1
$\begingroup$

I wonder is it possible to construct a generalized linear modelin in that way,

  • First, I will exclude the intercept term, which is standard for GAMs.
  • Second, I will predict my interested dependent variable(let say y1) with a parametric model using my dependent variable let say x.
  • Third, after eliminating the effect of parametric model estimates from the dependet variable, I will have a new dependent variable y2
  • Then, I will predict y2 by using x by a nonparametric method.

My aim is trying to capture the effect of x on y first by a parametric model. Then, capturing the remanining effects of x on y by a non parametric method, which are not captured by parametric model.

Is this a valid method?

I will be very glad for any help. Thanks a lot.

Namely:

Stage 1:

B=average(y)

y1=y-B

Stage 2:

calculate f(y1,x) where f is a parametric model(y1 is dep. and x is indep. variable)

Stage 3:

y2=y1-f(y1,x)

calculate y_pred=g(y2,x) where g is a nonparametric procedure(y2 is dep. and x is indep. variable)

Namely; y_pred=g(y1-f(y1,x),x)

$\endgroup$
  • $\begingroup$ It seems like you could specify the model as $y=x_1+x_2+\text{the rest that you do by gam}$, and that would do the whole process in one shot. $\endgroup$ – Dave May 30 at 19:10
  • $\begingroup$ @Dave, I revised my post. Maybe this time it may be more clear. $\endgroup$ – oercim May 30 at 19:51

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.