Dear list members,
If I understand it well, including random effects in a model (i.e. using
mixed models) allows us to compare data coming from different sampling units
and to overcome independence assumptions. I am looking for significant
correlations between environmental and biological variables with a data set
of sampling points coming from different sites. Since I have sampled the
frequency of one plant species and the cover of surrounding land uses at two
different dates at every site (8 sites), I consider that my data are not
independent because species frequency at a latter time depends on what it
was before at each site. Sampling points at every site are not the same,
however, so I do not think this is a time series design. I built a model
putting the species frequency as the dependent factor, the land use variable
as the fixed factor and the site as the random factor in a glmer model using
binomial family. Later on, I compared the deviance of this model with the
one of the 'null' model, where the fixed factor is constant (~1), to get
some kind of goodness of fit (r2). Do you think that this is a correct
approach? am I accounting for the grouped nature of my data set by doing so?
Thank you very much!
best regards,
Javier
[[alternative HTML version deleted]]