Statalist


[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]

st: xtscc vs. cluster - once again; was heteroskedasticity


From   "Doctor Theopolis" <theopolis@10base-t.com>
To   statalist@hsphsun2.harvard.edu
Subject   st: xtscc vs. cluster - once again; was heteroskedasticity
Date   Wed, 25 Jun 2008 14:11:29 +0200

Dear statalisters,

A similar question was already discussed, (see:
http://www.stata.com/statalist/archive/2007-05/msg00181.html) but still
I'm not sure how to proceed with my case. I want to estimate a
fixed effects model, two-way (including time dummies). So the first question is,
does it make sense to use xtscc in this case, as  -as Daniel remarked -
including time fixed effects  would be a parametric way of  correcting
for cross-sectional
dependence? So I could use FE two-way with cluster() option and be on
the safe side regarding heteroskedasticity and correlation within and
between panels? But I wonder if this approach might produce  too
conservative standard error estimations in a FE regression?

The unbalanced panel I use consists of n=10.000 cross-sectional units
with up to t=20 time-periods, (and I don't know if this matters, but as
I use FE regression only observations with time-variant features are
used, so the estimation of the coefficients of interest are based on
about n=300);

When I use the oneway FE regression, the standard errors of xtscc are
biggest but nearer to the panel-robust SEs (cluster -option).  When I
use two-way FE for all models the standard error estimates of  xtssc are
between the normal FE and FE with panel robust SE; in this case the
xtssc SE are about a third lower compared to the panel-robust SEs.
Referring to the discussion before that would mean that when including
the time-dummies the remaining correlation between cross-sectional units
is negative (????)-whatever that may mean?
So I'm not sure which SE estimations to use? Could I rely on xtscc SE?

I tried to test for cross-sectional dependence with xtcsd but I can't
extend matsize in Stata sufficiently.

Any help or hints are appreciated
Natasha
*
*   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/



© Copyright 1996–2014 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index