they fit the model on one portion and evaluate it on another. If you'd tested your Chapter 22 regression model on a held-out sample and the $R^2$ dropped dramatically, that would have been a sign of overfitting.