Hi,
I apologize in advance if these questions are too basic, but I’m a new
comer to the mixed-effects models world:
I’m analyzing several time-series of different length of fish catches
in several rivers spread along a latitudinal gradient and modelling
them with a random intercept and slope model as follows:
Model1<-
lme
(Y
~
X1
+
X2
+
X1
*X2
,random=list(river=pdDiag(~X1)),data=mydata,correlation=corAR1(form=~1|
river))
Where Y is the dependent variable (catches), X1 is a continuous
variable (temperature) and X2 is a factor variable (presence/absence
of dams). My question is related with the assumption that there is no
correlation between residuals of different time series, but that could
be violated if fishes on one river are affecting those on other
rivers. Calculating the correlation coefficients between the residual
time-series show some degree of correlation between proximate rivers.
Whatever the biological/environmental reason for this could be, is
there any way of allowing both temporal and spatial correlation in the
same model? Is it my reasoning correct?
My second question is more general: Does the stratification variable
used in variance functions like varIdent, or optionally in varPower,
varExp…, need to appear as well in the fixed and/or the random part of
the model? Or, one could use a completely different stratification
variable?
Many thanks for your expert input.
Jaime.
[[alternative HTML version deleted]]