The -xtgee- manual says that Stata (v.8.2) "drops panels with too few
observations".
I want to specifically ask about lag-dependent correlation structures, e.g.,
corr(stationary1)
corr(ar1)
I have panels that have occasional missing observations. In other words,
some panels may have observations at times {1,2,3,4} but one panel may have
only times {1,2,4} and another may have only time {2}. If I fit
-xtgee ..., i(id) t(time) corr(ar1)-
Stata drops the latter two panels with the comment:
note: observations not equally spaced
modal spacing is delta t = 1
2 groups omitted from estimation
First, the panel w/ times {1,2,4} seems to have some info about the
correlation (from the data of the first two timepoints). In any case, even
if both these panels are uninformative re the correlation parameter, I do
not understand why they have to be dropped entirely (they do have info
about the marginal means). Does that mean that we cannot have missing data
if we want to use such correlation structures? That seems unreasonable. The
correlation estimates from the panels that are informative could be applied
to all panels (since we are estimating a common correlation across panels)
and the panels with missing observations could be fully used. Is it perhaps
that the scoring equations do not go through with unbalanced data? Could
someone outline what I am missing?
Thanks in advance.
Constantine
The documents accompanying this transmission may contain confidential
health or business information. This information is intended for the use of
the individual or entity named above. If you have received this information
in error, please notify the sender immediately and arrange for the return
or destruction of these documents.
