Use a F-test to compare the model with interaction terms and the model without. If the p-value associated with the F-test is not significant, the interaction term is not relevant and should be excluded. Planned comparisons can test for marginal effects (holding the other factor at its average value) or combined effects (which implies an interaction). Make sure the test meets the assumptions of planned comparisons. You can construct the test by hand, however R has a function in the `lsmeans` package. ```R library(lsmeans) pairs(lsmeans(lmod, "<factor_as_string>"), adjust="none") ``` Ignore the tests that weren't planned for or adjust the p-values to account for multiple comparisons. To adjust, provide an argument to the `adjust` parameter. ```R pairs(lsmeans(lmod, "<factor_as_strings"), adjust="bonferroni") ``` Or use the `TukeyHSD` function. ```R TukeyHSD(aov(lmod), which="<factor_as_string>") ```