st: RE: heteroskedasticity-robust standard errors using "xtivreg2, fe"

["Schaffer, Mark E" <M.E.Schaffer@hw.ac.uk>]

Nina,

-xtivreg2- with the "robust" option implements standard Eicker-Huber-White-sandwich-etc. heteroskedastic-robust SEs.

Stock & Watson (2008) showed that these are inconsistent, but only in a large-N fixed-T context.  The size of the bias is decreasing in T, so if you have decent number of observations in the time series dimension, it might not be much of a problem.

-xtivreg2- has an undocumented option "sw" that implements the Stock-Watson bias-corrected heteroskedastic-robust SEs.  It's undocumented because at the time I programmed it, I wanted to replicate a published example but couldn't find one.  Feel free to use this option, but caveat emptor.

And if someone out there knows of a published example of these Stock-Watson SEs employed on an openly-available dataset so that I could replicate, please let me know!

HTH,
Mark

> -----Original Message-----
> From: owner-statalist@hsphsun2.harvard.edu 
> [mailto:owner-statalist@hsphsun2.harvard.edu] On Behalf Of 
> Nina Neubecker
> Sent: 09 February 2012 08:28
> To: statalist@hsphsun2.harvard.edu
> Subject: st: heteroskedasticity-robust standard errors using 
> "xtivreg2, fe"
> 
> Dear Stata listers,
> 
> I am currently estimating a fixed effects model on cluster 
> data (38 countries reporting data on 4 occupations). Since 
> the number of clusters is below the often cited critical 
> value of 50, I would like to estimate "simple" 
> heteroskedasticity-robust standard errors (and not 
> cluster-robust st.e.). As this is no longer possible with the 
> official STATA command "xtreg, fe" in STATA 11, I searched 
> for alternative commands and learned about the module 
> "xtivreg2". As it says in the help-file to this module, 
> "xtivreg2" can also be used to estimate a simple fixed 
> effects model without endogenous regressors.
> (In a later step I will also treat one of my regressors as
> endogenous.)
> I am now wondering whether "xtivreg2, fe" with the options 
> "small" and "robust" for a simple fixed effects model reports 
> the bias-corrected standard errors proposed by Stock & Watson 
> (2008) to tackle the problem that the estimated standard 
> errors à la White (1980) are inconsistent when applied after 
> mean-differenced estimation if the number of observations per 
> cluster is small (see Cameron & Miller, 2010, "Robust 
> Inference with Clustered Data")?
> 
> Thank you very much in advance for your help!
> 
> Nina Neubecker
> 
> *
> *   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.

Heriot-Watt University is the Sunday Times
Scottish University of the Year 2011-2012



*
*   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/