Patrick Giraudoux
2009-Nov-04 11:24 UTC
[R] What happen for Negative binomial link in Lmer
Seems the message below and the thread have reveived no attention/answer. The output presented is quite tricky. Looks like if lmer (lme4 0.9975-10) has accepted a negative binomial link with reasonable estimates, although it was not designed for... What can one think about result validity ? Best Patrick Message: 34 Date: Thu, 29 Oct 2009 06:51:24 -0700 (PDT) From: "E. Robardet" <e.robardet@gmail.com> Subject: Re: [R] What happen for Negative binomial link in Lmer fonction? To: r-help@r-project.org Message-ID: <26113408.post@talk.nabble.com> Content-Type: text/plain; charset=us-ascii Thank you for your answers, I have an exemple of that i was using: m1a<-lmer(atpos~ninter+saison+milieu*zone+(1|code),family=neg.bin(0.429),method="Laplace",data=manu) summary(m1a) Generalized linear mixed model fit using Laplace Formula: atpos ~ ninter + saison + milieu * zone + (1 | code) Data: manu Family: Negative Binomial(log link) AIC BIC logLik deviance 125.1 147.6 -54.57 109.1 I think It was the version lme4 0.9975-10. Unfortunately, I have this version no more available on my computer.. I wonder if this old results are still valid.. Ben Bolker wrote:> > > > > > > > ROBARDET Emmanuelle wrote: > >> >> >> >> Dear R users, >> >> I'm performing some GLMMs analysis with a negative binomial link. >> >> I already performed such analysis some months ago with the lmer() >> >> function but when I tried it today I encountered this problem: >> >> Erreur dans famType(glmFit$family) : unknown GLM family: 'Negative >> >> Binomial' >> >> >> >> Does anyone know if the negative binomial family has been removed from >> >> this function? >> >> I really appreciate any response. >> >> Emmanuelle >> >> >> >> >> > > > > I would be extremely surprised if this worked in the past; to > > the best of my knowledge the negative binomial family has > > never been implemented in lmer. One could in principle > > do a glmmPQL fit with the negative binomial family > > (with a fixed value of the overdispersion parameter). > > glmmADMB is another option. > > Can you say which version etc. you were using??? > > > > Follow-ups should probably be sent to r-sig-mixed-models@r-project.org > > .... > > >-- View this message in context: http://www.nabble.com/What-happen-for-Negative-binomial-link-in-Lmer-fonction--tp26013041p26113408.html Sent from the R help mailing list archive at Nabble.com. [[alternative HTML version deleted]]
Patrick Giraudoux wrote:> > Seems the message below and the thread have reveived no attention/answer. > The output presented is quite tricky. Looks like if lmer (lme4 0.9975-10) > has accepted a negative binomial link with reasonable estimates, although > it was not designed for... > > What can one think about result validity ? > >Well, I can believe that lme4 0.9975-10 may have allowed families that were not hard-coded (e.g., negative binomial with a fixed overdispersion parameter). I would expect the results to be reasonable. However, lme4 has gone through a lot of changes. If you have an old copy around and your results pass all the sanity checks you can think of, I might go ahead and use them -- otherwise you will have to deal with glmmADMB, WinBUGS, AD Model Builder (or lme4 with quasipoisson family) ... -- View this message in context: http://old.nabble.com/Re%3A-What-happen-for-Negative-binomial-link-in-Lmer-tp26194594p26198718.html Sent from the R help mailing list archive at Nabble.com.