st: heckman model "would simplfy to OLS regression"

[Jochen Späth <jochen.spaeth@iaw.edu>]

Dear Statalisters,

trying to run a heckman twostep estimation with bootstrapped standard errors I get the error message "Dependent variable never censored because of selection: model would simplify to OLS regression; an error occurred when bootstrap executed heckman". 

What is my mistake?

-To clarify, I have panel data on the firm level that looks like this:

firmid	year	size	gr_size	selected

1		2000	2	.		1
1		2001	0	.		1
1		2002	29	-0,2762532	1
1		2003	22	0,435318	1
1		2004	34	-1,041454	1
1		2005	12	0,5108256	1
1		2006	20	.		0

2		2000	25	.		1
2		2001	0	.		1
2		2002	20	-0,6931472	1
2		2003	10	0,6931472	1
2		2004	20	0,4700036	1
2		2005	32	-0,9007866	1
2		2006	13	.		0

with gr_size beeing the 1-year continuos growth rate of size (i.e. the difference of size logarithms). gr_size ist the dependent variable of the outcome equation. Since it is only observed if the firm continues to live I want to separate this selection bias via the heckman procedure. To separate firms that are just temporarily no visible in my data (size == 0 for those years) from end-of-firm-life observations I generated the dummy variable selected which is 0 if a firm does not continue to live in the next year and 1 otherwise.
The heckman command is thus of the form -heckman gr_size regressors, twostep select(selected = regressors from outcome equation + 1 continuos instrument) vce(boot, reps(500) strata(firmid))-.

Could it be problematic that the list of my regressors contains many dummy variables and constants so that the participation equation cannot be identified due to little variation in the data? Or what else is my mistake (e.g. wrong coding of the selected variable or alike)?

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/