Association Rules - Interpreting and Comparing Results
When comparing the results of applying association rules (e.g., Association Networks, 2D, 3D; see also Computational Procedures and Terminology) to those from simple frequency or cross-tabulation tables (see Basic Statistics), you may notice that in some cases very high-frequency codes or text values (items) are not part of any association rule. This can sometimes be perplexing.
To illustrate how this pattern of findings can occur, consider this example: Suppose you analyzed data from a survey of insurance rates for different makes of automobiles in America. Simple tabulation would very likely show that many people drive automobiles manufactured by Ford, GM, and Chrysler; however, none of these makes may be associated with particular patterns in insurance rates, i.e., none of these brands may be involved in high-confidence, high-correlation association rules linking them to particular categories of insurance rates. However, when applying association rules methods, automobile makes which occur in the sample with relatively low frequency (e.g., Porsche) may be found to be associated with high insurance rates (allowing you to infer, for example, a rule that if Car=Porsche then Insurance=High). If you only reviewed a simple cross-tabulation table (make of car by insurance rate) this high-confidence association rule may well have gone unnoticed.