Hello,
I have two gamm question (I am using gamm in mgcv).
1. In have, say 5 time series. Monthly data, 20 year. The 5 time series are
from 5 stations. The data are in vectors, so I have fitted something along the
lines of:
tmp<-gamm(Y ~ s(Year,by=station1)+s(Year,by=station2)+
s(Year,by=station3)+s(Year,by=station4)+
s(Year,by=station5)+
factor(station)*factor(month),
correlation=corAR1(form=~MyTime|station),
famliy=gaussian)
station is just a long vector with ones, twos, threes, fours and fives.
MyTime defines the order of time (and has values 1 to 240)
This model fits a Year smoother on each station and, and has one
auto-regressive parameter (whihc is about 0.3). How woul I allow for 5
different AR1 parameters (one per station)?
So far so good. It runs..and get the output. The problem is that the errors
from station 1 are correlated with those of station 2 (as they are close in
space). Same holds for other stations. The cross-correlation is about 0.5. How
do I build in this between-station correlation?
So..I have within station auto-correlation (dealt by the AR1 parameter), and
between station correlation.
Question 2: Why is Simon Wood using in his 2006 book only ML estimation and
not REML? I thought that REML was used to compare models with different random
components and ML estimation models with different fixed components?
Kind regards,
Piet Bell
---------------------------------
[[alternative HTML version deleted]]
______________________________________________
R-help@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
and provide commented, minimal, self-contained, reproducible code.
