st: Re: Wondering why default estimation in xtmixed was changed from REML

[ymarchenko@stata.com (Yulia Marchenko, StataCorp LP)]

Bill Simpson <wsimpson@hbs.edu> asks what prompted the change in the default
estimation method of -xtmixed- from REML to ML in Stata 12:

> Does anyone know why xtmixed changed from REML to MLE for the default
> estimation method in version 12?  My understanding is that REML estimates of
> variance components are less biased than MLE.

As of Stata 12, the -xtmixed- command supports sampling weights and
cluster/robust standard errors.  These features are only available with ML
estimation.  As such, the default estimation method of -xtmixed- was changed
to ML.  

For small samples, Bill is right that the REML estimates of variance
components are less biased than the ML estimates.  So, if you have a small
number of groups, you would want to switch to REML estimation by specifying
the -reml- option with -xtmixed-.  When the number of groups is large, the
REML and ML estimates will be close.


-- Yulia
ymarchenko@stata.com
*
*   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/