ESTIMATING THE SUBJECT BY TREATMENT INTERACTION IN NON-REPLICATED CROSSOVER DIET STUDIES