.
.
isprovidedbythestrong sea-
sonal trends; variables in summer months are highly correlated with each other and
between years,andareclosetobeingindependentofvariables forwintermonth tem-
peratures. Spring and fall months are weakly correlated with all temperature
variables.
For a matrix of cases by variable (the standard spreadsheet format), we might be
tempted to use a fluctuation diagram for an initial overview; since mosaicplots take
categoricaldataasinput,andfiteachtileaccordingtothenumberofcaseswithineach
combination, we can only visualise whole numbers. Instead of counting the number
of cases within each combination of variables, we now use weights for each case and
sum those. he size of a tile then is given by the total sum of weights for the cases of
eachcombination.Plotslikethesearesometimescalledpermutation matrices (Bertin,
) or Bertin plots (Falguerolles et al.,
).
For case-by-variables matrices, we therefore use column labels and row labels
as two categorical variables and the entries of the matrix give us the weights. For
the Mammals' Milk data (Hartigan,
), we have a variable “mammal”, consist-
ing of a list of the mammals' names: HORSE, ORANGUTAN, MONKEY, DONKEY,
HIPPO,CAMEL, BISON,..., and asecond variable “ingredients”, with categories:
