You can also say there is some mistitling there. The colloquial term
"robust" got stuck to the particular form of the standard errors --
the sandwich estimator based on Taylor series linearization of the
estimating equations, even though the estimator is not robust in the
strict sense of Swiss robustness theory (bounded influence functions).
This is the appropriate estimator in a general situation where the
data are not i.i.d. from the model you are estimating, where the "not"
part can apply to: (i) not the first i. of i.i.d.: not independent
(e.g., if you have clusters, or, with some extra push to -newey-
standard errors, autocorrelation); (ii) not the second i. of i.i.d.:
not identically distributed (e.g. there's heteroskedasticity, and this
is where the term "robust" is generally coming from, I guess); (iii)
the model does not hold -- and survey statisticians are used to
dealing with fixed characterisitics, so their approaches to
statistical models like regression is to say that there are fixed x's,
there are fixed y's, and there are population level parameters that
are estimated from the random sample. So instead of putting a lot of
trust into say assumption of linearity in the regression, you just say
that for any population, there is the best fitting surface, and with
our sample estimates, we are trying to get to that surface. The
variability then is only due to sampling, not due to the random errors
in the regression. So the standard errors reported for complex surveys
stem from a different paradigm, although numerically and
algorithmically they are essentially the same as the
Eicker-Huber-White standard errors in regression.
On 4/4/07, Maarten buis <[email protected]> wrote:
--- Eva Gottschalk <[email protected]> wrote:
> does anyone know how I can get non-robust standard errors after
>
> reg xy [pweight=weight] ??
>
> (Stata calculates robust standard errors but since I don't have any
> clusters, I don't see the need of it.)
Robust standard errors are not only for clusters, and you do want
robust standard errors after pw, even if you don't know that yet.
see for instance:
http://www.stata.com/statalist/archive/2005-01/msg00299.html
--
Stas Kolenikov
http://stas.kolenikov.name
*
* For searches and help try:
* http://www.stata.com/support/faqs/res/findit.html
* http://www.stata.com/support/statalist/faq
* http://www.ats.ucla.edu/stat/stata/
