Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: cluster robust SE options in 2SLS


From   Austin Nichols <austinnichols@gmail.com>
To   statalist@hsphsun2.harvard.edu
Subject   Re: st: cluster robust SE options in 2SLS
Date   Mon, 23 Mar 2009 10:51:30 -0400

Shang Ha <shang.ha@yale.edu>:

I would ignore the comment from Joshua A Shindell at
http://www.stata.com/statalist/archive/2009-03/msg01218.html on more
substantive grounds--the cluster-robust VCE is useful in a huge
variety of situations, of which serial correlation within panel is
merely one instance.  I would even say the majority of cases deserving
of cluster-robust VCE corrections are cross-sectional applications;
for example, any survey, or regression using covariates measured at
different "levels" e.g. individual outcomes regressed on
state/province policy variables.  However, the cluster-robust VCE is
biased downward in "small" samples, i.e. small number of clusters
(also severely unbalanced clusters), so in fact if the cluster-robust
VCE is inappropriate for your case, it will not necessarily "give a
more conservative of an estimate of your standard errors" as Joshua A
Shindell claims--quite the opposite is possible.

See also http://ideas.repec.org/p/boc/usug07/07.html
*
*   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/



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