Re: st: Heavy representation in a sample

[Maarten buis <maartenbuis@yahoo.co.uk>]

--- On Sun, 20/3/11, Claude Francoeur wrote:
> I would like to know if there is any econometric problem
> when making an OLS regression using a sample that is heavily
> represented by a subset within a variable? For instance, we
> are using country legal systems as an explanatory variable :
> 65% is Anglo-Saxon and the rest is divided between French,
> German and Scandinavian. Is there any econometric problem
> using such a sample? If so, what are the methods to correct
> it?

You'll have a statistical problem in the sense that such a 
pattern will cost you power. Effects are comparisons of groups.
So you need an estimate of the expected value of your dependent
variable for Anglo-Saxon, French, German, ans Scandinavian 
countries, and the differences (or ratios) between these means
are your effects. If any of these means is measured with little
precision (because there are only a few observations within 
that type of country), then that will have a negative impact on
the precision with which you can measure the effect. If you are 
not very sure about one of the means, then you are also not sure
of the difference between that mean and another mean. It is not 
that you estimates or standard errors will be biased, it is 
just that the standard errors will be larger then you would like.
The solution is to get more data on those countries where data
is sparse, but since the number of Scandinavian, German, etc. 
countries is prety much fixed, that is likely not an option. 
So, you'll just have to live with the lack of power. 

Hope this helps,
Maarten

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

http://www.maartenbuis.nl
--------------------------



      

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/