[R] GAMS and interactions

[simona.racioppi]

Dear all,
I am fitting a generalized additive model with several discrete and continuous predictors, using the library mgcv. 

Basically, I have two questions:

1.
If I model the response variable y with f(x), z and f(x):z, where f(x) has 1 edf and f(x):z has 4 edf, can I model x parametrically, that is considering x and not f(x), and keep in my model f(x):z (note: x and z are continuous)? In other words, if I am going to keep f(x):z, should I keep f(x) to be consistent, even if f(x) has 1 edf??

2.
If I am using a thin plate regression spline, f(w,t), should I put f(w) and f(t) in the equation? I think I should consider f(w,t) only 
 but I am not sure that is equivalent to f(w) + f(t) + f(w,t).

Thank you very much,

Simona




------------------------------------------------------
Leggi GRATIS le tue mail con il telefonino i-mode™ di Wind
http://i-mode.wind.it/