Parts One and Two of this series illustrated four problems with using a model-building approach when data snooping. This column will present an alternative approach for data snooping that is of proven utility. This approach is completely empirical and works with all types of data.

The model-building approach

When carrying out an experimental study we want to know how a set of input variables affect some response variable. In this scenario it makes sense to express the relationships with some appropriate model (such as a regression equation). The experimental runs will focus our attention on these specific input and response variables, and the model will answer our questions about the relationships.

 …