Statalist The Stata Listserver

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

Re: st: panel data: xtrege, re vs. xtreg, fe vs. regress...cluster

From   Christer Thrane <>
Subject   Re: st: panel data: xtrege, re vs. xtreg, fe vs. regress...cluster
Date   Mon, 12 Mar 2007 20:12:38 +0100

In the 2003 book entitled "Handbook of data analysis", by Bryman/Hardy (Sage), Trond Petersen has a chapter on Random versus Fixed effects.

His advice mirror Phillipp's. That is, report the RE estimator, the BE estimator and the FE estimator because they tell us different things about the data structure.

Personally I find this more satisfying than the "mechanical" the-RE-and-the-FE-differ-use-FE-advise, a la Hausman.

Best regards,

----- Original Message ----- From: "Philipp Rehm" <>
To: <>
Sent: Sunday, March 11, 2007 11:20 PM
Subject: Re: st: panel data: xtrege, re vs. xtreg, fe vs. regress...cluster

Hi Mike,

I usually find it helpful if people offer the results from three different estimators: a RE estimator (using, say MLE or GLS), a between estimator, and a within (i.e. FE) estimator. The RE is just a weighted average of the within and between estimators. This allows the reader (and author) to see:
- whether the effect holds across firms (between estimator, which - essentially - is just using the mean values over years for each firm)
- whether the effect holds across time within firms (FE estimator)
- or both (RE estimator, or my interpretation thereof)

OLS (with sandwich estimator) just doesn't allow you to figure out these details.


You may find the following book useful, which I can highly recommend:
Sophia Rabe-Hesketh and Anders Skrondal (2005): Multilevel and Longitudinal Modeling Using Stata (

Michael Pfarrer wrote:

Hi Everyone,

I know this topic has been debated for years on the listserv with many helpful responses, but I thought I'd ask again to help me clarify things.

I have a dataset of 291 firms over 15 years (N=4365). What, in essence, is the "best" model to run for this analysis? I understand the difference between fe and re and the use of the Hausman test, which, for me, is n.s. But is there really an inherent difference between xtreg, re and regress, cluster(id)? The OLS regress, cluster(id) normally gives me "better" results, but I certainly want to understand if it is "correct". I also understand that xtreg, re can decompose into OLS. So, if you had these data across this panel with a continuous DV and continous and binary IVs, what would you run? The DV is the Cumulative Abnormal Return from a 3-day event window (-1, +1)and the IVs are firm financial characteristics--ROA, volume, reputation dummy, etc.

Many thanks,

Michael D. Pfarrer
Department of Management and Organization
Robert H. Smith School of Business
University of Maryland
College Park, MD 20742
Phone: 301.653.0458

* For searches and help try:
*   For searches and help try:

© Copyright 1996–2017 StataCorp LLC   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index