Dear FSL users, I have a question about ANOVA analysis that I hope some of you can help me with. I have data from a 3-factor design. Each factor has 2 levels. I have run the first level analyses with 8 EVs, each defining one condition ("cell" in the 2x2x2 paradigm): A1 B1 C1 A1 B1 C2 A1 B2 C1 A1 B2 C2 A2 B1 C1 A2 B1 C2 A2 B2 C1 A2 B2 C2 I have also run a higher-level standard GLM analysis to calculate simple effects, and found an interesting difference between the two contrasts (A1B1C1 > A1B2C1) and (A2B1C1 > A2B2C1). A referee suggested I test if this difference is statistically significant. So, is the effect (B1C1>B2C1) significantly different between A1 and A2? To do this, I guess I should run a 3-factor ANOVA as described on the support site. The problem (?) is: The third factor, C, is defined by the subjects´ responses (task accuracy). Two of my conditions, A2B1C2 and A2B2C2 were very easy and therefore contain very few trials - in several subjects no trials at all. These empty conditions are not of particular interest to me, since I mainly want to compare B1C1>B2C1 across the two levels of A. But I will have to include all conditions in my factor analysis. So my question is: Is it still possible to run an ANOVA? If so, how do I deal with the empty factor levels? I can’t see how ANOVA would produce any meaningful results in this case, but maybe some of you can. Or perhaps you know of an alternative way to test the interaction effect I described above? Any suggestions are welcome! Kind regards, Hanne