On Sat, Dec 8, 2012 at 4:13 AM, John Antonakis <John.Antonakis@unil.ch> wrote:
> Right. A model cannot have negative degrees (-1) of freedom; it is obvious
> that it is not defined. Though in these cases it would be good if Stata
> refused to estimate it or gave an explicit message that the model is
> undefined.
Yes, definitely. -factor-, for instance, gives a solution at a Heywood
case but it quite appropriately whines. It would be good if -sem- did
the same thing. I definitely think it should give the solution because
that's often helpful for figuring out what the identification problem
is, but there should be a hefty cautionary note. While I have the
strong habit of checking the size of the standard errors of models I
fit and often output the parameter VCE to check for ridiculous values,
many folks do not.
*
* For searches and help try:
* http://www.stata.com/help.cgi?search
* http://www.stata.com/support/faqs/resources/statalist-faq/
* http://www.ats.ucla.edu/stat/stata/
