Re: reliability

On 3 Jun 2005 09:30:21 -0700, mcam54@xxxxxxxxxxx wrote:

> I am trying to calculate test retest reliability for psychological
> measurement scales. I assume that the intraclass correlation is the
> most appropriate measure. But previous studies on similar scales often
> seem to employ Pearson's r. Which one do I choose. And, if I choose
> the ICC, which model do I choose and do I use absolute agreement or
> consistency.
>
> Thanks! I have ordered Norman and Streiner's text on this topic which
> should help.

There are two Threads on the subject in May, 2005,
in sci.stat.consult, with 18+4 posts.

Google-groups on < group:sci.stat.* Pearson ICC >
for those, and more.

I strongly recommend Pearsons plus t-tests for doing
the actual scale development, with ICC used for summary
*if* that is the local convention; or, perhaps, if a single-
number summary is desired; or if you need a multi-rater
figure for (say) a power analysis. "Choosing the model" is
one of the hazards of communicating with the ICC.

--
Rich Ulrich, wpilib@xxxxxxxx
http://www.pitt.edu/~wpilib/index.html
.