[R] R package: autocorrelation in gamm

Guillermo Aravena Cuevas guillermo.aravena at ehu.es
Thu Oct 16 14:32:44 CEST 2008

Dear users
I am fitting a Generalized Additive Mixed Models (gamm) model to  
establish possible relationship between explanatory variables (water  
temperature, dissolved oxygen and chlorophyll) and zooplankton data  
collected in the inner and outer estuarine waters. I am using monthly  
time-series which are auto-correlated.

In the case of the inner waters, I have applied satisfactoryly (by ACF  
and PACF) an corAR1() as follows:
gamm.zoo <- gamm(zoo~s(water.temperature) + s(dissolved.oxygen) +  
s(chlorophyll), data=datos,family=gaussian,corr=corAR1() )

But in the case of the outer estuarine waters, our data has a seasonal  
auto-correlation. I have tried to applied corARMA(p,q), but it is not  
completely satisfactory because I have an autocorrelation of order 1  
and 12 as an seasonal ARIMA=(1,0,0)(1,0,0)

I have tried to model it with

corr=corARMA(p=12, fixed=c(NA,0,0,0,0,0,0,0,0,0,0,NA,NA)))

However, I am not completely sure I am modelling what I should.

I suspect the autocorrelation structure in the error term has been  
taken into account with corAR1/corARMA, but I cannot work it out how  
it is the random structure. I do not need to incorporate a random  
structure in my model, only need to take into account the  
autocorrelation term structure in the error term, but I am not sure  
that the model aforementioned is doing what I am trying to do.
Any help would be appreciated.

Thanks in advance

More information about the R-help mailing list