st: Problem-Estimation results-Heckman's twostep vs. maximum likelihood

[Zhou Yu <zyu@usc.edu>]

Hi there,

I am a novice Stata user.  I have problem with estimation results.

The twp-step and maximum likelihood estimations gave me two very 
different results.  The results of the twostep is more reasonable to me, 
as the coefficients are expected.  However, I was told that the ML is 
more efficient.  The last iteration of the ML was concave.

After taking off some independent variables, the two procedures would 
produce similar results.

What is the problem?  How to make the two estimations consistent?

Many thanks,

Zhou


Two-step:
-------------+----------------------------------------------------------------
mills        |
      lambda |   -.051415   .0295235    -1.74   0.082      -.10928 
 .00645
-------------+----------------------------------------------------------------
         rho |   -0.08543
       sigma |  .60181893
      lambda | -.05141501   .0295235
------------------------------------------------------------------------------

ML:

-------------+----------------------------------------------------------------
     /athrho |  -1.391146   .0148953   -93.39   0.000    -1.420341 
-1.361952
    /lnsigma |   -.264034   .0041028   -64.36   0.000    -.2720753 
-.2559927
-------------+----------------------------------------------------------------
         rho |  -.8834229   .0032705                       -.88967 
-.8768451
       sigma |   .7679474   .0031507                      .7617969 
.7741476
      lambda |  -.6784223   .0049747                     -.6881726 
-.668672
------------------------------------------------------------------------------
LR test of indep. eqns. (rho = 0):   chi2(1) =  2729.86   Prob > chi2 = 
0.0000
------------------------------------------------------------------------------


*
*   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/