Gang Wrote:
-------------------------------------------------------

>
> > I'm still confused. Aren't we interested in
> reliability though? So, if we are looking for
> > reliability, wouldn't that suggest we'd be
> interested in low ICC values? Is this correct?
> > I was under the impression that high ICC values
> indicate strength of reliability, reproducibility,
> no?
>
> My understanding about the notion of
> reliability/reproducibility is this: the ICC for
> session is also the correlation coefficient
> between any two measures (from two different
> subjects, for example) coming from the same
> session. And it has nothing to do with the
> difference between the two sessions, which is
> measured by the fixed effect.

Gang, I revisited 3dICC_REML recently and I have some more questions:

We input regression coefficients into 3dICC_REML

Here is my setup:

Output:ICC_output
MASK: mask_group+tlrc.
Clusters:5
Subj Session InputFile
01 one sess-01-a+tlrc
02 one sess-02-a+tlrc
03 one sess-03-a+tlrc
,
,
,
01 two sess-01-b+tlrc
02 two sess-02-b+tlrc
03 two sess-03-b+tlrc
.
.
.
01 three sess-01-c+tlrc
02 three sess-02-c+tlrc
03 three sess-03-c+tlrc
.
.
.

In terms of the ICC values, we were initially looking at the Session output. We wanted to see what activated clusters appeared with respect to session. We assumed that signal amplitude consistent across sessions would represent high reliability and a fairly high ICC value. We calculated voxel-wise ICC values using the ANOVA framework with absolute agreement (Shrout and Fleiss 1979, McGraw and Wong 1996). The ICC maps we generated from the ANOVA framework are strikingly similar to the ICC maps from 3dICC_REML for subjects...not session. Which would make sense based on the definition of ICC, no? Variability driven mostly by the between-subjects variability would reflect a high ICC value. That would also mean low amount of variability contributed by session, indicating similar amplitude values across 3 sessions? The lack of variation across session would indicate the majority of the variability is coming from noise or between-subjects. Am I understanding this correctly?

Thanks,

Michael



Edited 1 time(s). Last edit at 07/31/2013 12:11PM by Gang.