A nice cautionary discussion of how to interpret the results of multiple-way ANOVAs that yield unexpected interactions.
Bishop Blog