--- On Thu, 26/3/09, Stephen Armah wrote:
> should I first test for auto-correlation and
> heteroskedasticity in Stata before testing for
> endogeneity or is is better to do the reverse?
Any such sequence you shoose will strictly speaking
be wrong, unless you can frame the sequence of
tests in way that is similar to (Marcus, Peritz,
Gabriel 1976). I wouldn't worry too much about that
though.
The biger problem is that testing of model assumptions
is a pretty horrible idea anyhow. The very purpose
of a model is to simplify reality, ergo the
assumptions are supposed to be wrong, otherwise
the model would be a lousy simplification.
However, we don't want the assumptions to be
too wrong, otherwise the results would not say much
either. Statistical testing is not designed for this
kind of tradeoff: The logic behind testing is that
a hypothisis is either true or false, while when
we do model selection we already know that the
assumption is false but we want to see whether an
assumption is either useful or not useful. For
this reason graphical investigations of the various
model assumptions are by far preferable
I know that this is a rant and that opions differ
on this. If a reviewer/editor/supervisor/peer asks
you for such a test, than you should just give it
to them. Just don't take those tests too serious,
and don't forget to look at the graphs.
-- Maarten
Marcus, R, E. Perity, and K.R. Gabriel. 1976. On
closed testing procedures with special reference
to ordered analysis of variance. Biometrika
63:655--660.
-----------------------------------------
Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen
Germany
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
*
* 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/
