Bookmark and Share

Notice: On April 23, 2014, Statalist moved from an email list to a forum, based at

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: st: Estimating firm level data on regional level data using a within estimator.

From   John Antonakis <>
Subject   Re: st: Estimating firm level data on regional level data using a within estimator.
Date   Sun, 06 Jun 2010 14:17:35 +0200

Hi Natasha:

Actually, you can have your random-effects cake and eat it too with a fixed-effects cherry on the top. See:

Mundlak, Y. (1978). Pooling of Time-Series and Cross-Section Data. Econometrica, 46(1), 69-85.

That is, suppose that x is the region varying covariate that you wish to estimate.

Taking the cluster mean of x and including as a covariate in the regression then essentially includes the fixed-effect. This procedure is nicely explained here:

Rabe-Hesketh, S., & Skrondal, A. (2008). Multilevel and Longitudinal Modeling Using Stata. College Station, TX: Stata Press.

I have an in-press paper where we explain it too (if you e-mail me off-line, I will send it to you as well as to anyone else who'd like copy):

Antonakis, J., Bendahan, S., Jacquart, P., & Lalive, R. (in press). On making causal claims: A review and recommendations. The Leadership Quarterly, 21(6).

Not including the fixed-effect would be a very fatal flaw to make; this obsession in econometrics is a well founded one because if the assumptions of the random effects estimator are not met (with the random-effects being orthogonal to the regressors) then the estimator is not consistent. More fields are paying attention to this matter, including sociology, and applied psychology/management. For further discussion see the following as well:

Halaby, C. N. (2004). Panel models in sociological research: Theory into practice. Annual Review of Sociology, 30, 507-544.



Prof. John Antonakis, Associate Dean Faculty of Business and Economics
Department of Organizational Behavior
University of Lausanne
Internef #618
CH-1015 Lausanne-Dorigny

Tel ++41 (0)21 692-3438
Fax ++41 (0)21 692-3305

Faculty page:

Personal page:

On 06.06.2010 13:56, Maarten buis wrote:
--- On Sun, 6/6/10, natasha agarwal wrote:
I estimate it with a within estimator, i.e. ur fixed
effects. When I do so the minute I include all the time dummies, FDI goes insig but thats my main variable of
interest. The problem that I have found out is that FDI
is behaving like a time dummy because all the firms
within a region for each year will have the same value
of FDI.

The problem with macro quantities is that they are usually highly correlated with other macro quantities and time, so it is hard to disentangle the effects. Moreover, there are only so many observations possible on the macro level. These two together means that often you have very little statistical power to play with.
The advantage of fixed effects regression is that it provides a
strong reason for believing that you controlled for some of the unobserved and possible confounding variables on the the region level. However, the price you have to pay is that you loose a lot of power; i.e. you loose all the information from the comparison between regions. Since you don't have a lot to begin with, that is very bad news.

Random effects don't allow you to controll for unobserved confounding variables on the region level (the random effects are
assumed to be uncorrelated with the observed variables), but it
does allow you to also use the information from the comparison
between regions.

Then you have rather pragmatic discipline specific reasons: e.g.
within economics they tend to be so obsessed with fixed effects
regression, that you will have a hard time getting a random effects model published.

Hope this helps,

Maarten L. Buis
Institut fuer Soziologie
Universitaet Tuebingen
Wilhelmstrasse 36
72074 Tuebingen

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

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