[R-sig-ME] Validation of mixed model for nested data

[Javier Martinez Lopez]

Dear list members,

I am performing mixed models to relate the axes of a correspondence analysis
of environmental variables and the axes of a multidimensional scaling
analysis based on species frequencies. I am using mixed models (function
'lme') because I have a temporal data set at 8 different sites, containing
only two dates per site, so that site is a random effect for the model, MDS
axe is the dependent variable and the environmental variables axe is the
fixed factor (random intercepts model ). As a result, I get the following
model and graph, which I attach hereby. The residuals graphs look OK. The
question is whether such model is meaningful taking into account that there
are only two points per group and each site has a very different intercept.
Could I even try a random slopes and intercepts model? My goal is to test
whether there is a general relationship between the environmental and
biological variables, not to predict it. Besides, is there a way of looking
for a goodness of fit for my model?

Thank you for any advice,

Javier
-------------- next part --------------
A non-text attachment was scrubbed...
Name: lme_mds2_cu1.png
Type: image/png
Size: 8260 bytes
Desc: not available
URL: <https://stat.ethz.ch/pipermail/r-sig-mixed-models/attachments/20101124/8a7b8466/attachment.png>