R help - Aug 2008 - output components of GAM

I would like to request help with the following:
I am trying to use a Generalized Additive Model (gam) to examine the density
distribution of fish as a function of latitude and longitude as continuous
variables, and year as a categorical variable. The model is written as:
 
gam.out   <-  gam(Density ~ s(Lat) + s(Lon) + as.factor(Year))
 
The fitted model prediction of the link function is gam.out$linear.predictors.
Presumably, gam.out$linear.predictors must be derived from some combination of
the original predictor variables (Lat, Lon, Year), their corresponding
coefficients and the intercept (gam.out$coefficients), and the smooth outputs
gam.out$smooth and/or gam.out$sp.
 
By comparison, for a glm model:
 
glm.out   <-  glm(Density ~ Lat + Lon + as.factor(Year))
 
this is simply:
 
glm.out$linear.predictors  =  glm.out$coefficients(“intercept”)  +
 glm.out$coefficients (“year”)  +  glm.out$coefficients(“lat”) x Lat  +
 glm.out$coefficients(“lon”) x Lon
 
My problem is that I cannot figure out how to get the equivalent from the gam
model. I would like to know how to ‘decompose’ gam.out$linear.predictors into
its components so that I can evaluate the effects of the different predictor
variables separately.
 
I would appreciate any comments that can help me with this.
 
Thank you,
 
Andreas Winter
Blacksburg, VA


      
	[[alternative HTML version deleted]]

is this gam::gam or mgcv::gam? If the latter, and if I understand what
you want, if gam.out contains your GAM model fitted using gam() from
mgcv, then:

out.term <- predict(gam.out, type = "terms")

gives you the contributions of each covariate on the response for the
fitted model. read :?predict.gam for details of what type = "terms"
actually returns.

Note that you should be wary of accessing the components of model fits
directly. Instead, get the various bits using accessor functions,
coefs() for $coefficients, fitted() for the fitted values etc. Often the
components don't hold exactly what you think but which are processed
correctly when accessed by the appropriate accessor function.

HTH

G

On Sat, 2008-08-02 at 18:47 -0700, Andreas Winter wrote:
> I would like to request help with the following:
> I am trying to use a Generalized Additive Model (gam) to examine the
> density distribution of fish as a function of latitude and longitude
> as continuous variables, and year as a categorical variable. The model
> is written as:
> 
> gam.out <- gam(Density ~ s(Lat) + s(Lon) + as.factor(Year))
> 
> The fitted model prediction of the link function is gam.out
> $linear.predictors. Presumably, gam.out$linear.predictors must be
> derived from some combination of the original predictor variables
> (Lat, Lon, Year), their corresponding coefficients and the intercept
> (gam.out$coefficients), and the smooth outputs gam.out$smooth and/or
> gam.out$sp.
> 
> By comparison, for a glm model:
> 
> glm.out <- glm(Density ~ Lat + Lon + as.factor(Year))
> 
> this is simply:
> 
> glm.out$linear.predictors = glm.out$coefficients(intercept) + glm.out
> $coefficients (year) + glm.out$coefficients(lat) x Lat + glm.out
> $coefficients(lon) x Lon
> 
> My problem is that I cannot figure out how to get the equivalent from
> the gam model. I would like to know how to decompose gam.out
> $linear.predictors into its components so that I can evaluate the
> effects of the different predictor variables separately.
> 
> I would appreciate any comments that can help me with this.
> 
> Thank you,
> 
> Andreas Winter
> Blacksburg, VA
> 
> 
>       
> 	[[alternative HTML version deleted]]
> 
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide
http://www.R-project.org/posting-guide.html
> and provide commented, minimal, self-contained, reproducible code.