Hi,
I am looking at the effects of two explanatory variables on chlorophyll.
The data are an annual time-series (so are autocorrelated) and the
relationships are non-linear. I want to account for autocorrelation in
my model.
The model I am trying to use is this:
Library(mgcv)
gam1 <-gam(Chl~s(wintersecchi)+s(SST),family=gaussian,
na.action=na.omit, correlation=corAR1(form =~ Year))
the result I get is this:
Family: gaussian
Link function: identity
Formula:
CPRChl ~ s(wintersecchi) + s(SST)
Parametric coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) 3.57000 0.05061 70.54 <2e-16 ***
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Approximate significance of smooth terms:
edf Est.rank F p-value
s(wintersecchi) 2.445 5 4.672 0.00887 **
s(SST) 2.408 5 4.301 0.01237 *
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
R-sq.(adj) = 0.676 Deviance explained = 75.4%
GCV score = 0.074563 Scale est. = 0.053781 n = 21
The result look good - significant, with a lot of deviance explained,
but I am not convinced the model is actually accounting for the
autocorrelation (based on the formula in the results). How can I tell?
Many thanks,
Dr Abigail McQuatters-Gollop
Sir Alister Hardy Foundation for Ocean Science (SAHFOS)
The Laboratory
Citadel Hill
Plymouth UK PL1 2PB
tel: +44 1752 633233
[[alternative HTML version deleted]]
______________________________________________
R-help@r-project.org 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.
