Tom,
In my own work I have relied on the following recent articles to understand the distinction between cluster-robust standard errors and fixed effects:
Rick L. Williams, A note on robust variance estimation for cluster-correlated data. 56 BIOMETRICS 645 (2000).
A. Colin Cameron et al., "Bootstrap-Based Improvements for Inference with Clustered Errors," UC-Davis Department of Economics Working Paper # 06-21, available at http://www.econ.ucdavis.edu/working_paper_info.cfm?pid=368.
The most influential recent reference is Marianne Bertrand et al., How Much Should We Trust Differences-in-Differences Estimates? 199 Q. J. ECON. 249-275 (2004).
Gabor K�zdi, "Robust Standard Error Estimation in Fixed-Effects Models." Working paper available at http://ideas.repec.org/e/pke76.html.
Clustering is probably necessary (meaning highly desirable) when intra-class correlation coefficients are "high". You can test for this using one-way random-effects ANOVA techniques, and in particular Stata's -iclassr- routine, combined with the -ems- option if your panels are unbalanced.
Jason
-----Original Message-----
From: [email protected] [mailto:[email protected]] On Behalf Of Thomas Erdmann
Sent: Thursday, January 18, 2007 12:10 PM
To: [email protected]
Subject: st: AW: re: panel re vs. fe model
Mark, thanks again for the further clarification.
Kit, Jason, in saying that I address time- and firm effects I omitted
"biasing SEs".
A reference which I used to learn about this is:
Petersen, 2006. Estimating Standard Errors in Finance Panel Data Sets:
Comparing Approaches, Working paper.
http://www.kellogg.northwestern.edu/faculty/petersen/htm/working.htm
