[R] doing zero inflated glmm for count data with fmr

[levyofi]

Thank you very much Ben.
I took you advice and run the lmer model with family="quasipoisson" and got
sigma~10. I guess this means that the data has overdispertion but not too
high (>15) for me to must use zero inflated model.
Am I right? 
I will also post to the r-sig-mixed-models specialty list...
Cheers,
Ofir.


Ben Bolker wrote:
> 
> 
> 
> levyofi wrote:
>> 
>> Hello R users,
>> Doing My PhD I collected count data which I believe is zero inflated. I
>> have run a statistical model with lmer and family=poisson and got
>> summary(model)@sigma=1 so I believe there is no overdispertion.  
>> 
> 
> You have been misled.  sigma is set to 1 by definition for the Poisson and
> binomial families.
> Try with family="quasipoisson" and see what you get.
> 
> 
> levyofi wrote:
>> 
>> I would like to use the fmr function from the 'gnlm' library but I just
>> cannot figure out from the examples in the help page and some forums out
>> there how to convert the lmer parameters to the one used in fmr...
>> 
>> I have these variables in the model:
>>   count: the number of logs in a foraging tray (this is the response
>> variable).
>>   ta: the ambient temperature at the foraging tray.
>>   habitat: the habitat type of the foraging tray.
>>   season: the season in which the experiment session took place (summer
>> or winter).
>>   moon: the moon phase (new or full).
>>   position: a random factor (I had 4 foraging stations)
>>   individual_id: a random factor indicating the individual foraged in the
>> tray.
>>   
>> This is the lmer parameters I have used:
>> model<-lmer(count~ta*habitat*season*moon + (1|individual_id) +
>> (1|position), data=countdata, family=poisson) 
>> 
> 
> I think (but am not sure) that "fmr" won't do what you want; it will fit
> zero-inflated
> neg binom, but not mixed-effect models.  "gnlm" in Lindsey's repeated
> package does
> mixed-effect models with neg binom, but not zero-inflation.  Are you sure
> you need
> zero-inflation after accounting for random effects?
> 
> glmm.admb in the glmmADMB package will do *most* of what you want, but ...
> not crossed random effects as you have specified above (it only allows for
> a single
> grouping factor, as far as I can see).
> 
> If you really want all of this (zero-inflated negative binomial, crossed
> random
> effects) your choices would seem to be (a) the full version of AD Model
> Builder
> (maybe?) or (b) WinBUGS ...
> 
>   I would strongly recommend that you forward further queries on this
> to the r-sig-mixed-models specialty list ...
> 
>   cheers
>     Ben Bolker
> 
> 
> 

-- 
View this message in context: http://www.nabble.com/doing-zero-inflated-glmm-for-count-data-with-fmr-tp23136570p23140450.html
Sent from the R help mailing list archive at Nabble.com.