Re: st: xtregar in STATA vs. proc mixed in SAS

[David Jacobs <jacobs.184@sociology.osu.edu>]

First, I don't know anything about SAS.  Is your -Proc mixed-analysis 
a SAS version of fixed-effects (in the econometric meaning of that 
term as exemplified by adding case dummies to an OLS model; not the 
multilevel or Hierarchial Linear Models alternative meaning).  In 
fact, I thought -proc-mixed- in SAS estimated multilevel models, so 
the term fixed-effects would have a different meaning in -proc-mixed- 
than it would in -xtregar-.

Second even if your SAS routine estimates econometic fixed-effects, 
much depends on calculation defaults.  For example, Stata offers 
about 6 or 7 different ways to compute the AR(1) correction; Stata 
may not use the SAS default method.

My experience is that the fixed-effects estimators in Stata, EViews, 
and Limdep give virtually identical results, but this does not hold 
for random-effects because programs typically vary in the default 
methods they use to estimate random-effects.  And this probably is 
the case for AR(1) corrections.  Any of these differences, of course, 
would explain your divergent results.

Dave Jacobs

At 03:52 PM 3/18/2010, you wrote:
> I estimated fixed effect models with AR(1) error strcture using both
> STATA and SAS but received significantly different results. Below are
> the programs that I used in each:
>
> STATA:
> tsset id time
> xtregar y x1 x2, fe
>
> SAS:
> Proc mixed method=reml data=xxx;
> class id time;
> model y = id x1 x2 /s;
> repeated/type=ar(1) subject=id ;
>
> Shouldn't I expect similar (if not identical) results from both
> software? Am I missing something? Thanks.
>
> Yuluen
>
> *
> *   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/

*
*   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/