Skip to content

desiderata for data manipulation

3 messages · Brian Ripley, A.J. Rossini

Original
Brian Ripley
# 
[...]
Only if someone is interested enough to write them....  Actually,
I thought parametric GLMMs were an unsolved research problem in
general, even in, say, the logistic case.  I know there are
suggestions in special cases (and there is a glmm function in one
of Jim Lindsey's packages) but all the simple ones can be seriously
incorrect AFAIK.
A.J. Rossini
# 
PBR> Only if someone is interested enough to write them....

Exactly, that's why I said eventually. 

    PBR> Actually, I thought parametric GLMMs were an unsolved
    PBR> research problem in general, even in, say, the logistic case.
    PBR> I know there are suggestions in special cases (and there is a
    PBR> glmm function in one of Jim Lindsey's packages) but all the
    PBR> simple ones can be seriously incorrect AFAIK.

I'll have to follow up on Jim's packages, which have always seemed
daunting, and run some comparisons.

As far as research problems -- yes, how to fit the semiparametric
version of the GLMMs is open, but for a fully parametric version,
there isn't much of an issue.

You are right that interpretation is severely problematic and needs
to be handled with care (Thomas has a nice presentation on that) but
they can be useful tools during data analysis.

Okay, maybe I should just take off a day or two in the next week and
see if most of Stata can be replaced by R :-), since we've made good
progress on a number of issues I had previously thought were
unresolved...

best,
-tony
A.J. Rossini
# 
AJR> As far as research problems -- yes, how to fit the
    AJR> semiparametric version of the GLMMs is open, but for a fully
    AJR> parametric version, there isn't much of an issue.

And I should've said -- for those models I'm interested in (which of
course, are in the simpler class).

Red-faced with foot-in-mouth...

best,
-tony