Ah--interesting, and I applaud the terminology change! Robust is
certainly an ambiguous description, as is sandwich, whereas het-robust
and cluster-robust are fairly specific (esp. when combined with a
reference to the literature). Maybe new vce() term hrobust or
hetrobust? Please don't make us spell out heteroskedasticity... ;)
McCulloch, J.H. 1985. "On Heteros*edasticity." Econometrica 53(2): 483.
http://www.jstor.org/stable/1911250
On Tue, Aug 25, 2009 at 11:45 AM, David M. Drukker<[email protected]> wrote:
> This post continues the thread about robust standard errors for the
> generalized least-squares estimator implemented in -xtreg ,re-.
>
> Eric de Souza <[email protected]> asked whether there will be
> separate -vce()- options on -xtreg, re- to produce consistent estimators of
> the variance-covariance of the estimator (VCE) when the errors are
> heteroskedastic and when the errors are serially correlated.
>
> The answer is yes, there will be separate options for these cases.
>
> We will add a new -vce()- option for -xtreg ,re- that will produce the
> observation-level robust estimator that is consistent in the presence of
> some types of heteroskedasticity with -xtreg, re-. This observation-level
> robust estimator was produced by -vce(robust)- in Stata 10.1.
>
> In the Stata 11 -xtreg ,re-, we changed the meaning of the -vce(robust)- to
> be a synonym for -vce(cluster panelvar)-. We made this change because
> researchers increasingly expect this definition of robust. As Cameron and
> Trivedi (2005, section 21.2.3) and Wooldridge (2002, section 10.4.2) make
> clear, this individual-level robust estimator of the VCE is the natural
> "robust" estimator when sampling vectors of observations from individuals in
> the population. This VCE estimator is robust at the individual-level
> instead of at the observation level. This individual-level robust estimator
> of the VCE is more attractive in most real-world cases than the
> observation-level robust estimator because sampling is performed at the
> individual level. This individual-level robust estimator is consistent in
> the presence of heteroskedasticity and autocorrelation.
>
> Summing up, when specified with -xtreg re-, -vce(robust)- will continue to
> be a synonym for -vce(cluster panelvar)-. We will add a new, as yet
> unnamed, -vce()- option that will produce the observation-level robust VCE
> estimator that is robust to some types of heteroskedasticity.
>
>
> David
> [email protected]
>
>
> References
> ----------
>
> Cameron, A. C. and P. K. Trivedi. 2005. Microeconometrics: Methods and
> Applications. New York: Cambridge University Press.
>
> Wooldridge, J. W. 2002. Econometric Analysis of Cross Section and Panel
> Data.
> Cambridge, MA: MIT Press.
>
*
* 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/
