Stata The Stata listserver
[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

Re: st: problem with ivreg2 with cluster option


From   Mark Schaffer <M.E.Schaffer@hw.ac.uk>
To   statalist@hsphsun2.harvard.edu, SHEHAJ Entala <E.Shehaj@staffs.ac.uk>
Subject   Re: st: problem with ivreg2 with cluster option
Date   Wed, 23 Mar 2005 22:43:05 +0000 (GMT)

Entela,

Quoting SHEHAJ Entala <E.Shehaj@staffs.ac.uk>:

> Hi,
> 
> I am using ivreg2 with kernel based estimation with gmm for a panel of
> 50 firms over 6 year period. Also I am trying ivreg2 with cluster option
> and gmm for the same panel, but the problem is that the F test for the
> overall significance of coefficients in the second stage show that they
> are not different from zero. While all other statistics concerning the
> validity of instruments are OK. The same regression with kernel based
> estimation instead of cluster is fine (in terms of all statistics
> including the F test of overall significance in the second stage). I
> will appreciate any suggestion about this problem.

This isn't really a "problem with ivreg2 with cluster option".  Rather, it's
an issue of how do you interpret these results.

The difference between kernel-based robust covariance estimation and
cluster-robust estimation is the following.  The kernel-based approach makes
more assumptions about the data for the asymptotics to work, and in
particular requires an assumption about how fast the autocorrelations die
out.  The asymptotics also need the time dimension to go off to infinity. 
The cluster-robust approach allows for a more general - indeed, arbitrary -
form of intra-group correlation.  The asymptotics for cluster-robust need
the cross-sectional dimension to go off to infinity.

That the results are different is not very surprising.  Which of the two is
suited to your data depends on some things that you haven't mentioned in
your email, e.g., how long the time-series dimension is (for the kernel
apporach), and what the degrees of freedom in the cross-section dimension
are, i.e., 50 observations minus the number of instruments (for the cluster
approach).  Which of the two (one, the other, both, or maybe even neither)
is appropriate is, I suppose, to some extent a judgement call.

Hope this helps.

--Mark

> Regards
> Entela Shehaj
> 
> The information in this email is confidential and is intended solely
> for
> the addressee. Access to this email by anyone else is
> unauthorised.
> 
> If you are not the intended recipient, any disclosure, copying,
> distribution or any action taken or omitted to be taken in reliance
> on
> it, except for the purpose of delivery to the addressee, is
> prohibited
> and may be unlawful. Kindly notify the sender and delete the message
> and
> any attachment from your computer.
> *
> *   For searches and help try:
> *   http://www.stata.com/support/faqs/res/findit.html
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
> 



Prof. Mark Schaffer
Director, CERT
Department of Economics
School of Management & Languages
Heriot-Watt University, Edinburgh EH14 4AS
tel +44-131-451-3494 / fax +44-131-451-3294
email: m.e.schaffer@hw.ac.uk
web: http://www.sml.hw.ac.uk/ecomes

-------------------------------------------------------------------
DISCLAIMER

This message is subject to http://www.hw.ac.uk/disclaim.htm 

-------------------------------------------------------------------

*
*   For searches and help try:
*   http://www.stata.com/support/faqs/res/findit.html
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index