R help - Nov 2004 - predict.gam problem

-----------------
Mac OS X 10.3
R 2.0.1
gam 0.9.2
-----------------

Greetings

I am facing difficulties in prediction on using Trevor Hasties GAM package. 

How can one interpret the response from predict.gam?

Documentation on the type-parameter:
  The default (link) produces predictions on the scale of the additive 
  predictors, ... 

  If "response" is selected, the predictions are on the scale of the
  response, and are monotone trans-formations of the additive predictors,
  using the inverse link function.

Does "on the scale of the response" ammount to residals? 

All I want to do is get the predicted response from running my GAM model
on new data. The results given are way off, but I am not able to dechipher 
the documentation, nor use any scale information.


Any help will be highly appreciated.


All the best

Jon Egil Strand




-- 

Jon Egil Strand
jes at luretanker.no
Phone: +47 45030081

I thought the documentation you quoted is rather clear.  The GAM is
something like:

g(Y) = sum f_j(x_j)

so predict.gam() returns sum f_j(x_j) by default, and if
type="response",
g^-1(sum f_j(x_j)).  This is similar to predict.glm, I believe.

HTH,
Andy
> From: Jon Egil Strand
> 
> 
> -----------------
> Mac OS X 10.3
> R 2.0.1
> gam 0.9.2
> -----------------
> 
> Greetings
> 
> I am facing difficulties in prediction on using Trevor 
> Hasties GAM package. 
> 
> How can one interpret the response from predict.gam?
> 
> Documentation on the type-parameter:
>   The default (link) produces predictions on the scale of the 
> additive 
>   predictors, ... 
> 
>   If "response" is selected, the predictions are on the scale of
the
>   response, and are monotone trans-formations of the additive 
> predictors,
>   using the inverse link function.
> 
> Does "on the scale of the response" ammount to residals? 
> 
> All I want to do is get the predicted response from running 
> my GAM model
> on new data. The results given are way off, but I am not able 
> to dechipher 
> the documentation, nor use any scale information.
> 
> 
> Any help will be highly appreciated.
> 
> 
> All the best
> 
> Jon Egil Strand
> 
> 
> 
> 
> -- 
> 
> Jon Egil Strand
> jes at luretanker.no
> Phone: +47 45030081
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! 
> http://www.R-project.org/posting-guide.html
> 
>