[R] glmmPQL and Convergence

From: <rab45+_at_pitt.edu>
Date: Sat 20 Aug 2005 - 12:54:37 EST


I fit the following model using glmmPQL from MASS:

fit.glmmPQL <-
glmmPQL(ifelse(class=="Disease",1,0)~age+x1+x2,random=~1|subject,family=binomial) summary(fit.glmmPQL)

The response is paired (pairing denoted by subject), although some subjects only have one response. Also, there is a perfect positive correlation between the paired responses. x1 and x2 can and do differ within each pair. Here is the output:

> summary(fit.glmmPQL)

Linear mixed-effects model fit by maximum likelihood  Data: fernando

       AIC BIC logLik
  30.51277 49.25655 -9.256384

Random effects:
 Formula: ~1 | subject

        (Intercept) Residual
StdDev: 8.284993 4.113725e-09

Variance function:
 Structure: fixed weights
 Formula: ~invwt
Fixed effects: ifelse(class == "Disease", 1, 0) ~ age + x1 + x2

                  Value Std.Error  DF   t-value p-value
(Intercept)   -35.01862 2.4414559 123     -14.3       0
age             0.59026 0.0441817 123      13.4       0
x1              1.39317 0.0000014  41 1000507.2       0
x2              0.93695 0.0000010  41  915150.3       0
 Correlation:
              (Intr) age        x2
age           -0.952
x1             0.000  0.000
x2             0.000  0.000 -0.057

Standardized Within-Group Residuals:
          Min            Q1           Med            Q3           Max
-2.939213e+00 -2.509951e-07 -1.169248e-07 2.999710e-06 3.825035e+00

Number of Observations: 168
Number of Groups: 125

The t-values are huge and the se's are correspondingly tiny. The model does a great job of discriminating between disease and no disease. But I have a feeling there is something wrong here. Is there something wrong with the type of model I'm trying to fit? If it weren't for the pairing I would just have used glm. Any insights would be appreciated.

Rick B.



R-help@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 Received on Sat Aug 20 12:59:42 2005

This archive was generated by hypermail 2.1.8 : Fri 03 Mar 2006 - 03:39:55 EST