[R-sig-ME] MCMCglmm parameter expansion prior for binary model?

Jose Valdebenito Chavez jov23 @end|ng |rom b@th@@c@uk
Sat Apr 20 14:53:21 CEST 2019

Hello all,

I am having troubles with the convergence of a binary model using MCMCglmm.

It�s a relatively simple model but it only has binary variables (0,1):

The response variable "camply_1" represents presence or absence of bacteria in several individuals.

�wmi_campy_1� is whether the couple of this individual was infected or not.

�quality�  is either the individual was came from place A or B, and �sex� is the sex of the individual.

mc1 <- MCMCglmm(campy_1 ~ wmi_campy_1 + quality + sex,

                random = ~id + population,


                family = "categorical",

                data = mated,


I struggled a bit to set the priors for the model, actually I tried several online but they did not work for me. However after fiddling with other examples (like this one: https://stat.ethz.ch/pipermail/r-sig-mixed-models/2017q4/026115.html) and reading Hadfield Course notes I ended up with this:

prior.ex<- list(G = list(

  G1 = list(V = 1, nu = 1000, alpha.mu = 0, alpha.V = 1),

  G2 = list(V = 1, nu = 1000, alpha.mu = 0, alpha.V = 1)),

                R = list(V=1, fix = TRUE))

When I run my previous models with this prior it gave me very good results. But then when I run this model (�mc1�), which was basically very similar, the diagnostic plots showed that the variable �wmi_campy_1� was not converging properly. Also in the outcome I could see the posterior going wrong.

                          post.mean  l-95% CI  u-95% CI eff.samp pMCMC

(Intercept)              -2.5012   -4.7919   -0.2275 1502.426    0.028 *

wmi_campy_1   -200.9957 -410.5838   -1.0065    3.155    0.001 **

qualityisland         -3.9733   -8.2205   -0.4725 1770.965    0.024 *

sexM                       0.4699   -1.4708    2.2580 1830.226    0.604

I think this can be solved modifying the parameter expansion (alpha.mu, alpha.V) of the prior but I am out of ideas at this point. Any recommendations?

Many thanks.


	[[alternative HTML version deleted]]

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