ICC - Between subject variability

Tophe

New Member
#1
Dear all,

I have a twofold question.
1) I'm doubting on how I calculated my between subject variability for a pre/post test design. What I did is dividing my subject mean square (MSb) by the grand mean. Is that correct?
2) I have normalized the data (for each subject, always same way fro normalizing the 2 tests). I get a higher ICC value with a higher between subject variability. Is this possible because isn't it the purpose of normalization to decrease variability between subjects?

Many thanks in advance for taking this request into consideration
 

Jake

Cookie Scientist
#2
1. No, it's not. For every subject, compute the average of their pre- and post-scores. Then compute the variance of these means. That's the between-subject variance.

2. What do you mean by "normalized"? Do you mean you applied some transformation to the raw data values so that their resulting distribution would be closer to normal? Do you mean z-scores? Do you mean something else?
 

Tophe

New Member
#3
Dear Jake, many thanks in advance for your fast answer. Regarding number 2, this a force measurement that I did. Therefore I divided all force measures by body-weight to allow comparison afterwards between several population.
 

Jake

Cookie Scientist
#4
I see. I think it is not surprising or unusual that the procedure you described would noticeably change the ICC. It's going to change both the between-subject variance and error variance by potentially quite a lot, so it seems difficult to say in advance what might happen to ICC. I see no reason why it should necessarily be lower.