On Thu, 14 Sep 2000 00:32:08 GMT, [EMAIL PROTECTED] (J Dungan) wrote:
> A colleague has asked me about what flavor of linear regression to
> use when the explanatory variable(s) are proportions. I know that
> logistic regression handles situations when the *response* variable
> is a proportion, but are there special considerations when the
> explanatory vars are proportions (0 to 1)? I know that such vars
> should be independent, and they cannot all be independent if they
> sum to 1.
...
If "compositional data" is the question, my stats-FAQ has references,
http://www.pitt.edu/~wpilib/statfaq/compfaq.html .
I looked at some of these, a few years ago. J. Aitchison writes
about (for instance) mineral compositions of geological deposits.
Recommendations included: using the logs of ratios of selected
components.
I don't know whether these relate at all to Bob Wheeler's comments.
--
Rich Ulrich, [EMAIL PROTECTED]
http://www.pitt.edu/~wpilib/index.html
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
