Dear all,
I have a few questions about proper IVs in 3SLS. I am estimating the
following model using reg3.
(1) X=a1*Y+a2*Z+a3*W+a4*Q1+a5*R1+e1
(2) Y=a1*X+a2*Y+a3*W+a4*Q2+a5*R2+e2
(3) Z=a1*X+a2*Y+a3*W+a4*Q3+a5*R3+e3
1. If W can be endogenous in all three equations, I can specify W as
endogenous using "endo" option after reg3. Then, Stata uses all exogenous
variables in the system (= Q1, Q2, Q3, R1, R2, R3) as IVs. My question is
that using all exogenous variables as IVs does not seem to follow the basic
principle of IV, that is, an IV should be correlated with endogenous
variable but not be correlated with error term. If I want to follow this
principle, IVs should be different in each of the three equations. Or, the
rules for IV may different in 3SLS because this question also applies to X,
Y, Z because they appear in two equations. For example, IVs for X should be
different in equation (2) and (3), but 3SLS uses all exogenous variables as
IVs for X.
2. If I think W can be endogenous in only equation 1 and 2, what can I do
with reg3?
Thank you in advance.
Mike.
*
* 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/
