Skip to content

Mixed model for count data with overdispersion

8 messages · Mehdi Abedi, Manabu Sakamoto, Christopher David Desjardins +2 more

Original
Mehdi Abedi
# 
Dear all,

I had quick search but it looks there is no simple way in lme4 or  nlme In
the case of overdispersion for count data,. How we can run mixed model for
count data with family of quasipoisson or maybe NB?

I my working on seeding emergence with 2 fixed factor (n=10) and i would
like to have my plot as replicate(n=5) as a random.

Warm regards,
Mehdi
Manabu Sakamoto
# 
Dear Mehdi,

You can use the function MCMCglmm in the package of the same name,
specifying family="poisson". MCMCglmm automatically accounts for over
dispersion in count data.

best regards,
Manabu
On 10 August 2015 at 06:54, Mehdi Abedi <abedimail at gmail.com> wrote:

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Mehdi Abedi
    

    

      
      #
    

  


  

      
Thanks Manabu,
It is a bit complicated for me but If i have this data:
Parameter: Totalseedling
fixed effect: Heatsmoke, cold
random effect: plot

I should do something like this?!


Model1<- MCMCglmm(Totalseedling ~ Heatsmoke *Cold, random =
~Plots,family="poisson", data = growthdata)
 summary( Model1)
It looks i can not get anova() here for output as well?


I am not familiar with other details in the MCMCglmm:

library( MCMCglmm)
Model1<- MCMCglmm(Totalseedling ~ Heatsmoke *Cold, random = ~Plot,
+ family = "poisson", data = growthdata, prior = prior,
+ verbose = FALSE, pr = TRUE)

Warm regards,
Mehdi

On Mon, Aug 10, 2015 at 12:48 PM, Manabu Sakamoto <manabu.sakamoto at gmail.com

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Christopher David Desjardins
    

    

      
      #
    

  


  

      
Hi,

You really should read about the MCMCglmm package before just using it. There are a couple of vignettes which I strongly suggest that you read prior to actually using MCMCglmm as they explain a lot.

https://cran.r-project.org/web/packages/MCMCglmm/vignettes/Overview.pdf <https://cran.r-project.org/web/packages/MCMCglmm/vignettes/Overview.pdf>
https://cran.r-project.org/web/packages/MCMCglmm/vignettes/CourseNotes.pdf <https://cran.r-project.org/web/packages/MCMCglmm/vignettes/CourseNotes.pdf>

Do note that you need to specify prior distributions or at least understand the default ones.

Chris

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        

      
      
    

        

  
  


  


      

    

    

      
      

        


  

        

      Mehdi Abedi
    

    

      
      #
    

  


  

      
Thanks Chris for lectures,
Working with MCMCglmm is like jumping from high school physics to Albert
Einstein lectures:). Hopefully i can digest this as a ecologist this
modeling part!
All the best

On Mon, Aug 10, 2015 at 1:46 PM, Christopher David Desjardins <

        
cddesjardins at gmail.com> wrote:

        

            

      
      
      
    

        

      
      
    

        

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Steve Walker
    

    

      
      #
    

  


  

      
An alternative is to use glmer with `family=Poisson` and an 
observation-level random effect.  I only skimmed this paper, but it will 
hopefully put you on to the main idea:

https://peerj.com/articles/616/

Cheers,
Steve

        
On 2015-08-10 5:27 AM, Mehdi Abedi wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

  
  


  


      

    

    

      
      

        


  

        

      Paul Johnson
    

    

      
      #
    

  


  

      
I'll second Steve's suggestion (which I think is the easiest, although assessing fit is tricky) and add another suggestion of fitting a negative binomial GLMM in glmmADMB.

Paul


Sent using CloudMagic<https://cloudmagic.com/k/d/mailapp?ct=pa&cv=7.0.42&pv=4.2.2>

        
On Mon, Aug 10, 2015 at 1:37 PM, Steve Walker <steve.walker at utoronto.ca<mailto:steve.walker at utoronto.ca>> wrote:

        
An alternative is to use glmer with `family=Poisson` and an
observation-level random effect.  I only skimmed this paper, but it will
hopefully put you on to the main idea:

https://peerj.com/articles/616/

Cheers,
Steve

        
On 2015-08-10 5:27 AM, Mehdi Abedi wrote:

            

      
      
      
    

        

      
      
    

            

      
      
      
    

        
_______________________________________________
R-sig-mixed-models at r-project.org mailing list
https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

  
  


  


      

    

    

      
      

        


  

        

      Mehdi Abedi
    

    

      
      #
    

  


  

      
Thanks Paul and Steve,
If i understood well you mean somethings like this!?:


library( glmmADMB)
Model2<- glmmadmb(Totalseedling~Heatsmoke *Cold+(1|Plots),data=growthdata,
family="nbinom1")
 summary(Model2)

Is it possible to get only main effect results with some function like
anova or similar in glmmADMB. I am mainly interested to only main effect(
Heatsmoke and Cold) than theirs levels results. I think anova function
doesn't play role in these advanced package :)
All the best,
Mehdi

On Mon, Aug 10, 2015 at 5:15 PM, Paul Johnson <paul.johnson at glasgow.ac.uk>
wrote: