[R] kernel regression

[Romain.Mayor at ville-ge.ch]

Dear R community,

I have tried to use the kernel regression in the "NP" package. I work in a
population biology study and the regression function ("npr") dealing with
continuous and categorical variables seems to be appropriate for such study
with non parametric data.
I have 3 questions:
First: Is there a way to have for each predictor a value indicating the
explanation of each predictor in the model. (not only the R2 for the whole
model but for each predictor).
Second: What should be the correct tolerance value for the bandwidth. I use
0.01.
Third: What is the suitable bootstrap repetition number ("but.num") in the
npsigtest function? I use the default value ("499"), with 7 variables(5
continous,2 categorical) and 1600 data point, it takes 23 hours to compute
the test and normally for publication purpose 999 bootstraps is the minimal
number of repetitions. Is it possible to do the test on a set of suitable
observation?

Could somebody also give me some references of publications using such
analysis in ecological domain?

Than you for answers and best regards.

Romain Mayor.