Re: st: RE: ivreg2 questions

[Robert Davidson <rhd773@gmail.com>]

Yes, I see now what it says in Section 8.10.  I was referring to
section 8.1 where it states the correlation between the two variables
must be high (before footnoting section 8.10) and I was asking if this
was a simplification.  Further, section 8.1 goes on to state that you
can test the potential strength of an instrument by simply regressing
the endogenous regressor on the proposed instrument.  However, the
example described earlier in that section did not have any control
variables in it, and later I realized after reading through the
chapter that it is only appropriate to do that when there are no other
variables in the model.

Theoretically, I find it hard to motivate an instrument that has
extremely weak correlation with the endogenous regressor even though
its partial correlation is strong when other variables are included in
the model.  But, that is not an econometric issue I suppose but rather
a rhetorical one.

On Tue, Mar 20, 2012 at 10:14 AM, Austin Nichols
<austinnichols@gmail.com> wrote:
> Robert Davidson <rhd773@gmail.com>:
> Can you quote the exact text you are reading?  IMEUS section 8.10
> makes quite clear that partial correlations are at issue (cf footnote
> 25), as do Angrist and Pischke.  The excluded instruments are always
> judged in the context of a larger regression; instruments that look
> good with one set of controls can look terrible with another set,
> either on the basis of weak instruments or overidentification tests,
> or both.
>
> On Mon, Mar 19, 2012 at 8:38 PM, Robert Davidson <rhd773@gmail.com> wrote:
>> Mark,
>>
>> Sorry for what is purely an econometric question at this point
>> (removed from Stata) but there is still one thing that I am
>> misunderstanding.  In every text I can read, it basically says the
>> instrument must be correlated with the endogenous regressor (including
>> Mostly Harmless Econometrics and an Introduction to Modern
>> Econometrics Using Stata to name 2 - the latter stating the instrument
>> must be highly correlated).  These texts do not state that the
>> instrument must have a high correlation with the endogenous regressor
>> with the effect of a set of controlling variables removed (partial
>> correlation).  Is this just a simplification on the part of these
>> texts or again is there something I am missing?  And does this
>> basically mean that the validity of an instrument is conditional on
>> the other independent variables included in the primary model and not
>> just the dependent variable and the endogenous regressor?
>>
>> Thank you again,
>> Rob
>
> *
> *   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/

*
*   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/