I measured nitrate concentration and primary production (PP) biweekly for
23 months in one headwater stream. I would like to use linear regression
to determine if PP is related to nitrate concentration. My dataframe is
called "data" and consists of the vectors Rdate, PP, and nitrate. Rdate
is the observation date in class "date" and PP is primary production.
I first fit a linear model with uncorrelated residuals using gls:
M.lm <- gls (PP ~ nitrate, data=data)
I then looked for evidence of auto-correlation using acf:
acf (residuals ( M.lm, type = 'normalized'))
The plot showed a clear pattern, so I reran the gls using an AR-1
auto-correlation structure:
M.AR1 <- gls (PP ~ nitrate, correlation = corAR1 (form =~Rdate),
data=data)
I then compared the AIC values of the M.lm and M.AR1:
AIC(M.lm, M.AR1)
M.AR1 had a much lower AIC score than M.lm, suggesting that it is the
better model.
I then inspected a plot of the standardized residuals against the fitted
values for M.AR1:
plot (M.AR1)
This plot shows a pattern of decreasing residual spread with increasing
nitrate concentration, a clear indication on non-homogeneity.
My questions are:
Is the M.AR1 model appropriate for the question I am trying to answer? All
examples of AR-1 residual structure I could find are used in the context
of testing for a time effect. I am not interested in a time effect. Is
is still appropriate to use the AR1 auto-correlation structure?
If M.AR1 is OK, should I be worried about non-homogeneity in the
residuals? If so, how should I go about dealing with it? Could I add a
new variance structure to the model (e.g., varExp).
The approach outlined here is drawn from my interpretation of chapters 4
and 6 of Zuur et al. 2009. Mixed effects models and extensions in ecology
with R.
Thanks,
Jake
Research Ecologist
US EPA
[[alternative HTML version deleted]]
______________________________________________
[email protected] 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.
