I am working with a data set containing nested observations (students in classes). The aim is to asses the effect of an intervention against a control condition. We have two classes in each condition at two time points (entry and exit).

Firstly, I want to assess the magnitude of the intra-class correlation to see whether variance stemming from level-2 units (classes) is sizable (compared to variance within classes).

For this purpose I would like to fit a null model (I believe it's called intercept-only model) considering only the difference between classes (no predictors of any other kind).

I have been using both, SPSS and RStudio software. It appears that I have fallen into the "dummy coding trap", showing singularity for the L2-predictor classes in the following model:

null.model