Objective
You have master data about your customers and want to know which customers are most likely to respond to a mailing campaign? Or perhaps you have results from a CRM study showing acceptance levels for your brand and you want to know which customers are at the greatest danger of being lost to the competition and in which circumstances?
Fig. 1: Influence of categorial data on the response rate to a mailing campaign
These are typical situations for the application of CHAID analysis (Chi-squared Automatic Interaction Detector). An easy-to-read tree diagram shows a hierarchy of those subgroups in which the responders or dissatisfied customers are most likely to be found. Thus, it is easy to ascertain whether or not there are attributes that can distinguish between the categories of the target variable. These attributes are arranged from the "trunk" to the "branches" according to their statistical significance.
Prerequisites
The variables must be present in categories (ordinal or nominal). Prior to the analysis, decisions must be reached about the degree of statistical significance to be applied and also about the number of cases at which groups should be separated or merged.
The analysis does not provide linear relationships in the form "the more x, the more y" as is the case with regression analysis, for instance. Statements are made about individual categories and combinations of categories.
