The example used a classic dataset to explore models relating car mileage to a set of design and performance features. One key insight was to work with a scaled version of the reciprocal of mpg rather than mpg itself. Another insight was to develop a parsimonious model, given the relatively small sample size and high ratio of variables to cases. A final insight was to create a predictor by taking the ratio of two predictors--hp and wt--rather than working with the manifest predictors.
Indeed, this was one of the points of the article by Henderson and Velleman, who cautioned against automated multiple regression model-building back in 1981! The model we ended up with is parsimonious, interpretable, and fits the data well.
In the next chapter, we turn to two important exploratory techniques: Principal Components Analysis and Factor Analysis.