>I am doing something very similar to Yvonne and am curious about the
specific advantages of xtmixed.
Hmmm, well I can't really do an entire course on HLM on a listserv (not
and keep my job :), but here's a quick bit. Let's say you have a survey
of employees in companies. You want to make generalizations about the
companies, which you consider to be a random sample from a
super-population of companies. The individual level survey information
needs to be modeled, but you want to take the fact that the data came
from separate companies into account, if only because employees in the
same company are exposed to the same environment and are thus likely to
be correlated compared to those in different companies. The real benefit
of the HLM over separate regressions per company is the fact that the
HLM "borrows strength" from the other companies and does so in an
appropriate manner. If you used company averages, that commits the
aggregation fallacy and typically leads to models that have GoF
statistics that are way too high. If you ignore the companies you're not
answering the question you wanted to in the first place and also are
treating your observations as if they were independent, which they are
not.
Take a look here first, or do more Googling.
http://www.cs.berkeley.edu/~russell/classes/cs294/f05/papers/gelman-2005
.pdf
This really only scratches the surface of what is a gigantic literature.
If you're interested in using Stata, check out the book by Rabe-Hesketh
and Skrondal on Stata Press and go from there.
There are other approaches and the HLM approach isn't the only one to
consider. For instance, there's a large literature on panel data in
econometrics; I'm less familiar with that.
JV
*
* 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/
