Thanks for the replys so far. I've given more information below.
In article [EMAIL PROTECTED] (Donald Burrill) wrote:
> To whom, for what purpose(s) ? The "several bivariate associations"
> part rather suggests that you'll want to be making comparisons,
> implicitly if not explicitly; and even if you don't, readers will.
We want to report the correlation between two methods of measurement for
about 20-30 different response variables. It isn't expected that the
methods will give the same individual or average values, but the degree
of linear correlation is useful in future large studies to make
adjustments for using the less costly method. Sample size is about
100-150.
Comparing the correlation sizes between different variables isn't of
special interest, since in future studies these variables will usually
be examined separately.
> What else will you report to give evidence of comparability?
> (For example, are all the variables in question equivalently (in any
> sense!) restricted in range?)
The variable values aren't restricted in range, and the associations are
approximately linear for most of the variables.
> Depends on (a) sample size, (b) the nature(s) of the departure(s) of
> interest (shape & size of effect, at least).
Most of the correlation sizes are about 0.4 to 0.7, but some are outside
this range.
There are theoretical reasons for reporting Pearson as opposed to
Spearman correlations in our application (eg for making adjustments), so
an issue is whether CIs on the former are worth reporting, or not.
--
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/
=================================================================
