R help - Sep 2009 - How to extract partial predictions, package mgcv

Dear package mgcv users,



I am using package mgcv to describe presence of a migratory bird species as 
a function of several variables, including year, day number (i.e. 
day-of-the-year), duration of survey, latitude and longitude. Thus, the 
"global model" is:



global_model<-gam(present ~ as.factor(year) + s(dayno, k=5) + s(duration, 
k=5) + s(x, k=5) + s(y, k=5), family = binomial(link = logit), data = 
presence, gamma =1.4)



The model works fine, suggesting that all the variables have strong, 
non-linear, influences on the probability of detecting the species. My main 
interest is in the effect "dayno" has on presence, given the inclusion
of
the other explanatory variables. Thus, I would like to extract the values of 
the partial prediction of "dayno" and its associated 2 standard errors
above
and below the main effect, as shown by the code "plot(global_model)".



I have tried to extract the values by using 
"fitted.values(global_model,dayno)", but when plotted, the figure
doesn't
look like the partial prediction for "dayno". Instead, it gives me a
very
scattered figure ("shotgun effect").



Has anyone tried to extract the partial predictions? If so, please could you 
advise me how to do this?



Thanks,



Staffan



P.S.. For comparison, please have a look at Simon Woods paper in R News, 
1(2):20-25, June 2001, especially the figures showing the partial 
predictions of Mackerel egg densities. Using those graphs as an example, I 
would like to extract the partial predictions for e.g. "s(b.depth)",
given
the inclusion of the other variables.

--
Staffan Roos, PhD
Research Ecologist
BTO Scotland

Staffan,

Take a look at ?predict.gam. You need to use type="terms" with se=TRUE
to get
the quantities that you need.

Simon

ps. with binary data,  method="REML" or method="ML" for the
gam fit are often
somewhat more reliable than  the default.

On Monday 14 September 2009 19:30, Staffan Roos wrote:
> Dear package mgcv users,
>
>
>
> I am using package mgcv to describe presence of a migratory bird species as
> a function of several variables, including year, day number (i.e.
> day-of-the-year), duration of survey, latitude and longitude. Thus, the
> "global model" is:
>
>
>
> global_model<-gam(present ~ as.factor(year) + s(dayno, k=5) +
s(duration,
> k=5) + s(x, k=5) + s(y, k=5), family = binomial(link = logit), data >
presence, gamma =1.4)
>
>
>
> The model works fine, suggesting that all the variables have strong,
> non-linear, influences on the probability of detecting the species. My main
> interest is in the effect "dayno" has on presence, given the
inclusion of
> the other explanatory variables. Thus, I would like to extract the values
> of the partial prediction of "dayno" and its associated 2
standard errors
> above and below the main effect, as shown by the code
"plot(global_model)".
>
>
>
> I have tried to extract the values by using
> "fitted.values(global_model,dayno)", but when plotted, the figure
doesn't
> look like the partial prediction for "dayno". Instead, it gives
me a very
> scattered figure ("shotgun effect").
>
>
>
> Has anyone tried to extract the partial predictions? If so, please could
> you advise me how to do this?
>
>
>
> Thanks,
>
>
>
> Staffan
>
>
>
> P.S.. For comparison, please have a look at Simon Woods paper in R News,
> 1(2):20-25, June 2001, especially the figures showing the partial
> predictions of Mackerel egg densities. Using those graphs as an example, I
> would like to extract the partial predictions for e.g.
"s(b.depth)", given
> the inclusion of the other variables.
>
> --
> Staffan Roos, PhD
> Research Ecologist
> BTO Scotland
>
> ______________________________________________
> 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.
-- 
> Simon Wood, Mathematical Sciences, University of Bath, Bath, BA2 7AY UK
> +44 1225 386603  www.maths.bath.ac.uk/~sw283