[Limdep Nlogit List] Panel Data model with endogenous binary and count variables?
Michael Kennedy
huggsmba at yahoo.com
Sun Nov 26 03:59:47 EST 2006
Hello World,
I got a linear panel data model that I can estimate by
random or fixed effects:
REGRESS ;Lhs=Y1
;Rhs=ONE, Y2, Y3, X1, X2,
;Panel
;Pds = 5$
However, Y2 and Y3 are most likely endogenous, and
looking at the Limdep 8.0 manual on page E8-45, looks
like the best way to handle this is by 2SLS for the
FEM. (Digression: Is there a problem using 2SLS for
REM? Also, the manual did not mention anything about
3SLS for FEM, so wondering if that is not possible).
(1) REGRESS ;Lhs=Y2
;Rhs= all Xs (reduced form)
;fixed
;keep =predictedY2
;Pds = 5$
(2) REGRESS ;Lhs=Y3
;Rhs= all Xs (reduced form)
;fixed
;keep =predictedY3
;Pds = 5$
(3) REGRESS ;Lhs=Y1
;Rhs=ONE, Y2, Y3, X1, X2,
;Inst = predictedY2, predictedY3
;Panel
;Pds = 5$
So that looks straightforward, however a complication
in my case is that Y2 is a binary variable and Y3 is a
count variable. So in the above, I would like to use
PROBIT in (1) and POISSON in (2) instead of REGRESS
and not change (3), but I doubt that makes sense. Any
suggestions on whether that makes sense, and if not,
the correct way to do it?
Thanks,
Michael
____________________________________________________________________________________
Want to start your own business?
Learn how on Yahoo! Small Business.
http://smallbusiness.yahoo.com/r-index
More information about the Limdep
mailing list