Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at statalist.org.


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

st: RE: Minimum number of observations per cluster for clustered standard errors in IV?


From   "Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>
To   "statalist@hsphsun2.harvard.edu" <statalist@hsphsun2.harvard.edu>
Subject   st: RE: Minimum number of observations per cluster for clustered standard errors in IV?
Date   Thu, 17 Oct 2013 15:23:38 +0000

JZ,

My understanding from Austin's work (BTW, you cite a 2007 paper by him - the Statalist FAQ asks you to provide a full reference in such cases) is that problems arise with unbalanced clusters (=some clusters have few obs, some have lots), not small clusters (=clusters with few obs) per se.

The easy way to see that small clusters per se are not necessarily a problem is by noting that the classic Eicker-Huber-White-sandwich-heteroskedasticity-robust covariance estimator is just a special case of the cluster-robust covariance estimator with 1 observation per cluster.

--Mark

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu [mailto:owner-
> statalist@hsphsun2.harvard.edu] On Behalf Of Jen Zhen
> Sent: 17 October 2013 14:38
> To: statalist@hsphsun2.harvard.edu
> Subject: st: Minimum number of observations per cluster for clustered standard
> errors in IV?
> 
> Dear listers,
> 
> I sent this question as an add-on to my earlier inquiry under a different
> headline and just realized that this way it may not reach its audience, hence I'm
> putting it into a different thread now:
> 
> I am trying to choose how to cluster my SEs in ivreg2. I understand from Nichols
> (2007) and Angrist and Pischke that I should pick a dimension with which I get
> at least 42 or 50 different clusters. Now that's in extremo easy to satisfy if I
> cluster by observation. But I guess there should also not be too few
> observations per cluster. But I haven't found anything on what is "too few"
> here.
> 
> Is there any evidence providing for a minimum number of obs per cluster, or a
> minimum on the average, median, P10 etc number? Any other useful criterion?
> 
> I suppose that when using 2 way clustering then each of my 2 dimensions
> should on their own satisfy the respective requirement?
> 
> Thank you so much and kind regards,
> JZ
> *
> *   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/


© Copyright 1996–2018 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   Site index