[R-sig-ME] MCMCglmm and Contrast Part 2

[Miguel Macias]

Hello,

I would like to run some custom contrast on MCMCglmm. I have read how to do
this in the link <https://www.nceas.ucsb.edu/system/files/mixedlab.pdf> and
I have read previous posting about this issue <
https://stat.ethz.ch/pipermail/r-sig-mixed-models/2011q1/012415.html> but
the suggestions given have not resolved the problem I am having. I will
refer to data from <
https://github.com/bbolker/asaglmm/blob/master/data/culcitalogreg.csv> and
follow Ben Bolker's example <
https://www.nceas.ucsb.edu/system/files/mixedlab.pdf> to explain my problem.

I am using MCMCglmm version 2.24 and R version 3.4.1

My problem is that when I try to run a model with contrast I don't get back
the contrast results in the fixed effects summary. Please see below.

1) Running the Model without contrast

# setting seed to get reproducible results
set.seed(1)

# Running model with 'singular.ok = FALSE'
culcmod4 <-
MCMCglmm(cbind(predation,nopred)~ttt,random=~block,family="multinomial2",
                     data=culcdat,verbose=FALSE, singular.ok = FALSE)

# getting the summary results
#                 post.mean  l-95% CI u-95% CI   eff.samp    pMCMC
#  (Intercept)     228.11      43.33      408.20       340.0        0.008
**
#   ttt2              -196.65   -340.14       -62.33       224.0
 <0.001 ***
#   ttt3              -164.87   -314.63       -22.33       246.5
 0.016 *
#   ttt4              -253.51   -397.38       -99.90       132.1
 <0.001 ***
#   ---
#   Signif. codes:  0 ‘***’ 0.001 ‘**’ 0.01 ‘*’ 0.05 ‘.’ 0.1 ‘ ’ 1

2) The contrast are constructed and assigned to the factor:

# Setting contrast to ttt which has levels 1, 2, 3, and 4
contrasts(culcdat$ttt) <- matrix(c(3,-1,-1,-1,
                                                     0, 1,-1, 0,
                                                     0, 1, 1,-2),
                                                  nrow=4,

dimnames=list(c("none","C","S","CS"), c("symb","C.vs.S","twosymb")))

3) Running the model with the contrast

# setting seed to get reproducible results
set.seed(1)

# Running model with 'singular.ok = TRUE'
culcmod4v2 <-
MCMCglmm(cbind(predation,nopred)~ttt,random=~block,family="multinomial2",
                     data=culcdat,verbose=FALSE, singular.ok = TRUE)

# getting the summary results: I get the same results as without the
contrasts

Why aren't the contrast working? I do not get any error messages indicating
that contrast were dropped from the model.
I can get the procedure above to work with glmer from the lme4 package but
my data has complete separation and I would like to use a bayesian model to
get around complete separation.

Thank you in advance,



-- 
Miguel Macias González

451 E. Health Sciences Dr.
4212A UC Davis Genome Center,
Davis, CA, 95616
Work: 530-752-8889
Cell: 530-760-7205

	[[alternative HTML version deleted]]