[R] post-hoc comparisons following glmm

[Spencer Graves]

	  The following appears to be an answer to your question, though I'd be 
pleased to receive critiques from others.  Since your example is NOT 
self contained, I modified an example in the "glmmPQL" help file:

(fit <- glmmPQL(y ~ factor(week)-1+trt, random = ~ 1 | ID,
+                      family = binomial, data = bacteria))
iteration 1
iteration 2
iteration 3
iteration 4
iteration 5
iteration 6
Linear mixed-effects model fit by maximum likelihood
   Data: bacteria
   Log-likelihood: -551.1184
   Fixed: y ~ factor(week) - 1 + trt
  factor(week)0  factor(week)2  factor(week)4  factor(week)6 factor(week)11
      3.3459650      3.5262521      1.9102037      1.7645881      1.7660845
        trtdrug       trtdrug+
     -1.2527642     -0.7570441

Random effects:
  Formula: ~1 | ID
         (Intercept)  Residual
StdDev:    1.426534 0.7747477

Variance function:
  Structure: fixed weights
  Formula: ~invwt
Number of Observations: 220
Number of Groups: 50
 > anova(fit)
              numDF denDF   F-value p-value
factor(week)     5   166 10.821682  <.0001
trt              2    48  1.889473  0.1622
 > (denDF.week <- anova(fit)$denDF[1])
[1] 166
 > (denDF.week <- anova(fit)$denDF[1])
[1] 166
 > (par.week <- fixef(fit)[1:5])
  factor(week)0  factor(week)2  factor(week)4  factor(week)6 factor(week)11
       3.345965       3.526252       1.910204       1.764588       1.766085
 > (vc.week <- vcov(fit)[1:5, 1:5])
                factor(week)0 factor(week)2 factor(week)4 factor(week)6
factor(week)0      0.3351649     0.1799365     0.1705898     0.1694884
factor(week)2      0.1799365     0.3709887     0.1683038     0.1684096
factor(week)4      0.1705898     0.1683038     0.2655072     0.1655673
factor(week)6      0.1694884     0.1684096     0.1655673     0.2674647
factor(week)11     0.1668450     0.1665177     0.1616748     0.1638169
                factor(week)11
factor(week)0       0.1668450
factor(week)2       0.1665177
factor(week)4       0.1616748
factor(week)6       0.1638169
factor(week)11      0.2525962
 > CM <- array(0, dim=c(5*4/2, 5))
 > i1 <- 0
 > for(i in 1:4)for(j in (i+1):5){
+   i1 <- i1+1
+   CM[i1, c(i, j)] <- c(-1, 1)
+ }
 > CM
       [,1] [,2] [,3] [,4] [,5]
  [1,]   -1    1    0    0    0
  [2,]   -1    0    1    0    0
  [3,]   -1    0    0    1    0
  [4,]   -1    0    0    0    1
  [5,]    0   -1    1    0    0
  [6,]    0   -1    0    1    0
  [7,]    0   -1    0    0    1
  [8,]    0    0   -1    1    0
  [9,]    0    0   -1    0    1
[10,]    0    0    0   -1    1
 > library(multcomp)
 > csimint(par.week, df=denDF.week, covm=vc.week,cmatrix=CM)

	Simultaneous confidence intervals: user-defined contrasts

	95 % confidence intervals

       Estimate  2.5 % 97.5 %
  [1,]    0.180 -1.439  1.800
  [2,]   -1.436 -2.838 -0.034
  [3,]   -1.581 -2.995 -0.168
  [4,]   -1.580 -2.967 -0.193
  [5,]   -1.616 -3.123 -0.109
  [6,]   -1.762 -3.273 -0.250
  [7,]   -1.760 -3.244 -0.277
  [8,]   -0.146 -1.382  1.091
  [9,]   -0.144 -1.359  1.070
[10,]    0.001 -1.206  1.209

 > csimtest(par.week, df=denDF.week, covm=vc.week,cmatrix=CM)

	Simultaneous tests: user-defined contrasts

Contrast matrix:
       [,1] [,2] [,3] [,4] [,5]
  [1,]   -1    1    0    0    0
  [2,]   -1    0    1    0    0
  [3,]   -1    0    0    1    0
  [4,]   -1    0    0    0    1
  [5,]    0   -1    1    0    0
  [6,]    0   -1    0    1    0
  [7,]    0   -1    0    0    1
  [8,]    0    0   -1    1    0
  [9,]    0    0   -1    0    1
[10,]    0    0    0   -1    1

Adjusted P-Values

       p adj
  [1,] 0.011
  [2,] 0.013
  [3,] 0.014
  [4,] 0.015
  [5,] 0.020
  [6,] 0.024
  [7,] 0.985
  [8,] 0.985
  [9,] 0.985
[10,] 0.997
 > sessionInfo()
R version 2.2.1, 2005-12-20, i386-pc-mingw32

attached base packages:
[1] "methods"   "stats"     "graphics"  "grDevices" "utils"     "datasets"
[7] "base"

other attached packages:
   multcomp    mvtnorm       MASS    statmod       nlme
    "0.4-8"    "0.7-2"   "7.2-24"    "1.2.4" "3.1-68.1"

	  If this does NOT answer your question (or even if it does), PLEASE do 
read the posting guide! "www.R-project.org/posting-guide.html".  I'd 
prefer not to have to guess whether you would think the example I chose 
was relevant.

	  hope this helps,
	  spencer graves

Michaël Coeurdassier wrote:

> Dear R community,
> 
> I performed a generalized linear mixed model using glmmPQL (MASS 
> library) to analyse my data i.e : y is the response with a poisson 
> distribution, t and Trait are the independent variables which are 
> continuous and categorical (3 categories C, M and F) respectively, ind 
> is the random variable.
> 
> mydata<-glmmPQL(y~t+Trait,random=~1|ind,family=poisson,data=tab)
> Do you think it is OK?
> 
> Trait is significant (p < 0.0001) and I would like to perform post-hoc 
> comparisons  to check  where the difference among  Trait categories but 
> I did not find  a solution  in R help list or others.
> 
> Thank you in advance for your help
> 
> Michael
> 
> ______________________________________________
> R-help at stat.math.ethz.ch mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help
> PLEASE do read the posting guide! http://www.R-project.org/posting-guide.html