Hi! I'm using Minitab to run ANOVAs for a recent psycholinguistic experiment. My problem is that some interactions aren't even close to significant in the overall model, but when I do Tukey comparisons, some conditions are significantly different (they don't share a letter). How can this be? Can I use these "significant" comparisons, and if so, what p-value do I report? Thanks!

P.S.: Sorry about the typo in the title. . . not sure how to change it.