R help - Mar 2010 - Problem specifying Gamma distribution in lme4/glmer

Dear R and lme4 users-

I am trying to fit a mixed-effects model, with the glmer function in
lme4, to right-skewed, zero-inflated, non-normal data representing
understory grass and forb biomass (continuous) as a function of tree
density (indicated by leaf-area).  Thus, I have tried to specify a
Gamma distribution with a log-link function but consistently receive
an error as follows:
>
total=glmer(total~gla4+(1|plot)+(1|year/month),data=veg,family=Gamma(link=log))
> summary(total)
Error in asMethod(object) : matrix is not symmetric [1,2]

I have also tried fitting glmm's with lme4 and glmer to other
Gamma-distributed data but receive the same error.  Has anyone had
similar problems and found any solutions?

Thank you for your input.  Best regards,
___________________________________
Matt Giovanni, Ph.D.
NSERC Visiting Research Fellow
Canadian Wildlife Service
2365 Albert St., Room 300
Regina, SK S4P 4K1
306-780-6121 work
402-617-3764 mobile
http://sites.google.com/site/matthewgiovanni/

Matthew Giovanni <matthewgiovanni <at> gmail.com> writes:
> 
> Dear R and lme4 users-
> 
> I am trying to fit a mixed-effects model, with the glmer function in
> lme4, to right-skewed, zero-inflated, non-normal data representing
> understory grass and forb biomass (continuous) as a function of tree
> density (indicated by leaf-area).  Thus, I have tried to specify a
> Gamma distribution with a log-link function but consistently receive
> an error as follows:
> 
> total=glmer(total~gla4+(1|plot)+
    (1|year/month),data=veg,family=Gamma(link=log))
> summary(total)
> Error in asMethod(object) : matrix is not symmetric [1,2]
> 
> I have also tried fitting glmm's with lme4 and glmer to other
> Gamma-distributed data but receive the same error.  Has anyone had
> similar problems and found any solutions?
 1. probably best to post questions like this to 
r-sig-mixed-models at r-project.org

 2. haven't seen this particular problem.  Can you please
provide a reproducible example (post your data, or a small
subset of your data, or a simulated example that displays
the same problem), and give the results of
the sessionInfo() function?

f <- factor(rep(1:10,each=10))
x <- runif(100)
dat <- data.frame(x,f)
library(lme4)
 [snip messages]
g1 <- glmer(x~1+(1|f),data=dat,family=Gamma(link=log))
  Generalized linear mixed model fit by the Laplace approximation 
  [...]
summary(g1)
  [ works fine]

sessionInfo()
  R version 2.10.1 (2009-12-14) 
  i486-pc-linux-gnu 

  [snip]

  other attached packages:
  [1] lme4_0.999375-32-2 Matrix_0.999375-38 lattice_0.18-3    

 3. zero-inflated data may not be particularly well-represented
by a Gamma distribution: if you actually have a significant number
of exactly-zero values, you may want to analyze your data in two
stages, first as a presence-absence problem and then as a conditional
density (i.e., what is the distribution of the non-zero values)?

Ben Bolker wrote:
> 
> 
>  3. zero-inflated data may not be particularly well-represented
> by a Gamma distribution: if you actually have a significant number
> of exactly-zero values, you may want to analyze your data in two
> stages, first as a presence-absence problem and then as a conditional
> density (i.e., what is the distribution of the non-zero values)?
> 
> 
Interesting idea. Do you know of a example where this was done (independent
of lmer)? We have similar data, were people are either symptom free (50%
with score 0), and the rest is smoothly distributed.

Dieter


-- 
View this message in context:
http://n4.nabble.com/Problem-specifying-Gamma-distribution-in-lme4-glmer-tp1676344p1676746.html
Sent from the R help mailing list archive at Nabble.com.

Dieter Menne <dieter.menne <at> menne-biomed.de> writes:
 
> Ben Bolker wrote:
> >  3. zero-inflated data may not be particularly well-represented
> > by a Gamma distribution: if you actually have a significant number
> > of exactly-zero values, you may want to analyze your data in two
> > stages, first as a presence-absence problem and then as a conditional
> > density (i.e., what is the distribution of the non-zero values)?
> [...] Do you know of a example where this was done (independent
> of lmer)?  [...]
  Nothing springs to mind, but it seems sensible.

  Ben Bolker

On Mar 21, 2010, at 5:16 PM, Ben Bolker wrote:
> Dieter Menne <dieter.menne <at> menne-biomed.de> writes:
>
>> Ben Bolker wrote:
>
>>> 3. zero-inflated data may not be particularly well-represented
>>> by a Gamma distribution: if you actually have a significant number
>>> of exactly-zero values, you may want to analyze your data in two
>>> stages, first as a presence-absence problem and then as a  
>>> conditional
>>> density (i.e., what is the distribution of the non-zero values)?
>
>> [...] Do you know of a example where this was done (independent
>> of lmer)?  [...]
>
>  Nothing springs to mind, but it seems sensible.
I thought this was what hurdle and ZIF models were supposed to handle  
gracefully?

-- 
David.
>
>  Ben Bolker

On Mar 21, 2010, at 5:16 PM, Ben Bolker wrote:
> Dieter Menne <dieter.menne <at> menne-biomed.de> writes:
>
>> Ben Bolker wrote:
>
>>> 3. zero-inflated data may not be particularly well-represented
>>> by a Gamma distribution: if you actually have a significant number
>>> of exactly-zero values, you may want to analyze your data in two
>>> stages, first as a presence-absence problem and then as a  
>>> conditional
>>> density (i.e., what is the distribution of the non-zero values)?
>
>> [...] Do you know of a example where this was done (independent
>> of lmer)?  [...]
>
> Nothing springs to mind, but it seems sensible.
I thought this was what hurdle and ZIF models were supposed to handle  
gracefully?

-- 
David.
>
> Ben Bolker