Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.
From | "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk> |
To | <statalist@hsphsun2.harvard.edu> |
Subject | st: RE: Changes to two way clustering ivreg2 update |
Date | Mon, 31 Jan 2011 13:27:28 -0000 |
Sam, There was a bug fix in 3.0.04 that might be the explanation. This is the one in the notes at the bottom of ivreg2.ado that refers to a bug relating to cluster+kernel, which is a specific case of 2-way clustering. In any case, if you contact me off-list with more details, I can investigate and see if I can track down the source of the change. Yours, Mark (ivreg2 co-author) > -----Original Message----- > From: owner-statalist@hsphsun2.harvard.edu > [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of > Samuel Bazzi > Sent: Monday, January 31, 2011 1:02 AM > To: statalist@hsphsun2.harvard.edu > Subject: st: Changes to two way clustering ivreg2 update > > Were there changes implemented in the updated -(xt)ivreg2- calculation > of standard errors in the case of two-way clustering? After updating > to the most recent version of -ivreg2- one week ago, I am unable to > replicate earlier results on standard errors from cross-country panel > regressions with clustering on country and year. Point estimates and > one-way clustering results remain unchanged. I've scoured the detailed > update/version log but found nothing immediately relevant to this > issue. The old version of -ivreg2- was 3.0.01 8Feb2010. The new > version is 3.0.05 17June2010. -xtivreg2- was in both cases the 1.0.11 > 17June2010 version. Prior estimates were therefore based on a pairing > of 1.0.11 -xtivreg2- and 3.0.01 -ivreg2-, while new estimates were > based on 3.0.05 -ivreg2-. Also, all estimates were performed on Stata > version 10.1. > > On a possibly related note, I now find that two-way clustering (with > the updated program) produces standard errors that are in some cases > smaller than and in all cases much closer to the one-way clustering > (on country) results. I understand how this can arise mathematically > but fail to grasp why the new two-way clustered std errors are so much > smaller than ones implemented with the older version of the program. > > Have there been any changes in internal programs across these versions > of -ivreg2-? The fact that neither the point estimates nor the one-way > clustered std errors change leads me to believe that something differs > in the underlying -(xt)ivreg2- programs used to generate the two-way > clustered var-covar matrix. > > Thanks for your consideration. > > -Sam > * > * 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/ > -- 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/statalist/faq * http://www.ats.ucla.edu/stat/stata/