0

I am a newbie here and my question is whether I should use a parametric or non-parametric post-hoc test based on the results from a generalized linear model; and if non-parametric is appropriate, how to conduct it.

Because this is about concepts, I'm not posting any data (unless it's necessary). I fitted generalized linear mixed-effect models with negative binomial function in Rstudio with lme4 package (glmer.nb model).

I understand that the GLM is for non-parametric data, but if I want to run a follow-up post-hoc test based on the model results, do I use parametric or non-parametric test?

I am able to conduct Tukey-adjusted parametric test of the model results using emmeans package (emmeans command) and multcomp package (cld command) to analyse significant differences between treatments. I.e. emmeans(modelX, "treatments").

However, if this is not appropriate and non-parametric test should be conducted instead, can someone enlighten me on how to do it using the model results and/or accounting for the random effects?

Thank you in advance!

zx8754
  • 52,746
  • 12
  • 114
  • 209
Hamham Rhapsody
  • 1
  • 3
  • Since your question focuses on statistical concepts rather than programming, it would be more appropriate to pose your question over at Cross Validated. See their list of on-topic subjects [here](https://stats.stackexchange.com/help/on-topic). – Z.Lin Mar 21 '19 at 02:48
  • Okay thank you, will do so next time! I just checked and figure out that since I already assumed a type of distribution with the model, I can stick with parametric post-hoc test. – Hamham Rhapsody Mar 21 '19 at 14:54

0 Answers0