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

Re: st: Two-Way FE Estimation

From   Asgar Khademvatani <>
Subject   Re: st: Two-Way FE Estimation
Date   Fri, 15 Aug 2008 11:12:09 -0600

Dear Maarten buis,

Thank you for your answer and time. But, I got the following problem in running your suggestion for implementing two-way FE:

As You agreed on that the following pseudo-code is a two-way FE implementation:

xtreg Y d1 d2 d3 X, fe
where Y is dependent variable(s) and X is independent variable(s), and (d1 d2 d3) are industry dummies.

I run this model and got the following error:

independent variables are collinear with the panel variable ind

This shows that to prevent collinearity, we do not need include industry dummies (d2 d2 d3) , since -xtreg- itself takes care of time periods and industries defined as panel id. This means, we need to run the above pseudo-code without industry dummies. It will be implicitly accounts for a two-way FE. By the way, we can not add time dummy as well, since T is enough large and will lose degree of freedoms.

I am wondering whether you are agree with my above discussion and suggestion in implementing FE model?


On Wed, 8/13/08, Maarten buis <> wrote:

From: Maarten buis <>
Subject: st: Re: Two-Way FE Estimation
To: "stata list" <>
Date: Wednesday, August 13, 2008, 4:00 AM

--- Asgar Khademvatani wrote to me privately:
> You meant, if I have 4 industry in my panel data set, then a two-way
> FE can be implemented as:
> xtreg Y X d1 d2 d3, fe where d1 d2 d3 are industry dummies and
> -xtreg- take cares of time periods? If so, Do not you think this is a
> double accounting for a one-way fixed effects without taking care of
> time periods? Please advise

First of all, it is better not to respond of list. This is discussed in
section 5 of the Statalist FAQ:

There are two ways of thinking about what you want: either you want to
control for industry and period, or you want to control for industry,
period, and the interaction between industry and period. Which one you
want is a subtantive choice and can only be made by the researcher.
What I proposed is the first option, and that boils down to the
pseudo-code you have given above (assuming you -xtset- on period
first). For the alternative you create a new id variable that is the
combination of period and time, and use that in xtreg. In pseudo-code:

egen id = group(period industry)
xtset id
xtreg Y X, fe

Hope this helps,

Maarten L. Buis
Department of Social Research Methodology
Vrije Universiteit Amsterdam
Boelelaan 1081
1081 HV Amsterdam
The Netherlands

visiting address:
Buitenveldertselaan 3 (Metropolitan), room Z434

+31 20 5986715

Send instant messages to your online friends
* For searches and help try:

* For searches and help try:

© Copyright 1996–2015 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   What's new   |   Site index