Susan Lingle <[email protected]> asked:
> It is appropriate to use mixed models for binary response variables, or
> even for linear response variables, when the groups usually consist of 1
> individuals and at most 2 individuals???
Jeph Herrin <[email protected]> suggested
to check if the intra class correlation is similar to zero.
Jay Verkuilen <[email protected]> suggested to use "robust" standard
errors with clustering by parent.
I suggest a third alternative, one with a more Bayesian flavor (a la
Bartels or Gelman):
estimate the multilevel model, which uses partial pooling, but also
the classical models with no or complete pooling.
In response to Jeph, the clustering effect is determined not only by
the size of the intraclass correlation coefficient but rather by the
design effect.
A brief reference is Linda Muthen's posting at
http://www.statmodel.com/discussion/messages/12/18.html?1190832976.
In response to Jay, the real issue is whether a multilevel model is justified.
For example, I think that only a multilevel model can separately estimate
the predictive effects of an individual predictor and its group-level
mean, which
sometimes are interpreted as "direct" and "contextual" effects of the predictor.
Anders Alexandersson
[email protected]
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
