GAMM4: In mer_finalize(ans) : false convergence (8)

saba <saba.ghotbi at ...> writes:

Hi

I have read your comment on ( In mer_finalize(ans)), and some questions
raised for me. It would be your kind if advise me about all 
or some of them:

I'm not sure who you're addressing (this is a mailing list), but I'll
try.

In introducing the random effect is it important that
 repeated measurements
exist per individuals? And why?

It depends on the model.  If the dispersion parameter is estimated
(as in a linear mixed model fitted with lmer or a Gamma or Gaussian
model fitted with glmer), then a one-measurement-per-individual
experimental design will probably end up confounding the dispersion
parameter (or residual variance in the case of lmer fits) with the
random effect.  Hopefully you'll get an error or a warning message in
this case, but it is possible to trick lme4.
  If the dispersion parameter is fixed (binomial/Poisson GLMMs) then
an observation-level random effect is a useful way to model overdispersion.
See http://glmm.wikidot.com/faq

Does R consider the repeated values in a group and calculate
 the variances?

Not sure what you mean here.  You may want to read e.g. 
Pinheiro and Bates 2000, or some other text on mixed models, for
the basic theory of what mixed-model software is estimating.

Best regards

saba