Bookmark and Share

Notice: On March 31, it was announced that Statalist is moving from an email list to a forum. The old list will shut down on April 23, and its replacement, is already up and running.

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

Re: st: Country and population weights in panel analysis

From   Jeffrey Wooldridge <>
Subject   Re: st: Country and population weights in panel analysis
Date   Sun, 24 Feb 2013 21:54:48 -0500

You are describing a correction for heteroskedasticty due to computing
averages over different population sizes. This is a pretty standard
problem -- treated, for example, in my book "Introductory
Econometrics: A Modern Approach," 2013 -- although the panel data
aspect raises some interesting issues. The main thing to remember is
that weighting by population is only efficient if the underlying
individual-level model satisfies the classical assumptions -- constant
variance and serial independence. The latter is very unlikely to be
true, and assuming away heteroskedasticity at the individual level is
too strong. So, I would do it with and without the population weights.
But in both cases compute the "cluster(id)" standard errors. These
will allow for heteroskedasticity of any form as well as serial
correlation. Hopefully you get similar estimates of the betas. The
standard errors will essentially tell you which approach is more


On Fri, Feb 15, 2013 at 1:36 AM, ARDE DE <> wrote:
> Dear All:
> We are running a panel analysis using aggregate data (like income,
> unemployment etc.) for 24 counties (regions) in some country. We have
> eight years of data. The counties differ substantially in size
> (population: from 500.000 to 20 million) such that the averages that
> we use in theory are not equally reliable, a fact that our model
> should reflect.
> We have two questions:
> 1) I know that in stata we can specify weights for panel regressions.
> But do you know about any references that may guide us to the correct
> modelling of population weights and give some more theoretical
> background? I was searching the panel books of Wooldridge (2010) and
> Matyas (2008). They are dealing with stratification issues which
> appears to be much more complicated than what I was looking for.
> 2) A fact that confuses me: Why do most authors not consider country
> weights in well-known cross country studies in the literature? Just
> because they tested for heteroskedasticity and found none? I searched
> a large economics journal database for cross country panel
> regressions, but the question of population weights seems not to be an
> issue. I am confused because I was thinking it is an important element
> of the model (even if empirically there seems to be homosked.)
> Any help/hints will be highly appreciated
> Arde
> *
> *   For searches and help try:
> *
> *
> *
*   For searches and help try:

© Copyright 1996–2016 StataCorp LP   |   Terms of use   |   Privacy   |   Contact us   |   Site index