Date:         Mon, 5 Oct 2009 06:36:23 -0700
Reply-To:     "cat.." <cat.b41@GMAIL.COM>
Sender:       "SAS(r) Discussion" <SAS-L@LISTSERV.UGA.EDU>
From:         "cat.." <cat.b41@GMAIL.COM>
Organization: http://groups.google.com
Subject:      Appropriate model for intra-class correlation coefficient
              calculation ?
Comments: To: sas-l@uga.edu
Content-Type: text/plain; charset=ISO-8859-1

Hi folks !

In a pediatric clinical trial, I have 2 scales that measure the same
quantity (let's call it Y) on each child in the trial: One based on a
questionnaire filled in by the child him/herself, the second one based
on a questionnaire filled in by his/her parent.

I need to calculate the Intra-class correlation coefficient between
these two scores (parent-rated versus patient-rated) to measure their
agreement.

That papers gives several formula, each being based on an ANOVA model
that fits the score variability on several sources of variability.

http://www3.uta.edu/faculty/ricard/COED/McGraw%20(1996)%20Forming%20inferences%20about%20ICCs.pdf

The paper gives a speific ICC formula (with 95% CIs) for each of the
models beneath:

1 - Score = overall mean + random patient effect + error
2 - Score = overall mean + random patient effect + random rater effect
+ interaction patient*rater + error
2a- Score = overall mean + random patient effect + random rater effect
+ error
3 - Score = overall mean + random patient effect + fixed rater effect
+ interaction patient*rater + error
3a - Score = overall mean + random patient effect + fixed rater effect
+ error.

I'm not sure what the model is that best fits my data and would like
your thought about it.

In fact, I'd go for the 3a one but the problem I have is that the
rater and the patient are not independant as the rater is either the
patient him/herself, or her/his parent. So, none of these models would
be valid.

Thanks for your input, especially if you have experience in this.

Cheers,

Catherine.