ABSTRACT
Marketers use decile analysis to assess their models regarding classification or prediction accuracy. This chapter presents two additional concepts of model assessment—precision and separability—and illustrates these concepts by further use of the decile analysis. It discusses the traditional concepts of accuracy for response and profit models and illustrates the basic measures of accuracy. The chapter introduces the accuracy measure used in marketing, known as Cum Lift. The traditional measure of accuracy is the proportion of total correct classifications (PTCC), calculated from a simple cross tabulation. There are several measures of prediction accuracy, all of which use the concept of error, namely, actual profit minus predicted profit. The mean squared error (MSE) is by far the most popular measure, but it is flawed, thus necessitating three alternative measures. Calculation of the decile analysis for a profit model is similar to that of the decile analysis for a response model with response and response rates replaced by profit and mean profit, respectively.
