[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
st: RE: fixed effects with robust standard errors
Credit where credit is due: this suggestion
was I believe mostly due to Mark Schaffer, as
an excavation of the multiple layers of the
posting will reveal. I just added an extra
comment, I believe.
> I am trying to do what Nick suggested in a posting a few
> years back (see
> below), i.e. to estimate a fixed effects model with
> heteroskedasticity-robust standard errors by transforming the data to
> deviations from means and then running the regress command, with the
> option robust specified.
> However, I believe I need to correct the degrees of freedom
> for the number
> of dummy variables this procedure implicitly estimates. Is it
> possible to
> do this within the regress command; should I (not being an expert
> programmer in the least) try to program it myself; or is
> there some other
> way to change the degrees of freedom used?
> From "Nick Cox" <email@example.com>
> To <firstname.lastname@example.org>
> Subject RE: st: robust st. errors and fixed effects
> Date Mon, 21 Oct 2002 15:50:47 +0100
> > > is there a possibility to estimate a fixed effect model,
> > controlling for
> > > heteroskedasticity.
> > > the comad robust and xtreg don`t work togetehr.
> > > Is there an other way to control for h.?
> > > thanks for your help
> > > peter
> > >
> > ...
> > As was suggested in a posting to Statalist earlier today,
> > you can get
> > -regress- to estimate a fixed effects model for you, and -regress-
> > will of course generate robust SEs.
> > The posting suggested estimating a "least squares dummy variable"
> > (LSDV) regression, with a dummy for each observational unit
> > (individual, firm, whatever). This is OK unless you have a lot of
> > different units, in which case you get more dummy variables
> > than you
> > can reasonable handle. A slightly more laborious but also
> > equivalent
> > method is to transform the data by putting it into mean-deviation
> > form, and then estimating on the transformed data.
> Mark mentioned -areg- in passing in his posting
> (abbreviated here): this is just to flag its
> availability given many dummy variables.
* For searches and help try: