I am using the lsmeans package for "post hoc" multiple comparisons and I read here: ANOVA - Do we need a global test before post hoc tests? that in general post hoc tests are valid even if the ANOVA result is not significant. The only exception is the protected Fisher Least Significant Difference (LSD) test. Would that be the case when using lsmeans without p-value adjustment method? For example:

lsm <- lsmeans(Model,  ~ Factor1)
cld(lsm, type = "response", sort=FALSE, Letters=c("abcdefg"), adjust="None")

And models are glm-Models using binomial, negative binomial or poisson(or gaussian) distribution.

  • 1
    $\begingroup$ The unadjusted comparisons if all means are equivalent to using the LSD in standard anova situations, and very similar in generalized linear models. So you would NOT want to do such unadjusted comparisons without at least having a significant anova F, and even then it is only weak protection of the error rate. $\endgroup$ – Russ Lenth Mar 31 '16 at 2:35
  • 1
    $\begingroup$ But you could do the Tukey adjustment and it'd be just fine w/o doing anova, provided all the assumptions hold. $\endgroup$ – Russ Lenth Mar 31 '16 at 2:37
  • $\begingroup$ Thank you very much for the answer! That was very helpful! $\endgroup$ – Pharcyde Mar 31 '16 at 16:09

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.