Graceful handling of empty cells (i.e., unbalanced designs)

[dgdillon]

I have a 2 (Memory status: new, old) x 2 (Accuracy: correct, incorrect) x 2 (Valence: negative, positive) recognition memory design. Occasionally, one (or more) of the cells in the design is empty; for instance, a participant may not have any incorrect responses to old, negative items. Accordingly, when I analyze these data in unfold (i.e., "y ~1 + cat(Memory Status) * cat(Accuracy) * cat(Valence)"), the model fails to converge. It would be great to have a tool that could (a) identify empty cells in the design and then (b) drop model terms involving those cells from the design matrix, so that the model could be estimated. For instance, for a subject with no incorrect responses to old negative items, the triple interaction should be dropped but the main effects and other interactions could still be modeled. I thought uf_imputeMissing might be able to handle situations like this but unfortunately that's not so. I can probably find a manual workaround but having a tool to do this more easily would be a great addition to an already outstanding toolbox.