I performed a 2-way ANOVA with one between- and one-within subject factor in SPM5. For the between-subject factor, I set SPM to allow inhomogeneity of variance; for the within-subject factor, I allowed both inh of var as well as non-independence of the factor levels. I decided to compare that to the situation when I assume variance was homogeneous and factor levels were independent. The dof and F-statistic thresholds for a given p were identical. When I plotted one image against another, it wasn't true that "corrected" F- values were always lower than the "uncorrected" ones ("corrected" referring to the inh of var and dependence corrections). Is it possible?