Laura:
You seem to be using the same variables in the selection
equation as in the wage equation. In that case
identification is solely based on the non-linear effect
of the variables on the probability of being selected.
However if most respondents have a probability of being
selected somewhere between 20 and 80%, than this effect
is approximately linear, so identification breaks down
and weird things like this can happen. Something similar
came up earlier today, see:
http://www.stata.com/statalist/archive/2006-10/msg00285.html
HTH,
Maarten
-----------------------------------------
Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands
visiting adress:
Buitenveldertselaan 3 (Metropolitan), room Z434
+31 20 5986715
http://home.fsw.vu.nl/m.buis/
-----------------------------------------
--- Laura Seelkopf wrote:
> I am having a problem with a heckman selection model for panel data. First
> stata 9 estimated it without problems after 34 iterations
>
> heckman y x x x, select (y=x x x)
>
> but then I needed to
>
> tsset
>
> my data in order to create a lagged variable. After this, the same heckman
> command as above did not work out anymore
>
> (no convergence achieved).
>
> I am wondering now, why this is the case and if/how it influences my
> estimation of the heckman model, because after I typed
>
> sort year
>
> the model could again be estimated.
*
* 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/
