I have 4 measurements for each of my 1000 subjects, and each subject has a bunch of parameters (p1, p2, ... p100), for each of which I would like to calculate an ICC (version: (3,1) according to Shrout & Fleiss article).

I could do so easily when having 2 sessions, for each parameter individually. However, 2 of those 4 measurements were acquired on one day, and the other 2 at another day, roughly a year after that. Further, the measurement instrument was slightly different for the first and second measurement of each day, such that there is one measurement of the two with measurement instrument type A, and one with instrument type B, on each of the two days, respectively.

On top of this, some of my subjects are siblings (therefore I assume higher dependency), and subjects were measured at different recruiting sites (e.g. some at Clinic 1, some at Clinic 2, but each subject or family member was measured at the same clinic).

This data structure is quite complicated, and I have never worked with multilevel modeling. Could someone point me in the right direction?

Also, I have not found anything useful for calculating ICC in such structured data. If I use search functions, I end up with comments on how to use ICC to describe variance in clustered data, but not how to calculate an ICC on my parameters of interest (p1 to p100). Thanks a lot for any idea!

Best, Roman

  • $\begingroup$ ICC is a proportion of variance explained statistic. It's level 1 variance divided by the sum of level 1 and level 2 variance. I recommend Raudenbush & Bryk (2002) as a good intro text. $\endgroup$ – Jay Schyler Raadt Feb 5 at 20:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.