Re: st: an estimation method question

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

--- On Tue, 16/3/10, Xiang Ao  wrote:
> There are different numbers of founders for each firm. 
> The smallest number is 2, it can be as many as 10
> co-founders.  As far as I understand, the number of
> categories need to be fixed in the methods you
> mentioned. 

That is right, but more importantly the categories need
to have the same meaning across observations. For example
imangine we are looking at the proprotion of spending in 
a firm on labour and capital (I am no business economist 
so forgive me if this doesn't make substantive sense). 
Across firms the proportion spent on labour has (sorta) 
the same meaning. The aim of such an analysis is to find
variables that make some firms spend more on labour and 
others more on capital goods.

I find it harder to come up with something similar for 
the different founders. Who would fall in the first 
category, who in the second, etc. (alphabetical order?). 
A solution would depend on what your unit of analysis 
is for your study, i.e. who do you want to study: the 
firms or their founders. 

If you want to study the firms, than I could imagine 
that you are interested in explaining differences in 
the "structure" of ownership. I would than try to 
operationalize that in one number per firm and use an 
appropriate regression like command to analyse that. 
For example, I could imagine someone interested in the 
share of the largest founder as a measure of 
concentration of ownership, in which case I would use 
the -glm- trick or -betafit- (zero or one proportions 
are no problem there as you stated that the minimum 
number of founders in your dataset is 2). Alternatively,
you could try other summaries of the concentration of
ownership like entropy. Don't take my reference to 
entropy too literaly, I only know that there are many
such measure, and entropy was the only one I could
remember.

If you want to study founders, I would for now forget
about the constaint that the proportions should add
up to one, and transform the data to a panel dataset 
where the observations are founders nested in firms. 
I would than try -xtgee- to model these proportions,
just like the -glm- trick. It will be a population
averaged model, so no individual level effects, 
which seems to be considered a problem in economics.
However, it would be a good place to start.

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/