Skip to content

issue with using different error terms within the same model

1 message · Basil Iannone

Original
Basil Iannone
# 
Dear R users,

I am trying to analyze a data set from an experiment where I measured
nitrate over time. The study was set up in 5 blocks with four plots in
each (For now let?s assume that Block is a Fixed effect). Within each
block I randomly assigned one treatment combination of two possible
treatments (?Mulch? and ?Tilling?) at two levels each (absent or present).
I also measured relative water content (?RWC?) in each plot 5 times over
the growing season to use as a covariate. So a model for an analysis had I
only measured nitrate one time would have a sample size of 20, have df for
the error term of 12, and look something like (I am not interested in RWC
interactions):

        Nitrate ~ Block + RWC + Mulch*Till

But I measured nitrate 5 times over the growing season. So now I have a
sample size of 100 and I want to analyze the following model.

        Nitrate ~ Block + RWC + Mulch*Till + Week + Week:RWC +
Week*Mulch*Till

The issue is that I cannot figure out how to code the larger model in R so
that the between subject terms (ones before ?Week?) are estimated using df
for the error term of 12 (N = 20) and the within subject terms (?Week? and
later in the model) using the df for the error term of 78 (N = 100) (This
value may not be correct). Currently the models I use in R, including aov
and gls (in nlme) produce pseudo-replicated results for the between
subject terms, i.e., they are calculated using the larger df value
assuming that N = 100 not 20.

I would prefer to use the ?nlme? package for this analysis because I am
familiar with how to include variance structures if they are needed and I
could treat Block as a random effect if I convince myself that it in fact
is.

Thanks in advance for any advice that anyone can offer.