I used the following model: model1= lmer(A~B*C+(1|block)) where A is a quantitative (continuous) variable and B and C are qualitative ones, with respectively 3 and 4 levels (A and B are crossed, 1 repetition of A-B combination treatments by block and 4 blocks).
anova(model1) gave significant results for A, B and the interaction A*B.
As post-hoc test, I would like to use pairwise.t.test (p.ajust.method="holm") in order to access which levels are significant. Is it the right method? If not, could anyone tell me why please?
Moreover, I read that people used lsmeans or ghlt. Does anyone knows what is the best between the 3 post-hoc tests and why? Thank you!
The decision between glht or lsmeans in your case is only a matter of convenience. Using one of them is considerably better than paired t-tests. For your paired t-tests you will either not use most of the data (for estimation of standard errors) or not be able to account for the combined effects of both factors.
Thank you Jochen for your answer, it helps a lot! So, do you mean that there is no difference in using glht and lsmeans ?
Ok, thank you!
In my lmer model, the 2 fixed factors and their interaction are significant. Then, when I performed a lsmeans as you adviced me, for ex. : lsmeans(model1,pairwise~C, ajust="tukey"). R gave me this note "Results may be misleading due to involvement in interactions". Does this mean that lsmeans does not consider the interaction in my model? Should I modify something?
- Badges
- Science topic