ABSTRACT
Lift charts (Barry and Linoff, 1997) are a graphical tool that both indicate the relative predictive power of different possible candidate models, and allow managers to quickly apply an estimated predictive model in order to address specific managerial questions. This chapter describes how lift charts are created and how they are used to profitably target customers in direct marketing applications. To make the description of how lift charts are created concrete, this note makes use of an example based on a random sample of 50 donors from the Canadian Charitable Society data set (CCS in the BCA data library) and the MixedCCS2 logistic regression model that is developed in the logistic regression tutorial in Chapter 5. The managerial use of lift charts is illustrated using a second example involving a regional telephone service provider’s sales campaign for DSL service. While the example approach makes the concepts behind the creation of lift charts more concrete, the disadvantage is that lift chart methods rely on there being many customers in the database, breaking those customers into a set of groups (traditionally ten groups), and then presenting averages for each of these groups. The use of only 50 observations allows us to reasonably look at all the data, but makes dividing the data into ten separate groups impractical for reasons of random sampling error. Consequently, instead of using ten groups in this note, the data will be divided into five groups. However, the concepts presented are applicable to an arbitrary number of groups.
