There are two ways to run Cluster in Clustan - interactively, or by specifying commands in a Clustan program.   For this tutorial we'll focus on the command language approach, because it's easier to illustrate on a webpage.

Here's a Clustan program which runs Cluster with mixed variable types and missing values:

read data, variables multi 1 cont 2-5 bin 6 missing 6 * -1.0, cases 9
2 .57 .53 .69 .43  0
2 .72 .54 .72 .41  0
3 .50 -1. .68 .46  0
-1 .47 .71 .84 .56 -1
1 -1. .49 .53 .42  0
1 .25 .38 .41 .33  0
3 .45 .68 -1. .45  0
1 .57 .70 .78 .55  1
2 .87 .82 .87 .51  0

read labels variables
Blood Ht/Wt FRG val ECG ind AntibodyCyst

cluster, method wards,
print fusions, centres min 2 max 4, members min 2 max 4

stop

The program calls three Clustan procedures - Read Data, Read Labels and Cluster.

Read Data defines 6 variables: the first is a multi-state or ordinal variable; variables 2-5 are continuous; and variable 6 is binary.  We also specify missing value codes of -1 for all 6 variables.  There are 9 cases to be read and the data are read in free-format; that is, successive values are separated by blanks.  Note that a missing value (code -1) occurs in all 6 variables.

Read Labels specifies alphabetical labels for the variables.

The third command runs Cluster directly on the data matrix to complete a Ward's analysis.   The options selected for printing are the fusion schedule, cluster centres and cluster membership for 2, 3 and 4 clusters.

A typical output for 3 clusters is as follows:

Table of cluster centres for 3 clusters
Cluster codes            1        2        3
Cluster weights          5.00     2.00     2.00

Variable Type Cluster:   1        2        3
Blood    Multi. 1 %      0      100      100
Observations/weights     5.       1.       2.
Blood    Multi. 2 %     60        0        0
Observations/weights     5.       1.       2.
Blood    Multi. 3 %     40        0        0
Observations/weights     5.       1.       2.
Ht/Wt    Cont. Mean      0.69     0.61     0.41
Observations/weights     5.       2.       1.
FRG val  Cont. Mean      0.47     0.50     0.36
Observations/weights     4.       2.       2.
ECG ind  Cont. Mean      0.72     0.87     0.13
Observations/weights     4.       2.       2.
Antibody Cont. Mean      1.27     1.49     1.10
Cyst     Binary %        0      100        0
Observations/weights     5.       1.       2.

We would like to draw attention to some special features in this table.  Cluster weights refer to the number of cases in each of the 3 clusters; they are decimal values because the cases could have been assigned differential decimal weights.

The number of observations/weights for each variable in each cluster is the sum of the weights for valid observations only; missing values have been ignored.  In the example, each variable has a weight total of 8, i.e. 9 cases less 1 missing value on each variable.

Cluster means are computed for continuous variables, but percentages of occurrence are computed for the multi-state and binary variables.

The tree corresponding to this hierarchical cluster analysis by Ward's method can also be displayed using ClustanGraphics:

Finally, we would emphasize that since Cluster operates directly on a data matrix we can use it to obtain a hierarchical classification for thousands of cases.  See very large surveys for details.