Post hoc tests in Walrus (Robust ANOVA)

[xiaoli.yu]

Hi there,
My reviewer asked me to specify the post hoc test used in Walrus (Robust ANOVA). Which specific tests were embedded here? I can't seem to find this information. Does anyone have a clue about this and provide a reference?

Thank you very much!

[MAgojam]

Hi, @xiaoli.yu

If you used the Trim Method, the post hoc is with the mcp2atm() function of the WRS2 package:
Author: Patrick Mair [cre, aut], Rand Wilcox [aut], Indrajeet Patil [ctb]

You can take a look here (page 18):
https://cran.r-project.org/web/packages ... s/WRS2.pdf

You can see the function call in jamovi/walrus here:
https://github.com/jamovi/walrus/blob/m ... a.b.R#L245

Here, if you want to see the code of the function:
https://rdrr.io/rforge/WRS2/src/R/mcp2atm.R
or also here:
https://github.com/cran/WRS2/blob/master/R/mcp2atm.R

Cheers,
Maurizio

Tip you can trash:
You can submit references to the refere (personally, in various revisions, I accepted this).

[jorge_hidalgo]

Hello!

I have a similar doubt when interpreting the post hoc test in Walrus while using bootstrapped-based robust ANOVAs.

Specifically, my doubt comes when reading the post hoc tests for an interaction. Note that I am using two factors: Group (two levels) and Discourse task (four levels). Nonetheless, in the table produced for the post hoc comparison of the interaction group*discourse task, it is not specified what is being compared to what. In the table, I can just see a group column (that names one of the two groups), another column that enlists the four tasks, and then the corresponding columns to psi-hat, p, and the confidence intervals.

Now, I can see there is one significant interaction corresponding to the elements pointed out in the first row, but from observing the plotting of my data I know that there are no significant differences between the means obtained in that task between the two groups. Although I was expecting statistical differences in the fourth task (corresponding to the fourth row) as the differences between the means is evident, it is marked as the least significant interaction.

Hopefully you can help me out with this doubt. I also hope my inquiry is not confusing. I tried being very detailed but I can further provide details if necessary.

Thank you in advance!

[reason180]

I would also have doubts. For example, in a Walrus (bootstrap) robust ANOVA on the Tooth Growth Data, I think there should be 15 post-hoc comparison. But the output produces only 3.

Untitledhdgfe.png (26.24 KiB) Viewed 167 times

[MAgojam]

For example, in a Walrus (bootstrap) robust ANOVA on the Tooth Growth Data, I think there should be 15 post-hoc comparison.

Hey @reason180,
you're right, there should be 15 comparisons.

When I last updated Walrus I added for the "Robust ANOVA", the third table for the post-hoc interation of the factors.
This was achieved by bringing into the module some Rand Wilcox specific functions released in the Rallfun-v41.txt archive file.
Rallfun can be found here: https://osf.io/xhe8u/files/osfstorage?view_only=
The functions used only return the comparisons you see in the tables and I have left it to a possible future update of Walrus.

Cheers,
Maurizio
https://www.jamovi.org/about.html

[reason180]

Oh. OK. So it looks like the other value of supp (i.e., "VC") is excluded completely from the interaction post-hoc test. So probably best to just remove the "interaction" post-hoc test table since, for whatever reason, it doesn't actually work right.

[jorge_hidalgo]

Alright! I understand. Thank you both for your explanation @reason180 and @MAgojam. I will indeed omit that interaction and seek for alternatives.

Kind regards,
Jorge Hidalgo.