[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
Re: st: Exact Identification and Over Identification
my personal idea is to use both instruments (maybe after checking their redundancy, see the help).
However, if you want to choose between them, and if I understood your model correctly, you don't have any problem for testing the validity. Suppose your endogenous are endo1 and endo2, the instruments for endo1 are instr1 and instr2, the instruments for endo2 are instr3 and instr4. If both endo1 and endo2 enter in the SAME model, then you cannot ask Stata to use instr1 and instr2 as instruments for endo1 ONLY and instr3 and instr4 as instruments for endo2 ONLY. Thus, you have 2 endogenous and 4 instruments, and you can test each of them separately with the orthogonality option.
Remember to refer to Stata commands within ordinary text using dashes (- -) to flag words.
At 02.33 08/04/2007 -0400, sabrina wang wrote:
>I would appreciate your comments on the following
>I have a model with two endogenous variables. I have
>two candidate IVs for each endogenous variable. I am
>using IVREG2 to estimate my model. My questions are:
>(1). Should I use both candidate IVs in my estimateion
>or should I check the validity of the two candidate
>IVs and use one IV for my model?
>(2). If I go for the latter case in (1), identifying
>one IV. Then how could i conduct Durbin-wu-hausman
>test in IVREG2? Since i have exact identification and
>the Orthog() option of IVREG2 will not work in this
>case. Could I use:
>IVREG dependentvar indepedentvars (endogenous
>vars=ivs), robust cluster(id)
>reg dependentvar indepedentvars, robust cluster(id)
>(3). When I am checking the validity of my IV. Could I
>just regress the residual from the structure model on
>the candidate IVs separately. If the coefficient on an
>IV is insignificant, then the IV is exogenous and
>valid assuming the IV already passed the relevance
>(4). a more general question is: which one is better?
>the exact identification or using more-than-one IVs.
>Thanks a lot,
* For searches and help try: