st: Re: Pls help me! question regarding panel fixed effect model

[Kit Baum <baum@bc.edu>]

sacrificial rutabaga	

Although Clive's willingness to help early on a Sunday morning is  
admirable, some of his advice is a bit off base.

(1) ivreg2 will NOT fit a fixed effect model. Mark Schaffer's  
xtivreg2 will do so (and is in fact a 'wrapper' for ivreg2, so will  
allow for all of the ivreg2 options).

(2) cluster-robust SEs allow errors to be heteroskedastic and  
arbitrarily correlated with one another within clusters (but  
independent of those in other clusters). A robust SE treatment allows  
unspecified arbitrary heteroskedasticity. Cluster-robust SEs can  
appear in a purely cross-sectional context. If you are in a panel  
context, you may well want to treat the prospect of non-independently  
distributed errors with something like the Newey-West estimator,  
which as Clive says would be invoked with bw(2)  [which allows for AR 
(1) errors].

(3) You cannot combine HAC estimation ("Newey-West") with cluster- 
robust SEs. Use one or the other. You can combine robust and cluster 
(varname) in any Stata command that allows cluster(), but 'robust' is  
redundant, as cluster-robust SEs are always robust.

Kit

Kit Baum, Boston College Economics and DIW Berlin
http://ideas.repec.org/e/pba1.html
An Introduction to Modern Econometrics Using Stata:
http://www.stata-press.com/books/imeus.html


On May 20, 2007, at 2:33 AM, Clive wrote:


> # If I use cluster option, does it correct for serial correlation and
> heteroskedasticity? I need to correct both in my estimation. What does
> intra
> group correlation mean?
Heteroskedasticity. As for the intra-group correlation, this link may
help to explain:

> As for which model you should use, to achieve what you need, why not
> use Baum, Schaffer and Stillman's -ivreg2-, downloadable from SSC?
> Then you could do something like this:
>
> . ivreg2 y x1 x2 d1 d2 d3, bw(2) robust cluster(group) small
>
> where the d-variables are your time-dummies. This fits an OLS
> fixe-effect model with Newey-West standard errors (although using both
> the -robust- and -cluster()- options might be overdoing things). I'm
> sure Kit, Mark and/or Steve will add further comments on this should
> they feel the need.
*
*   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/