[R-sig-ME] nlme: varIdent

[Ben Bolker]

D Gill <oceancurrents at ...> writes:

> 
> Hello,  I am a student working on fisheries data using a mixed linear
> model. I have a question about the varIdent function. I have information on
> 140 fishers at nine sites, where the only random effect is for the site. As
> I do not have equal variances between sites, is it correct to use the
> varIdent argument for the sites? Do I lose any information by doing it this
> way?
> 
> M1<-lme(resp~......,random = ~1|Site,weights = varIdent(form =~1|Site) ))
> 

  This seems reasonable to me.  However, note that you might
want to consider (variable*site) interactions of any predictor variables
that vary within sites (see e.g. Schielzeth and Forstmeier 2009) [e.g.
if your measurement is total catch, effort is a covariate, and CPUE
varies among sites, you would want effort|Site in your random
effects specification].
   I'm assuming you have a single measurement for each fisher -- 
otherwise you should also add a random effect for fishers ...

  Ben Bolker