[R-sig-ME] Doubt about including random effects or not

V. Coudrain v_coudrain at voila.fr
Tue Nov 12 09:00:54 CET 2013


Dear Ben Bolker,

Thank you very much for your help.

> * I don't quite understand why you don't just calculate average
> specialization per site; presumably Isolation and habitat amount are
> site-level covariates? If you have different sample sizes per site, you
> could calculate the mean and std. dev. of specialization and use the
> weights= argument to inverse-variance weight ... (see Murtaugh 2007
> _Ecology_ for arguments in favor of aggregating when analyzing nested
> designs). (It's possible that one of your covariates varies within site,
> which would make this aggregation infeasible.)

Just to avoid confusion: isolation is a factor with two levels: isolated or not isolated (15 sites each), and habitat amount has a single value pro site. These two 
variables have limited collinearity.

I thought about using the mean, but what I am really interested in is to know is the distribution of specialization index, i.e., if species with high specialization occur 
mostly in connected sites compared to isolated sites, or if their abundance increases with increasing habitat amount. Just taking the mean won't show me this. 
However by reading you I think that looking at the spread (variance) of the specialization index may be a solution. Increasing variance should indicate the 
presence of species with more extreme specialization values. But I am not sure how to test for this, and if I can specify a one-tailed test (if variance towards 
higher specialization values is related to the isolation and habitat amount covariates).

> * I also don't quite understand why you expect specialization to be
> directly proportional to abundance? Is abundance a species:site-level
> covariate, or an overall (site-level) covariate (I think the latter)? I
> would consider just putting insect abundance in as a covariate (i.e.
> allow for some dependence, don't require direct proportionality)... ?
> 

No I think, this s an error from my side. Anyway think I misunderstood offset, which should anyway be used for a response variable that is a count (not the fact 
in my case. As you suggest it, it is certainly more appropriate to put insect abundance as a covariate.

Valérie


___________________________________________________________
Qu'y a-t-il ce soir à la télé ? D'un coup d'œil, visualisez le programme sur Voila.fr http://tv.voila.fr/programmes/chaines-tnt/ce-soir.html



More information about the R-sig-mixed-models mailing list