Hi,
I'm working with binomial data and I've got a problem of convergence with
this function (glmmPQL). The problem is that for one of the groups (the
control group) all the observations are zero, so that makes the probability
of the binomial for that group zero. I think that that's the cause of the
variables having large standard errors.
Is there some way of avoiding this problem?
And, how is it that a constant effect is so difficult to analyse in
statistics?, I mean that there is information in the sample that the methods
available seem unable to "extract". The sample says "there is no effect in
this group" but you can't take advantage of the information. Does this mean
that zero variability is zero information?.
Sorry if the question seems a bit silly but bear in mind that I'm a student
yet....
Thanks
|