Date: Thu, 18 Dec 2008 06:31:29 -0500
Reply-To: Jim Groeneveld <jim.1stat@YAHOO.COM>
Sender: "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From: Jim Groeneveld <jim.1stat@YAHOO.COM>
Subject: Intraclass Correlation alternatives
Hi friends,
I am searching for ways to calculate the Intraclass Correlation and found
the SAS macro IntraCC.sas of Robert M. Hamer.
The macro produces 6 types of ICCs, based on Shrout & Fleiss (and 2
comparable ones based on Winer).
The 6 types can briefly be described and discriminated as:
Estimating the reliability of: a single rating the mean of several
ratings
--------------------------------- --------------- ----------------------
-----
* Raters for each subject
are selected at random 1 4
* The same raters rate each case.
These are a random sample. 2 5
* The same raters rate each case.
These are the only raters. 3 6
More info can be found in the following urls:
http://ourworld.compuserve.com/homepages/jsuebersax/icc.htm
http://support.sas.com/kb/25/031.html
http://www.nyu.edu/its/statistics/Docs/intracls.html
We have the following design: a limited number of raters (say 5) judge 16
subjects on a single characteristic (or more independent ones) twice.
I determine the ICC per rater using the above macro based on his two
judgments of the 16 subjects. This is the rater's ICC between his (pairs
of) judgments. I do this for each rater separately. It represents within
rater reliability / agreement.
I determine the overall ICC based on the average judgment of each pair of
judgments of a subject per rater. This is the ICC between raters. It
represents between rater reliability / agreement.
My question is which of the 6 ICCs I have to regard applicable for both
cases. My guess is nr. 3 for the first case and nr. 6 for the second one.
Do you agree?
An additional question is how to test these correlations for significance.
Can you point me to documents with formulas and/or SAS code? I didn't find
anything (yet).
Regards - Jim.
--
Jim Groeneveld, Netherlands
Statistician, SAS consultant
home.hccnet.nl/jim.groeneveld
