An Iterative Approach to Classification Analysis

This paper introduces a totally new approach to classification analysis. Part of its appeal stems from the fact that it provides a method for complying with Proposition 103's requirement that variables be considered in a specific order. Our paper is presented using private passenger automobile insurance; but, the technique could be used with any line of insurance. The analysis is based on a statistical procedure known as CHAID, introduced in 1980. CHAID analyzes a single attribute of a population (the dependent variable) based on other attributes (predictor variables). CHAID iteratively subdivides the population into classes having significantly different values for the dependent variable. This paper first describes some of the limitations of current classification analysis. We then introduce CHAID, and discuss how it overcomes these limitations. Because this is an introductory article, we have presented the statistical concepts of CHAID without fully explaining the underlying theory. Next, we present a CHAID analysis of "live" private passenger automobile experience. We then derive credibility weighted relativities for the CHAID classes. Finally, we briefly discuss actuarial and operational implications of a CHAID-derived class plan, and additional areas of needed research.