Decision Rules for Increasing the Rate of Successfully Classified Respondents

Abstract

Discriminant analysis is one of the more commonly used methodsfor classifying individuals. In this article, the authors discuss strategies for increasing the percentage of correctly classified subjects (i.e., the hit rate). Through their study of 303,000 credit card customers of a corporation, who were sent offers to purchase insurance policies through the mail, the authorsfound that by varying the decision rulesfor classification one could more accurately predict response rates and establish cutoff points for acceptance that best meet the researcher's needs. The results of this study demonstrate that one can increase the hit rate from one of less than .2% (the random response rate) to one of more than .6% by appealing only to those prospective respondents ranked in the top 10o according to classification scores. Such a result is a large improvement over the typical output available through some of the more commonly used software packages for classification of individuals.