Eric Turkheimer <[EMAIL PROTECTED]> wrote:
> What I want to do is to plot the variance of a variable as
> a smooth function of the values of another variable. It would
> seem as though most of the methods for bivariate smoothers
> should work, but I can't find any theory on the matter.
[...]
I suggest approaching the problem indirectly: first construct
a model of the joint distribution of the variables involved,
then derive the conditional distribution from that.
The joint distribution could be modeled by, say, a mixture of
Gaussians. In that case, the conditional distributions are
again mixtures of Gaussians, with mixing weights which vary
from location to location. Then it is easy to compute the
conditional variance as a function of location.
I have some notes on this problem in Appendix E of my
dissertation, which you can find at
http://civil.colorado.edu/~dodier/publications.html. See also
Sec. 6.7 for an example. (I'm sure that all this has been
extensively published, but I wanted to collect my notes into
one place.) If you want more help, just ask.
Regards,
Robert Dodier
[EMAIL PROTECTED]
-----------------------------------------------------------
Got questions? Get answers over the phone at Keen.com.
Up to 100 minutes free!
http://www.keen.com
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================