In article [EMAIL PROTECTED] (Donald Burrill) wrote:
> Sounds like a prediction or calibration kind of problem. As Joe Ward
> pointed out, raw regression coefficients, and standard errors of
> measurement, are more stable than correlation coefficients.
Yes, that's right. Regression will be important in our application,
too. But, as others have have remarked, the bivariate normality
assumption is not needed for regression. Our current problem is
really whether there is any value in putting CIs on the correlation
coefficients.
> That you have not imposed artificial restrictions on the range of the
> variables does not imply that their range is not restricted. After
> all, you were asking about bivariate normality with respect to a
> (presumably) real (and therefore finite) data set. There may be
> natural constraints on their range(s).
The subjects were human, so, yes there are natural constraints. But,
these don't result in distributions that are incompatible with random
variables, if that is the issue you were raising.
PS Thank you to Eric for the references.
--
Znarf
Sent via Deja.com http://www.deja.com/
Before you buy.
=================================================================
Instructions for joining and leaving this list and remarks about
the problem of INAPPROPRIATE MESSAGES are available at
http://jse.stat.ncsu.edu/
=================================================================
