st: Heckman with covariates that perfectly predict selection

["K Ssdjk" <dfjkldf3@hotmail.com>]

I am running a heckman selection model (stata7, two-step estimation)
and have some RHS vars which perfectly predict censoring. While I get
estimates, the parameter on the perfect predictor seems to be the
opposite sign from what I would expect. For example I have an
indicator (Dmil) which is a perfect predictor: when Dmil=1 the
dependent variable is always observed. However the estimated
selection effect is negative,

...
select
  Dmil   -48.28     .
...

I understand the large value but why not positive? The rest of the
estimates are plausible and rho=-0.15. Any help/suggestions would be
appreciated!

KS

PS There might be a write-up in STB-43 about this but I could not
find it.

_________________________________________________________________
MSN 8 helps eliminate e-mail viruses. Get 2 months FREE*. 
http://join.msn.com/?page=features/virus

*
*   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/