A careful reading of the help for -version-
may be helpful.
In general, do not assume that setting
the -version- back to some previous version
is equivalent to going back to that version
in all respects. Rather it gives a handle
to re-create _selected_ previous behaviour.
Nick
[email protected]
Dev Vencappa
> David,
> thanks a lot for your help indeed. As you rightly mentioned,
> I think my dataset does not satisfy the new default
> nrtolerance. Specifiying nonrtolerance in the command did
> indeed induce convergence after 24 iterations (this at the
> same number of iterations as what my friend obtained with
> stata 8.0). However, what worries me is that at convergence,
> we both obtained the message "(not concave)".
>
> A second issue that puzzles me is that you mentioned that
> following the 5 November 2003 update, the -nrtolerance(1e-5)-
> became the default
> convergence criterion. If that is the case, then how is
> possible that when I specify version 8.0 within stata and run
> the command, I am still faced with this problem of
> non-convergence? Is it the case that once I have updated from
> Stata 8.0 to 8.2, my software automatically overrides the
> previous nrtolerance to the new one, and this becomes the
> default no matter what version I specify when running the
> frontier command with maximise options? Or am I wrong in
> thinking that nrtolerance is comparable with any other
> command which are outdated in the new version and can be run
> only in an older version and, therefore, should not have
> produced this problem of non-convergence in the first
> instance, when I tried to run the command under version 8.0?
>
> Finally, I will attempt you suggestion of multiplicative
> rescaling of the variables. I did try changing the starting
> values before I posted this query, but that did not work either.
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
