Graphics Reference
In-Depth Information
Table
.
.
he hospital data
Length of stay(in years)
Visit frequency
-
-
+
Σ
Regular
43
16
3
62
Less than monthly
6
11
10
27
Never
9
18
16
43
Σ58
45
29
132
Table
.
.
he hospital data, corrected for the column margin
Length of stay (in years)
Visit frequency
2-9
10-19
20+
Regular
0.74
0.36
0.10
Less than monthly
0.10
0.24
0.35
Never
0.16
0.40
0.55
Σ1.00
1.00
1.00
Table
.
.
he hospital data, corrected for the row margin
Length of stay(in years)
Visit frequency
-
-
+
Σ
Regular
0.69
0.26
0.05
1.00
Less than monthly
0.22
0.41
0.37
1.00
Never
0.21
0.42
0.37
1.00
In addition, Haberman (
) notes that this pattern is not significantly different
in the “less than monthly” and “never” strata. From the row-standardized table (see
Table
.
), it seems indeed that LOS is homogeneous with respect to these two visit
frequency strata.
Although far from optimal, contingency tables are frequently visualized using
grouped bar plots (see Fig.
.
) or even by means of
-D bar charts (see Fig.
.
).
It seems hard to detect the aforementioned pattern in these, especially in the
-D
plot, where the perspective view tends to distort the true proportions of the bars.
In the following, we will introduce three graphical methods that are better suited to
contingency tables.
Mosaic Displays
12.2.1
Mosaic displays were introduced by Hartigan and Kleiner (
,
) and extended
by, among others, Friendly (
,
,
). hey visualize the observed values
of a contingency table by area-proportional tiles, arranged in a rectangular mosaic.
he tiles are obtained by recursively partitioning and splitting a rectangle. In the fol-
lowing, we describe the main concepts of mosaicplots; Chapter III-
in this topic
