ABSTRACT
Predictive models are mathematical models—i.e., equations—that allow us to predict some variable of interest from one or more other variables that are believed to be related. The simplest case is that of a "best-fit line" through a scatterplot. It discusses the whiteside data frame nicely illustrates many of the main ideas. The term "linear regression" refers to problems like this: even if some of the functions involved are nonlinear, the model fitting problem remains a linear regression problem if the parameters appear linearly. Two key points are, first, that the partitioning strategy just described is applicable to all predictive model types, and, second, that it is very easy to implement in R. The key feature of a linear regression model with interactions between predictors is that the influence of one of the predictors on the response variable depends on the value of the other predictor.
