To measure emotion differentiation, which refers to the precision of an individual to accurately describe and distinguish between different emotional states, many studies have used a between-person index based on the Intra Class Correlation (ICC) across measurements. Often based on ESM data (which includes many measurements per participant) an ICC between different emotion items is calculated. A high ICC means low differentiation, a low ICC indicates high differentiation.
My problem with this measure is that you aggregate all the information from the within level to the between level. You lose all information on the within level, which might be interesting variance. Especially when you are interested in situational predictors.
To create a proxy for emotion differentiation I thought it would also make sense to simply calculate the variance of the different emotion items for each occasion. High variance would then imply more differentiation in the moment or occasion, whereas low variance would imply little differentiation.
In a multi-level framework the intercept for each person on a between level would represent the mean of each of these variances. As a kind of validity cheack, I correlated the ICC measure with the average variance of each person and found in 3 different simulated/example datasets a correlation between -.35 and .45. Which is way lower than I expected.
Can anyone explain this correlation? Is it a conceptual logical correlation? Both measures in my opinion should give an indication of differentiation of scores. Therefore I am a bit flustered by this...
Please correct me in any of my arguments above, and if you have ideas I would love to hear them!