[R-sig-ME] Error in .local(object, n, verbose, ...) : Update not yet written
bbolker at gmail.com
Fri Nov 23 13:54:06 CET 2012
joana martelo <jmmartelo at ...> writes:
> Many thanks Ben!
> What I would like is to get the parameter estimates and confidence intervals
> based both on the fixed and random effects, and not just on the fixed
> effects. Is parametric bootstrapping the only alternative?
> Thanks again!
Yes, unless you use Bayesian methods (MCMCglmm etc.), or
glmmADMB with MCMC sampling ('pseudo-Bayesian'), or you
manage to get your GLMMs working with profiling in the
development version of lme4 (experimental!)
(As always I would be happy for enlightenment from other
list readers if they have any ideas! Is there a straightforward
way to do this in SAS or AS-REML?)
> mcmcsamp has never worked for GLMMs and at this point probably never will,
> because writing a reliable post-hoc MCMC sampler (i.e., one that doesn't get
> stuck at low parameter valuables or have other undesirable behaviour) has
> turned out to be really, really hard.
> If you really want the posterior distribution of the parameters, it should
> be pretty easy to fit your model using MCMCglmm. If you want confidence
> intervals it's unfortunately harder than it should be, but you might try
> parametric bootstrapping ... (the pbkrtest package has parametric
> bootstrapping, but it is designed for parameter testing via model comparison
> rather than for finding posterior/sampling distributions of parameters).
> Ben Bolker
More information about the R-sig-mixed-models