Response measurement and optimization of direct mailings

Beginning with 1995, we discuss different studies that deal with response measurement and optimization of direct mailings. Most of these studies analyze data sets from mail order companies or charities. We classify various dependent and predictor variables and—w.r.t. the latter distinguish static and dynamic effects. Response models are divided into parametric and flexible models. Besides, we analyze important modeling aspects, i.e., latent heterogeneity and endogeneity. Optimization methods are presented according to whether they refer to static or dynamic objectives. Based on these modeling aspects we evaluate the different studies. Considering various studies of model evaluation it becomes evident that logit models frequently constitute a good choice. However, Bayesian neural nets and Tobit models turn out to be good alternatives. As predictor effects are concerned results vary. Authors do not completely agree on which variables are the most important. Furthermore, signs and significances of predictors vary across studies. The majority of studies neglect latent heterogeneity and endogeneity. Finally, results show that there are still plenty of interesting research possibilities, such as a comprehensive evaluation of models or new specifications of (mailing) variables.