[R-sig-ME] Sampling methods for MCMCglmm using cengaussian family

Joshua Wiley jwiley.psych at gmail.com
Sun Sep 30 18:08:05 CEST 2012

Hi Robin,

Jarrod may be along at some point with a more definitive answer but:

If you have not already, you should read:
They are excellent and I think you may find Chapter 7 informative.

Of course you have a posterior distribution for latent scores.  You
always will, although if you had no random effects other than
residuals and all observed noncensored gaussian, it would be a
prodigiously boring posterior with no variability across draws, but
there all the same.  In your case, for censored cases it will vary by

To answer your questions.

1) I am not sure.
2) Yes, MH is being used.  Again the technical details on the course
notes provide some more information on sampling schemes.
3) As far as I can tell you cannot.  You can set it via the tune
argument.  You set NULL so the default start is used

 if (is.null(tune)) {
        AMtune = c(rep(FALSE, nG), rep(TRUE, nR))
        for (i in 1:nR) {
            tune[[i]] = diag(nfl[nG + i])

this is passed to .C, but as near as I can tell, is not returned from
the R MCMCglmm() function.  So it would seem you do not extract it.  A
quick poke around MCMCglmm.cc, suggests it is the 34th
argument---probing deeper there may get you some traction if you
really want.

Side comment, you are using small variances on the prior for your
fixed effects.  Just wanted to point it out if not intentional.



On Sat, Sep 29, 2012 at 3:17 PM, Robin Jeffries <rjeffries at ucla.edu> wrote:
> Hello,
> I am using MCMCglmm to model a time to event outcome where individuals can
> be either left or right censored so I am using the "cengaussian" family. I
> have successfully run the model
> test <- MCMCglmm(cbind(low, up) ~ g0 + age0,
>          family="cengaussian", data=sex, tune=NULL,
>          prior = list(R = list(V=1, n=0.002),
>                       B = list(mu=rep(0,3), V=diag(3))))
> Looking at the form of the density function from the MCMC Overview
> documentation, I don't see liabilities. Combine that with having no random
> effects, leads me to think that the program would use Gibbs sampling.
> However, when I add a "pl=TRUE" to the above code, I get a posterior
> distribution of liabilities, and the program displays the acceptance ratio
> for latent scores (this is displayed regardless of the value of pl). Both
> seem to indicate that MH sampling is being used.
> So my questions are
> 1) Where are the liabilities being used in the density?
> 2) Does the display of the acceptance ratio for latent scores definitively
> mean that MH sampling is being used?
> 3) If MH is being used, how can I extract the covariance matrix for the
> proposal distribution that was used?
> Thank you,
> -Robin
> Robin Jeffries
> MS, DrPH Candidate
> Department of Biostatistics,
>         [[alternative HTML version deleted]]
> _______________________________________________
> R-sig-mixed-models at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-sig-mixed-models

Joshua Wiley
Ph.D. Student, Health Psychology
Programmer Analyst II, Statistical Consulting Group
University of California, Los Angeles

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