Model Selection: Finding the Right Fit.

Abstract

There are numerous possible goals for building statistical models. Those statistical goals, the associated model types, and each statistical tool involved in model building come with its own assumptions and requirements. In turn, these requirements must be met if we are to ensure that our models produce meaningful, interpretable results. However, beyond these technical details is the intuition, and the additional set of tools and algorithms, used by the statistician, to build the contextually appropriate model: not only must we build an interpretable model, we must build a model that answers the particular question at hand and addresses the particular goal we have in mind. In this column we discuss the methods by which statisticians build models for description, risk factor identification, and prediction.