I believe you are referring below to -espoisson- from package st0057 (from
http://www.stata-journal.com/software/sj4-1), which I don't think is of use
to you. If the RHS trade variables are defined only for multinationals, then
only multinationals will appear in the regression, so you have a
non-representative sample of all firms, and a dummy won't help you. Is that
what you mean?
If you want to run a Poisson regression with an endogenous continuous
variable, and you have an instrumental variable, use the control function
approach described in Wooldridge, EACSPD 19.5.1, p664 (see
From: liza jabbour [mailto:Liza.Jabbour@univ-paris1.fr]
Sent: Saturday, August 20, 2005 8:00 AM
Subject: st: endogeneity and count models.
I am estimating a count model and I have an endogeneity problem. However my
endogenous varaibles are not dummies but continuous variables. I want to
estimate the impact of trade variables on innovation. The trade variables I
have are limited to multinational firms while the innovation activity is
not. The endogeneity of those variables comes from the fact that more
innovative firms have a higher probability of becoming multinationals.
I know that there is a GMM model that treat endogeneity in count models, but
I don't think that it is available with STATA. However STATA estiamtes a
"FIML estimation of an endogenous switching model for count data" where the
endogenous variable needs to be a dummy.
If I include in the "FIML estimation of an endogenous switching model for
count data" estimation a dummy for multinational firms and treat it as
endogenous, will this be sufficient to control for the endogeneity of the
* For searches and help try: