Hello statalisters,
I know something related to this topic came up some time ago, but I'd
love to propose it again, because it could be interesting to discuss
about this in the following case.
I have a latent ordered response model such as:
y* = a + bX + c(k) + e
Where b and X are, respectively, a vector of parameters and a vector
of variables at individual and local-level and microregion-level,
while c(k) represents macro-regional dummy variables. In practice
these are K regional (fixed) effects. This is the model I come up
with, because one of the most important thing I'd love to know is if
there exists a significant difference across regions in the reported
y.
The number of K regional effects can be 7 (macro-regions) or 33 (micro-regions).
Now, if I remember well this problem is recognized as incidental
parameters problem is Panel data. Even if I am dealing with a
cross-section, I might have the same inconsistent estimators for my
dummies.
Some information on the data set might help here. I have about 1,500
observations. If I consider the 7 macro-regions, the # of obs. varies
between a minimum of 107 to a maximum of 395. However, if I consider
the 33 micro-regions, the # of obs. varies between 2 and 171.
I'd love to get some advice (such as if it is maybe better to change
and estimate a random effect model or a multileveland if so, why?) and
some references about this - incidental parameters problem with cross
section.
Thank you very much to all of you in advance!
Mirko
*
* 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/