Restrictions for General Linear Models
main topic  

If your general linear model includes nested factors or random factors, or if it uses binary coding, the following are restrictions on the procedures that you can perform.

·    If your model contains random factors, you must specify a hierarchical model in the Model dialog.

·    Stepwise procedures are not available for models that contain random factors.

·    Comparisons are not available for terms that contain or interact with random factors. Nesting is a form of interaction. Suppose the model contains A B C A*B, where both A and C are random, and B is fixed. Because A*B is a term in the model, no multiple comparison results for B are available even though B is a fixed factor. However, if the model does not include A*B, multiple comparison results are available for B.

There is a special case for a balanced design that has two factors. Suppose the model contains A B A*B, where A is random, and B is fixed. Even though A*B is in the model, you can perform multiple comparisons for B.

·    Comparisons is disabled if you choose (1, 0) coding and the model is non-hierarchical. To enable Comparisons for this case, choose (-1, 0, +1) coding or specify a hierarchical model.

·    Factorial plots cannot display nested factors.

·    Predict cannot calculate confidence intervals when the model contains random factors.

·    Predict is not available for models that contain nested factors.

·    Response Optimizer is not available for models that contain nested factors.