ABSTRACT
Traditional validation of a marketing model uses a holdout sample consisting of individuals who are not part of the sample used in building the model itself. This chapter points to the weaknesses of the traditional validation and then presents a bootstrap approach for validating response and profit models as well as measuring the efficiency of the models. The data analyst's first step in building a marketing model is to split randomly the original data file into two mutually exclusive parts: a calibration sample for developing the model and validation or holdout sample for assessing the reliability of the model. As marketers use the Cum Lift measure from a decile analysis to assess the goodness of a model, the validation of the model consists of comparing the Cum Lifts from the calibration and holdout decile analyses based on the model. The bootstrap method is a computer-intensive approach to statistical inference. The bootstrap also falls into the class of nonparametric procedures.
