[R-sig-ME] Meta-analysis for heritability using MCMCglmm?

Jarrod Hadfield j.hadfield at ed.ac.uk
Fri Dec 26 07:58:13 CET 2014

Hi Jackie,

The data are not binomial they are continuous: a beta distribution is  
probably most appropriate for continuos observations bounded by 0 and  
1. However, although heritabilities are bounded by 0 and 1,  
heritability estimates are not necessarily so, depending on the method  
of inference (for example it would be possible to get a negative  
parent-offspring regression, either by chance or through certain types  
of maternal effect).

We have just finished a meta-analysis of h2 estimates and just treated  
them as Gaussian. The distribution of the residuals wasn't far off and  
I think the conclusions are robust to the distributional assumptions.  
Have you checked your residuals - do they look badly non-normal?



Quoting Ken Beath <ken.beath at mq.edu.au> on Wed, 24 Dec 2014 12:30:03 +1100:

> If you have the original data giving the numerator and denominator for the
> proportion then it is binomial data, and can be modelled in a met-analysis.
> I don't know if this can be done with MCMCglmm but should be possible with
> STAN, JAGS or BUGS. All will require a bit of effort in setting up the
> model.
> On 24 December 2014 at 07:17, Jackie Wood <jackiewood7 at gmail.com> wrote:
>> Dear R-users,
>> I am attempting to conduct a meta-analysis to investigate the relationship
>> of narrow-sense heritability with population size. In previous work, I have
>> used MCMCglmm to conduct a formal meta-analysis which allowed me to account
>> for the effect of sampling error through the argument "mev". This was
>> relatively easy to do for a continuous response variable, however,
>> heritability is presented as a proportion and is therefore bounded by 0 and
>> 1 which clearly changes the situation.
>> In fact, I am not actually certain if it possible to conduct a formal
>> weighted meta-analysis on the heritability data using MCMCglmm. I have seen
>> elsewhere where data presented as a proportion (survival, yolk-conversion
>> efficiency for example) has been logit transformed and fitted using a
>> Gaussian error distribution (though this was done using REML rather than
>> Bayesian modelling) but I don't know if this is a legitimate strategy for a
>> formal meta-analysis using heritability as a response variable since any
>> transformation applied to the heritability data would also need to be
>> applied to the standard errors?
>> I would greatly appreciate any advice on this matter!
>> Cheers,
>> Jackie
>> --
>> Jacquelyn L.A. Wood, PhD.
>> Biology Department
>> Concordia University
>> 7141 Sherbrooke St. West
>> Montreal, QC
>> H4B 1R6
>> Phone: (514) 293-7255
>>         [[alternative HTML version deleted]]
>> _______________________________________________
>> R-sig-mixed-models at r-project.org mailing list
>> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models
> --
> *Ken Beath*
> Lecturer
> Statistics Department
> Phone: +61 (0)2 9850 8516
> Building E4A, room 526
> http://stat.mq.edu.au/our_staff/staff_-_alphabetical/staff/beath,_ken/
> CRICOS Provider No 00002J
> This message is intended for the addressee named and m...{{dropped:15}}

More information about the R-sig-mixed-models mailing list