?
Dear R users,
I'm currently analysing some data with the gam function from the mgcv package.
I'm looking at the relationship between spatially referenced budburst dates
(recorded as number of days from January 1st) and two continuous variables, and
their interaction, where they are found. I'm particularly interested in testing
the significance of the interaction, whilst accounting for spatial trends in
the data.
I've set up my model as such, with year as a random factor (10 years of data):
M1<-gam(budburstday~ti(v1)+ti(v2)+ti(v1,v2)+s(Easting,Northing,k=1000)+s(year,bs="re"),family=scat,data=phen)
I initially set up the model with the default Gaussian family, but the qqplot
looked a bit heavy tailed with the points on the left of the plot particularly,
curving down below the straight line. I have just found the (family=scat -
scaled t for heavy tailed data) option in a newer version of mgcv, which it
states are:
"for regression type models dependent on a single linear predictor, and with a
log likelihood which is a sum of independent terms, each coprresponding to a
single response observation. Usable only with gam, with smoothing parameter
estimation by "REML" or "ML" (the latter does not integrate the unpenalized and
parameteric effects out of the marginal likelihood optimized for the smoothing
parameters)."
Question: I can find very few examples using this scat family, and was hoping
someone could tell me whether the model specified above is theoretically sound
(i.e. is it okay to include these different smooth types (ti, s & bs="re") with
this family)?
Any advice would be greatly appreciated. Thanks in advance.
Robin
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org mailing list -- To UNSUBSCRIBE and more, see
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.
