"Exploratory Analysis of Marketing Data": A Reply

The major problem is that Crocker misinterprets the intent of our article. Before considering the specific points he raises, let me restate the basic intent: regression analysis is commonly used in the development of predictive models in marketing. Some evidence now exists to suggest that, in certain situations, this practice may not be optimal. An alternative method, tree analysis, may lead to better predictive models in these situations. We tested the above hypothesis in a situation where regression analysis is often used in an exploratory manner-the site location problem. For example, Kotler [4, p. 442] implied that the Rayco Company was following good practice when it used stepwise regression to develop a model to predict volume at retail locations. And I am aware of at least one oil company which has used stepwise regression for location of service stations. In our judgment, problems with the data in the site location problem should tend to render the exploratory use of regression analysis less effective than the use of trees. In short, we were attempting to demonstrate that there are situations where trees are expected to be more effective than regression. I would now like to consider the specific points raised in Crocker's article.