Re: st: Constructing a variable from standard deviations

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

--- On Mon, 22/11/10, M.P.J. van Zaal wrote:
> However, you guys claim that the estimates from this
> procedure would be meaningless.

We did not say that. This is the solution proposed by Stas
and I said about it: "Stas' solution works, but is 
substantively awkward". That is not the same as meaningless.
It means that you can use it, but than you'll have to show
in your paper why a model that assumes constant residual 
variance still gives you unbiased estimates of differences
in residual variance across groups. The simulation I gave
earlier shows that that is the case. If you still want to
use Stas' solution you need address this counterintuitive 
step in your analysis. I would start by defining exactly 
what that residual variance is, and go over the exact 
definition of the residuals and their variance in linear 
regression, see what happens with those residuals in case
of heteroskedsticity, and than try that with the variances
that you want to estimate. Looking for proofs like that is
a sequential process, so there is no guarantee that such 
an initial plan works, but this is how I would start.
(Actually I would want to avoid this, and use either my
or Kit's solution, but that does not mean that Stas' 
solution is meaningless).

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/