Interactions between categorical terms in GAM (mgcv)

Question

I'm trying to fit a model using the mgcv package and I need to include interactions between categorical terms. I know how to implement interactions between continuous and categorical variables as

s(x1, by=c1)

but I don't know how to change this for two categorical terms and I couldn't find anything.

Answer 1

You can add parametric terms to the formula just as you would in a linear model or a GLM.

gam(y ~ f1 + f2 + f1:f2 + s(x1), data = df, method = "REML")

or via the shorthand

gam(y ~ f1 * f2 + s(x1), data = df, method = "REML")

would be a model with main effects plus interactions of the two categorical variables f1 and f2 plus a smooth effect of continuous variable x1 .

Anything you can do with parametric terms in lm() or glm() can be used in the formula for gam() , gamm() , bam() , and gamm4::gamm4() .