Re: st: first stage results using ivregress with vce(cluster)

[Suryadipta Roy <sroy2138@gmail.com>]

Dear Mark and Nick,
Thank you very much for the excellent suggestions!

Suryadipta.

On Thu, Mar 13, 2014 at 7:40 AM, Schaffer, Mark E <M.E.Schaffer@hw.ac.uk> wrote:
> Souvik, Suryadipta,
>
>> -----Original Message-----
>> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-
>> statalist@hsphsun2.harvard.edu] On Behalf Of Suryadipta Roy
>> Sent: 13 March 2014 10:57
>> To: statalist@hsphsun2.harvard.edu
>> Subject: Re: st: first stage results using ivregress with vce(cluster)
>>
>> Souvik,
>> The other option (probably the preferred one to me) is to use - ivreg2
>> - which comes with a wide array of tests on instrumental validity,
>> i.e. testing for weak instruments as well as overidentification. Try -
>> help ivreg2 - and the command that is most pertinent to your query is
>> -condivreg- .
>
> Actually, if Souvik is going to go down the weak-identification-robust route, -weakiv- rather than -condivreg- is more useful for Souvik's problem.  -weakiv- supports non-iid VCE options such as cluster-robust, unlike -condivreg-.
>
> --Mark
>
>>
>> Best wishes,
>> Suryadipta.
>>
>> On Wed, Mar 12, 2014 at 8:49 AM, Souvik Banerjee
>> <bobobanerjee@gmail.com> wrote:
>> > Yes, by relevance I meant testing for weak instruments. I will try it
>> > out. Thanks.
>> >
>> > Best,
>> > Souvik
>> >
>> > On Wed, Mar 12, 2014 at 8:40 AM, Alfonso Sanchez-Penalver
>> > <alfonso.statalist@gmail.com> wrote:
>> >> If by relevance of the instruments you mean that they are not weak, which is
>> what I think you mean, that is what is tested with -estat firststage- which takes
>> into consideration the clustering in the vce. To try it out, do the 2SLS estimation
>> with and without vce(cluster clusterid), and try the test right after each
>> estimation. You ought to get different results.
>> >>
>> >> Best,
>> >>
>> >> Alfonso Sanchez-Penalver
>> >>
>> >>> On Mar 12, 2014, at 8:20 AM, Souvik Banerjee
>> <bobobanerjee@gmail.com> wrote:
>> >>>
>> >>> Hi Alfonso,
>> >>> I am following up on your response yesterday. Thanks for your help. I
>> >>> am aware of the formal tests for endogeneity and instrument validity.
>> >>> I was actually referring to the test for relevance of the
>> >>> instrument(s), which to my knowledge is just based on the first stage
>> >>> results.
>> >>>
>> >>> Thanks again.
>> >>>
>> >>> Best,
>> >>> Souvik
>> >>>
>> >>> On Tue, Mar 11, 2014 at 11:03 AM, Alfonso Sánchez-Peñalver
>> >>> <alfonso.statalist@gmail.com> wrote:
>> >>>> Hi again Souvik,
>> >>>>
>> >>>> to test for endogeneity and the validity of the instruments more formal
>> tests are preferred than just inferencing on the results from the first stage. See
>> -help ivregress_postestimation-. Both -estat endogenous- and -estat overid-
>> provide Wooldridge's (1995) robust score tests. Also -estat firststage- accounts
>> for the robustness of the VCE if I'm not wrong.
>> >>>>
>> >>>> Wooldridge, J. M. 1995.  Score diagnostics for linear models estimated by
>> two stage least squares.  In Advances in Econometrics and Quantitative
>> Economics: Essays in Honor of Professor C. R. Rao, ed. G. S. Maddala, P. C. B.
>> Phillips, and T. N. Srinivasan, 66-87.  Oxford: Blackwell.
>> >>>>
>> >>>> Best,
>> >>>>
>> >>>> Alfonso Sánchez-Peñalver, PhD
>> >>>>
>> >>>>
>> >>>>> On Mar 11, 2014, at 10:13 AM, Souvik Banerjee
>> <bobobanerjee@gmail.com> wrote:
>> >>>>>
>> >>>>> Hi Alfonso,
>> >>>>> Thanks for the response. I agree that  -ivregress 2SLS- is probably
>> >>>>> applying the clustering in the second stage only. I, however, feel the
>> >>>>> standard errors should be clustered in the first stage as well because
>> >>>>> one would need the "correct" standard errors from the first stage
>> >>>>> regression to test for the relevalence of the instrumental variable
>> >>>>> (z) (which I need to test).
>> >>>>>
>> >>>>> I apologize for the confusion with the clustering syntax - it is a
>> >>>>> typo. The correct code should be:
>> >>>>>
>> >>>>> ivregress 2sls y x (x1 = z), first vce(cluster x2)
>> >>>>> and then manually run:
>> >>>>> regress x1 x z, cluster vce(cluster x2)
>> >>>>>
>> >>>>> Thanks again.
>> >>>>>
>> >>>>> Best,
>> >>>>> Souvik
>> >>>>>
>> >>>>> On Tue, Mar 11, 2014 at 9:50 AM, Alfonso Sanchez-Penalver
>> >>>>> <alfonso.statalist@gmail.com> wrote:
>> >>>>>> Hi Souvik,
>> >>>>>>
>> >>>>>> What you're saying is normal. The first stage regression(s) is only useful
>> to predict the values of the endogenous variable(s) so that they can be used in
>> the second stage, so for 2SLS estimation purposes the standard errors of the
>> first stage are not that important. Therefore my understanding is that -
>> ivregress 2SLS- only applies the clustering for the standard errors in the second
>> stage, not the first.
>> >>>>>>
>> >>>>>> Having said that, I'm confused with your syntax. You say that you use -
>> vce(cluster clustervariable)- but I don't see this option in your clarifying
>> example but rather -cluster(x2)-.
>> >>>>>>
>> >>>>>> Best,
>> >>>>>>
>> >>>>>> Alfonso Sanchez-Penalver
>> >>>>>>
>> >>>>>>> On Mar 11, 2014, at 9:09 AM, Souvik Banerjee
>> <bobobanerjee@gmail.com> wrote:
>> >>>>>>>
>> >>>>>>> Dear Stata users,
>> >>>>>>> I have a question about verifying the first stage regression results
>> >>>>>>> obtained using the "first" option with "ivregress 2sls" with those
>> >>>>>>> obtained by manually running the first stage regression (using
>> >>>>>>> "regress") of the endogenous variable on the IV and other covariates
>> >>>>>>> when I use the "vce(cluster clustervariable)" option to allow for
>> >>>>>>> clustering. The coefficients are the same in both cases, however the
>> >>>>>>> standard errors are different - which should not be the case. Does
>> >>>>>>> anyone have any idea why this might happen? However, if I do not
>> allow
>> >>>>>>> for clustering, the standard errors are the same in both cases - which
>> >>>>>>> is correct.
>> >>>>>>>
>> >>>>>>> To clarify, I run:
>> >>>>>>> ivregress 2sls y x (x1 = z), first (cluster x2)
>> >>>>>>> and then manually run:
>> >>>>>>> regress x1 x z, cluster (x2)
>> >>>>>>>
>> >>>>>>> Here, y is binary variable, x is vector of covariates (x1, x2, ...
>> >>>>>>> xk), x1 is the endogenous variable (binary) and z the instrumental
>> >>>>>>> variable.
>> >>>>>>>
>> >>>>>>> Thanks,
>> >>>>>>>
>> >>>>>>> Souvik
>> >>>>>>> *
>> >>>>>>> *   For searches and help try:
>> >>>>>>> *   http://www.stata.com/help.cgi?search
>> >>>>>>> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/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/faqs/resources/statalist-faq/
>> *   http://www.ats.ucla.edu/stat/stata/
>
>
> -----
> Sunday Times Scottish University of the Year 2011-2013
> Top in the UK for student experience
> Fourth university in the UK and top in Scotland (National Student Survey 2012)
>
>
> We invite research leaders and ambitious early career researchers to
> join us in leading and driving research in key inter-disciplinary themes.
> Please see www.hw.ac.uk/researchleaders for further information and how
> to apply.
>
> Heriot-Watt University is a Scottish charity
> registered under charity number SC000278.
>
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/faqs/resources/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/faqs/resources/statalist-faq/
*   http://www.ats.ucla.edu/stat/stata/