[R-sig-ME] planned comparisons-lmer
mengzhu@y@n @ending from vuw@@c@nz
Thu Sep 20 10:32:13 CEST 2018
I am running planned comparisons (using glht function in multcomp package) to test the differences between groups.
What I understand is that we have to apply corrections when we do more than one pairwise comparison, and it is not surprising that p-values will change (normally it becomes larger) when having more pairs.
I am interested in about 20 pairs of comparisons. Two hypotheses are being tested in the planned comparisons. Is it legitimate to have two sets of planned comparisons based on my hypotheses? My thought is no... But does anyone have an opinion on this?
I have tried keeping all comparisons in one set of comparisons using corrections like Tukey, repeated-t and Fisher's LSD Test, and Holm-Bonferroni. The p-values using LSD are the smallest followed by Tukey and Holm-Bonferroni. Reading from posts and discussions online, it seems none of post hoc test is best. But how should we justify ourselves when choosing a correction?
Have a nice day!😊
[[alternative HTML version deleted]]
More information about the R-sig-mixed-models