[R-sig-ME] HPDinterval MCMCglmm and using transformed variables in MCMCglmm

[Jarrod Hadfield]

Hi,

Have you loaded lme4? This masks HPDinterval from coda. Try:

coda::HPDinterval(AMstem3$Sol[,"Dist_fr_A"])

At the moment only transformations (eg I(x^2)) only work left of the ~.

Cheers,

Jarrod


On 12 Jul 2010, at 12:13, Rebecca Ross wrote:

> Hi Jarrod,
>
> Thank you for the quick reply.
>
> I tried that one already and it hasnt made a difference (also I  
> haven't closed down R since it worked last night).
>
> I guess...close R down and start again?!
>
> Re logs - presumably I(x^2) is also not recognised (I had a go but  
> the x2 term did not appear in the Sol output)?
>
> Cheers,
>
> Rebecca
>
>
>
> ________________________________________
> From: Jarrod Hadfield [j.hadfield at ed.ac.uk]
> Sent: Monday, July 12, 2010 11:30 AM
> To: Rebecca Ross
> Cc: r-sig-mixed-models at r-project.org
> Subject: Re: [R-sig-ME] HPDinterval MCMCglmm and using transformed  
> variables in MCMCglmm
>
> Dear Rebecca,
>
> You probably need to load MCMCglmm (library(MCMCglmm)) or at least
> load coda. At the moment the response needs to be logged externally:
>
> my.data$ylog<-log(my.data$y)
>
> but the next version will "fix" this.
>
> Cheers,
>
> Jarrod
>
>
> On 12 Jul 2010, at 10:31, Rebecca Ross wrote:
>
>> Dear ME users,
>>
>> Apologies if this is a naive question, I am new to using MCMC glmm.
>> I have been working through the tutorial, using some of my own data,
>> and I thought I had followed the steps correctly. The model I have
>> been fitting is:
>>
>> AMstem3<-MCMCglmm(log(stem)~Dist_fr_A,random=~Block
>> +animal,pedigree=Ped,data=SapRO,prior=prior1.3,
>> family="gaussian")
>> Briefly, I want to see how the Distance variable (continuous)
>> affects stem weight, which I have log transformed as the raw data is
>> heteroscedastic. Block is field block, and I am fitting the animal
>> random effect to take into account different relatedness between
>> individuals (I'm not at the moment trying to estimate Va).
>>
>> When I tried this model last night, I could get the 95% CIs using
>> this function
>> HPDinterval(AMstem3$Sol[,"Dist_fr_A"])
>> HPDinterval(AMstem3$VCV)
>>
>> When I tried this morning, I get the following error message
>> Error in function (classes, fdef, mtable)  :
>> unable to find an inherited method for function "HPDinterval", for
>> signature "mcmc"
>> Can anyone explain what  has gone wrong? I've tried changing the
>> terms in the model, but that doesn't help, so I think I must have
>> changed something in R elsewhere? Or else missing something
>> completely obvious.
>>
>> Question 2: I dont think MCMCglmm is actually logging my stem
>> variable because the fixed effect parameter estimate from the model
>> above (AMstem3) are about a factor of 10 out from the fixed effect
>> estimate from
>> a similar model, fit in lme4, without the animal effect and very
>> similar to the same model using the raw stem weight (see below). Do
>> I need to specify the log transform differently?
>>
>> AMstem3c Dist_fr_A  posterior mode: -0.004236659
>> versus:
>> stem5a<-lme(log(stem)~Dist_fr_A,random=~1|
>> Block,data=SapRO,na.action=na.omit)
>> Fixed effect Estimate: Dist_fr_A   -0.0004442 (SE=0.00006136)
>> and:
>> stem5b<-lme(stem~Dist_fr_A,random=~1|
>> Block,data=SapRO,na.action=na.omit)
>> Fixed effect Estimate: Dist_fr_A   -0.004342 (SE=0.0006714)
>>
>> Thank you very much for your help,
>>
>> Rebecca Ross
>> DPhil Candidate,
>> University of Oxford.
>>
>>
>>
>>
>>      [[alternative HTML version deleted]]
>>
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
>>
>
>
> --
> The University of Edinburgh is a charitable body, registered in
> Scotland, with registration number SC005336.


-- 
The University of Edinburgh is a charitable body, registered in
Scotland, with registration number SC005336.