[R-sig-ME] Modelling missing data in MCMCglmm()
Jarrod Hadfield
j.hadfield at ed.ac.uk
Tue Oct 13 09:11:12 CEST 2009
Hi Wayne,
If you have missing data in your response(s) then MCMCglmm samples
them under the assumption of an MAR (missing at random) process. The
terminology is confusing and MAR should not be confused with MCAR
(missing completely at random) which is what most people think of when
making the statement "missing at random". Under MAR the missing
species data are updated conditional on the model so that if the
phylogenetic signal is high and the species with the missing data have
close relatives, then the missing species data are weighted
(appropriately) to the data of their sibling taxa.
If there are missing data in the fixed effects MCMCglmm will
terminate. Sometimes if there are a lot of missing data for a fixed
predictor it can be handy to move it into the response:
cbind(y_1, y_2)
if missing data occur in y_1, then y_2 can help predict the missing
data if a relationship exists between the two responses.
This scheme is better than using point estimates for the missing data
because the uncertainty in their values is integrated out.
Jarrod
Quoting Dawson Wayne <wayne.dawson at ips.unibe.ch>:
> Hi R users,
>
> This is more of a forward planning question on my part, as I am still
> constructing this particular dataset, but I want to use MCMCglmm to
> test for factors contributing to naturalisation success of alien
> plants, whilst including a phylogeny. I will have missing data in this
> dataset, especially seed size for a number of species. I have read
> through the MCMCglmm vignette and package pdf and cannot see an example
> of how to include code for modelling missing data. To model missing
> seed sizes per species, I would like to use the mean and variance of
> seed size for congeners.
>
> I'm guessing they would be included as priors somehow, but I'm unsure
> how this coding would work. Any suggestions/further reading advice
> appreciated.
>
> Cheers,
>
> Wayne
>
> --
> Dr. Wayne Dawson
> Institute of Plant Sciences
> University of Bern
> Altenbergrain 21
> 3013 Bern
> Switzerland
> +41 (0)31 631 49 25
>
> _______________________________________________
> 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.
More information about the R-sig-mixed-models
mailing list