[R-sig-ME] partly nested repeated measurements
Quentin Schorpp
quentin.schorpp at ti.bund.de
Tue Sep 15 14:45:16 CEST 2015
Hello,
I made a field study, where i took samples for soil fauna.
The Design is complicated and i want to make sure that i choose the
right model formulation.
The study comprises 18 sites (SiteID) at 7 locations (loc),
The number of sites per location varies between 1 and 5
On each site i took 4 samples (SiteID, too).
I sampled 12 of the 18 sites in two consecutive years (Sampling
Campaigns = SC). Hence I've got repeated measurements.
These fields are assigned to one of 4 Age Classes (AC) of a
perennial energy crop. Age Class is derived from the year of establishment.
Each Age Class has 3 Replicates.
Another Class consisting of an annual energy crop was sampled in 3
Replicates in year one and 3 different replicates in year two.
However I want to treat this Class as Age Class Zero.
The SiteID is nested in the Age Class and partly nested in SC (only for
SiteID 13-18)
The Aim of the investigation was to substitute time for space. So i want
to know if there is an development throughout the Age Classes (i.e.
increase in abundance, different community structures, Gain in
individual weights, increase in functional groups, etc).
And if there is an development, I want to know if it is resembled in the
two consecutive years of sampling.
My Model formula is:
model <- glmer(Y ~ AC*SC + (1|SiteID), family="poisson", data)
or in a more general coding:
model <- glmer(Abundance ~ Trt*time + (1|subject),
family="poisson", data)
Is this the right way?
The random effect accounts for repeated measurements.
I did not consider location, since it was not fully factorial.
The Analysis increases in complexity regarding the response variable:
Indiviudal weights have a right skewed distribution, but contain zeroes
(gamma distribution is not possible here)
Abundance of nematode families were identified in subsamples of 100
Individuals in total per sample, hence family abundances represent
rather proportions.
Abundances of families are community data and should be analysed with
multivariate methods.
I read a lot of posts and book chapters for the Analysis of longitudinal
data, however there is a vast amount of methods and method fine tuning
and again and again i face difficulties, error messages, many different
results from different approaches. Now, I just want to know the truth,
but i've got severe problems to take a decision on which Analysis to
choose.
Kind regards,
Quentin
More information about the R-sig-mixed-models
mailing list