Identification of Groups
There are a number of techniques that help us to identify inherently different groups of records. They differ to the techniques above because they concern groups of variables within records, rather than variables across records.
These are useful in segmentation and description of markets.
Correspondence Mapping
This interdependence technique produces a perceptual map describing the relationship between nominal variables, usually in two dimensions. The distances between category points describe the relationships: similar categories are placed close together, dissimilar categories far apart. Correspondence is a measure of association and is generally an exploratory technique. It is useful for showing, say, both brands and respondent characteristics on one map.
Discriminant Analysis
Used to describe samples into groups by predicting the likelihood that a record belongs to a particular group based on a number of independent variables.
The groups are pre-defined.
Cluster Analysis
Here the technique is used to define groups, but the groups are not pre- defined. It operates on as many variables as you wish, though often factor analysis variables are used. Records are grouped together on the basis that the members of each group are more like each other than they are like any member of another group. The number of groups obtained can be chosen, but in practice a given number of groups provides the most useful division.
AID/CHAID
AID techniques (automatic interaction detection) are methods of identifying key drivers (which individual variables segregate the data?) and niche markets (how large are groups with shared characteristics?). They belong to the family of techniques known as decision trees.
An overall dependent variable is identified first, and a number of likely predictor variables. The basis of division is to capture the greatest amount of variance (or difference) between the subsequent groups. Each sub-group is then sub-divided again, using the variable with the greatest variance. The outcome is an hierarchical tree of smaller and smaller groups, describing the characteristics and size of each group. |
