Hi,
I have a 2 factors experiment with data neither normal nor
homogeneous. I am thus using the R package nparLD to perform an interaction model with non-parametric data. The interaction is significant and I wonder whether it would be correct to interpret it using regular Tukey/Bonferroni post-hoc tests?
Thanks in advance for your help,
G.
I imagine you cannot use Tukey HSD, etc. in this situation. I'm not really familiar with the nparLD package.
My recommendation is use an aligned ranks approach. If so, be sure to read the vignettes at the following pages about the correct way to conduct the post-hoc comparisons for an interaction term.
http://depts.washington.edu/madlab/proj/art/
https://cran.r-project.org/web/packages/ARTool/index.html
The package ART is easy to use but it does not allow to perform the analyses if there are missing data..any other idea?
Cannot remove the missing data or the structure of my df will not be consistent anymore :)
I guess the honest question is: what do you expect the analysis --- any analysis --- to do with the missing data ?
Perhaps an entirely different approach would be better, like a mixed effects model for an appropriate generalized linear model (gamma regression, and so on).
Dear Salvatore, this is exacty what I am finally doing, a mixed effect model where the missing data do not represent a problem, tks for your answers!
- Badges
- Science topic
- Similar topics
- Philosophy
- Esthetics
- Interpretation