On 3 May 2002 11:11:49 -0700, [EMAIL PROTECTED] (F.
Goldhammer) wrote:

> Yes, I am expecting interesting changes of the concentration
> performance in the 18 minutes, which describe a negative linear trend.
> And there are differences between persons (slope).
> Please could you tell me a �test for trends between persons� to get a
> measure of reliability.

James Diamond posted a reference that has a couple
of usable ideas, even though I find the sections of Winer 
to be hard to read and misleading.  That was:  4.5 and 4.6
in Winer.  Whatever you can read,  is probably a good
start.  However, your data don't start out with the intended
replication/ duplication that  Winer talks about in 4.5. 

You have  TRENDs.

The formula for trend in 4.6  is not easy, though, and
(a) Winer is a bad example when he *accepts*  that
cubic trend with testing the "combined quadratic and
cubic"  and deciding that linear is enough.
(b) Winer does not go so far as to describe the 
trend and the within-subject similarity in contrast
to the between subject difference.   He refers to 
group trends, and not to what you need, which 
is *individual*  trends.

Here is an adaption: something that is feasible for
you to perform and to defend.  

Figure out what it is that makes an intelligible 
contrast, and one that you can compute:  
linear trend; first six versus last six; or whatever.
This can be a variable called "Decline".

Now, looking at the points that make up Decline, 
take (say) alternating points, to make *two*  variables,
with half the points in each -- so that you end up with 
two versions of Decline, D1 and D2.  Now you 
can compute the correlation between D1 and D2.

The reliability of the Scale with the original Decline
is greater than the correlation between D1 and D2:
Here is where you can use a formula for looking at the
reliability of a scale "with twice the length" (as in section 4.5).

This is a counterpart of "internal reliability".
Probably, it is an overestimate of that more sensible
value, test-retest reliability, but it puts an upper limit
on that value (which might be useful to know).


-- 
Rich Ulrich, [EMAIL PROTECTED]

http://www.pitt.edu/~wpilib/index.html
.
.
=================================================================
Instructions for joining and leaving this list, remarks about the
problem of INAPPROPRIATE MESSAGES, and archives are available at:
.                  http://jse.stat.ncsu.edu/                    .
=================================================================

Reply via email to