[R-sig-eco] MCMCglmm and power analysis

[Nicholas Lewin-Koh]

Hi Krista,
You might look at posterior predictive p-values and Bayes factors,
However, you are mixing your metaphors so to speak. You can't really
talk about power in a Bayesian context, It really is a frequentest
concept dealing with classical error rates. You can think about how much
information from the data modifies the posterior, but that has more to
do with model identification. You really need to decide if you are
going to be a frequentist or a Bayesan.

Nicholas 
> ------------------------------
> 
> Message: 3
> Date: Fri, 25 Feb 2011 20:58:13 -0500
> From: "Nichols, Krista M" <kmnichol at purdue.edu>
> To: "r-sig-ecology at r-project.org" <r-sig-ecology at r-project.org>
> Subject: [R-sig-eco] MCMCglmm and power analysis
> Message-ID:
> 	<8B3D151076714145AE9D54B657B3C118D67696F499 at VPEXCH06.purdue.lcl>
> Content-Type: text/plain
> 
> Hi All,
> 
> I have a question about using MCMCglmm with simulated data for a power
> analysis of heritability.  I'm specifically using MCMCglmm and another
> package called PEDANTICS to simulate phenotypes across a pedigree; the
> response data are binomial.  I need to calculate power over these
> simulated data sets, but am having a hard time grasping how to do this in
> a Bayesian analysis.  I understand that I can compare models in MCMCglmm
> with the DIC, but as far as I know, there is no formal way to test
> whether there is a 'significant difference' between two models with the
> DIC that would be analogous, say, to a likelihood ratio test in a normal
> linear mixed model.  Does anyone know that this is indeed true (i.e.
> whether the DIC has statistical properties that would allow formal
> testing for significance so that I could use this for a power analysis),
> or if there is some other way to determine the significance in comparing
> models for such a power analysis?  Note that I'm specifically interes!
>  ted in testing the 'significance' of a random effect (animal ID from a
>  pedigree analysis), so looking for overlap with zero using HPDinterval
>  will not work, as random effects are constrained to be greater than zero
>  in MCMCglmm.
> 
> Any help, thoughts on this will be greatly appreciated!
> 
> Thanks,
> Krista
> 
> 
> Krista M. Nichols
> Associate Professor
> Purdue University
> Departments of Biological Sciences & Forestry and Natural Resources
> 915 W State Street
> West Lafayette, IN  47907
> 765.496.6848 (w)
> 765.494.0876 (f)
> 
> 
> 	[[alternative HTML version deleted]]
> 
> 
> 
> ------------------------------
> 
> _______________________________________________
> R-sig-ecology mailing list
> R-sig-ecology at r-project.org
> https://stat.ethz.ch/mailman/listinfo/r-sig-ecology
> 
> 
> End of R-sig-ecology Digest, Vol 35, Issue 17
> *********************************************
>