Mark -
Thanks for pointing out all those "maximize" options - I tried the
"difficult" option and voila! - I do get the same results in the two fits.
However the log likelihood was lower than for the noconst and no "difficult"
run. I did check likelihood values through the iterations with and w/o a
constant and with and w/o "difficult". Here's what happened on the last few
iterations:
A: with _cons
Iteration 6: log likelihood = -182.0929 (not concave)
Iteration 7: log likelihood = -178.55296
Iteration 8: log likelihood = -178.40032
Iteration 9: log likelihood = -178.39898
Iteration 10: log likelihood = -178.39897
B: with _cons and "difficult"
Iteration 35: log likelihood = -178.96001 (not concave)
Iteration 36: log likelihood = -178.95998
Iteration 37: log likelihood = -178.95993
Iteration 38: log likelihood = -178.9599
Iteration 39: log likelihood = -178.95986 (not concave)
Iteration 40: log likelihood = -178.95985
C:
nocons
Iteration 0: log likelihood = -201.65183 (not concave)
Iteration 1: log likelihood = -189.26711
Iteration 2: log likelihood = -182.49504
Iteration 3: log likelihood = -178.24615
Iteration 4: log likelihood = -177.73176
Iteration 5: log likelihood = -176.86157
Iteration 6: log likelihood = -176.84049
Iteration 7: log likelihood = -176.83953
Iteration 8: log likelihood = -176.83953
D: nocons and "difficult"
Iteration 0: log likelihood = -201.65183 (not concave)
Iteration 1: log likelihood = -188.39399 (not concave)
Iteration 2: log likelihood = -182.04631 (not concave)
Iteration 3: log likelihood = -179.40337
Iteration 4: log likelihood = -178.48942
Iteration 5: log likelihood = -178.39969
Iteration 6: log likelihood = -178.39897
Iteration 7: log likelihood = -178.39897
The results for A and D were the same, but C had the highest log likelihood
and converged quickly. B had a lower log likelihood and had the most trouble
converging.
So if we really are looking for a global optimum, C would be the winner -
but since rho-hat came out essentially zero for this case, it too is
suspect. I guess this revolves back to the warning on p.26 (Stata 7 H-P
manual). Thanks again for your help on this.
Al Feiveson
-----Original Message-----
From: Mark Schaffer [mailto:[email protected]]
Sent: Friday, January 24, 2003 10:57 AM
To: [email protected]
Subject: Re: st: -heckman- with model reparameterization
Al,
From: "FEIVESON, ALAN H. (AL) (JSC-SD) (NASA)"
<[email protected]>
To: "'statalist'" <[email protected]>
Subject: st: -heckman- with model reparameterization
Date sent: Fri, 24 Jan 2003 10:41:56 -0600
Send reply to: [email protected]
> Hi - I am attempting to run a heckman selection regression model with
three
> indicator variables as the only independent variables. In particular they
> are ipre, iin and ipost, denoting membership in one of three "phases".
>
> First, I run -heckman- with a constant and ipre omitted (because
> ipre+iin+ipost=1 for all observations):
> . heckman y iin ipost,select(iin ipost) nolog
<snip>
I've had problems on occasion with getting -heckman- to finish at the
global maximum. My guess is that you're having a similar problem.
What happens if you don't -nolog- the output?
It was a little while ago but I think the way I dealt with it was to
fiddle with the maximize options of -heckman-. The -difficult- option
might do the trick.
Hope this helps.
--Mark
Prof. Mark E. Schaffer
Director
Centre for Economic Reform and Transformation
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS UK
44-131-451-3494 direct
44-131-451-3008 fax
44-131-451-3485 CERT administrator
http://www.som.hw.ac.uk/cert
*
* 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/
*
* 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/
