We can also use local regression fits as building blocks in a GAM, using the lo() function.

gam.lo <- gam(wage ~ s(year, df = 4) + lo(age, span = 0.7) + education, data = Wage)
plot.Gam(gam.lo, se = TRUE, col = "green")

Here we have used local regression for the age term, with a span of 0.7. We can also use the lo() function to create interactions before calling the gam() function. For example,

gam.lo.i <- gam(wage ~ lo(year, age, span = 0.5) + education, data = Wage)

fits a two-term model, in which the first term is an interaction between year and age, fit by a local regression surface.

We can make predictions from gam objects, just like from lm objects, using the predict() method for the class gam. Here we make predictions on the training set.

preds <- predict(gam.m2, newdata = Wage)

Questions

• Fit a GAM on the Boston dataset to predict medv with a local regression for the lstat term, with a span of 0.5. Also add a degree-3 smoothing spline function of rm. Store the result in the variable gam1.
• Calculate predictions for medv on the training set. Store the results in the variable preds.

Assume that:

• The MASS library has been loaded
• The Boston dataset has been loaded and attached

(ignore: test Dodona)