Skip to content

Ordinal categorical variable as a random effect in MCMCglmm

3 messages · Manabu Sakamoto, Jarrod Hadfield

#
Dear Manabu,

Could you explain what Quality is? If it is ordered it is hard to see  
why you would be fitting it as random effect? Is this really your  
response variable?

Cheers,

Jarrod



Quoting Manabu Sakamoto <manabu.sakamoto at gmail.com> on Wed, 29 Oct  
2014 11:16:48 +0000:

  
    
#
Dear Jarrod,

Thanks for your response. Quality is a ranking in the
completeness/missingness of the data; a higher quality data point is scored
higher. This is an attempt to control for potential missing information
from the response variable. The response is a count (hence the family being
Poisson), and the idea is that for each taxon on a phylogeny, there is a
count variable, but that count could potentially be under-estimated based
on the Quality of the data associated with each taxon. But in reality, this
potential under-estimation is unknown and unmeasurable so Quality is just
one attempt to control for this somewhat-known uncertainty.

many thanks,
Manabu
On 3 November 2014 14:22, Jarrod Hadfield <j.hadfield at ed.ac.uk> wrote:

            

  
    
#
Dear Manabu,


I suggest

prior<-list(R=list(V=1, nu=0.002), G=list(G1=list(V=1, nu=1, alpha.mu  
= 0, alpha.V = 25^2)))

m1<-MCMCglmm(counts~Quality, random=~Taxon, ginverse=list(Taxon=Ainv),  
family="poisson", prior=prior, ....)

as a first stab.  MCMCglmm can't handle ordered predictors, so you  
could try just fitting Quality as a standard factor, or fitting it as  
continuous?

Cheers,

Jarrod


Quoting Manabu Sakamoto <manabu.sakamoto at gmail.com> on Mon, 3 Nov 2014  
14:35:40 +0000: