Troy Payne wrote:
>Does anyone know of a good HLM6 to Stata 10 comparison for estimating
>multilevel models? I assume both have strengths and weaknesses - a
>side-by-side comparison would be very helpful to me.
There will be small differences due to the algorithms' implementations
as you already noted.
Stata 10 may not have the full list of covariance matrices that HLM has
and it won't run Penalized Quasi-Likelihood Generalized Linear models
(no real loss). However, HLM does have the nifty "higher order Laplace"
for some GLMs which hasn't been implemented elsewhere.
HLM has sandwich estimators for multilevel models which Stata 10 doesn't
seem to do.
I'm not sure what of the various DF calculation methods Stata uses to
compute DF in multilevel models. SAS MIXED documentation has five I
believe (Satterthwaite, containment, Kenward/Roger, between/within,
residual). I believe HLM uses between/within, i.e., simply counts how
many vars and observations appear at different levels. Stata does
something else but I'm not sure what exactly, Satterthwaite I suppose
but I've never cared enough to find out.
Maybe Bobby Gutierrez can chime in and provide a definitive answer. :)
JV
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
