Separate three-level models were run, with one level of cross-classification each, to see whether the intercepts and slopes would be drastically different from those in the final models reported in Table 4. The two analyses used the same model. The only difference was in the sample size and for the model predicting Grade 10 performance where the intervening level of cross-classification at Grade 8 was omitted. The results of these cross-classified models matched those of the final cross-classified models thus arriving at similar conclusions on the overall significance pattern of the predictors.