I have conducted an experiment where participants listened to 4 conversations (t1, t2, t3, t4) in 3 versions (S, M, L). They heard them in three listening blocks with all 4 conversations present in each block in one randomly selected version. After hearing the conversation participants left a rating on a scale from 0 to 6.
My hypothesis is that a personality trait assessed by a questionnaire (EPT) can serve as a predictor to how well a participant distinguishes between S, M and L versions. Means for the ratings of the S (3.098), M (3.010), L (2.832) versions are quite close to each other, but the difference is statistically significant.
What would be the best way to construct differentiation index / score to test my hypothesis? In ideal situation (S rated as 6, M rated as 3, L rated as 0) I'd like the differentiation index to show that S, M and L are maximally spaced out.
I tried an index that would just show the difference between S and L version, binary index that would show that the participants differentiated between two different versions of the same dialogue. No correlation with EPT score observed using those measures as a differentiation index.