JASP Bayes RM Anova - matched models

SamL

I'm very new to Bayes, and I am trying to run a Bayesian repeated-measures ANOVA to complement NHST analyses.

The RM ANOVA is 3x2, with Treatment (Control/Stress) X Visual Field (LVF/RVF) X Stimuli (Word/Non-Word).

The frequentist approach is revealing main effects of VF and Stimuli but no interaction effects:

When trying to find information about whether to select "Across all models" or "Across matched models" when computing the Bayesian RM ANOVA, I read that "Across matched models" is more similar to a frequentist RM ANOVA as it only compares to models with the same predictors.

When I run the Bayes RM ANOVA selecting "Across matched models", it appears there is substantial evidence in support of the interaction:

I'm surprised to see such a large BF (41) when it did not reach significance in the frequentist analysis (although the effect size appears large).

However, when I select the other option "Across all models", evidence in support of the interaction is anecdotal:

I guess I'm confused how the options can return such different results. I've read a few other posts querying this, but I'm still unclear what the best approach is.

So, my questions are:

1) Can somebody possibly provide a clearer definition explaining the difference between these two options?

2) Is one of these options more comparable with a frequentist RM ANOVA than the other?

Thanks in advance!

EJ

OK:

1. For the model comparison table, the results are easier to interpret with "best model on top"
2. When you look at the model comparison table, you see that we need the main effects, but it is not clear whether we need the interaction
3. When you select "matched models", models that have a specific interaction are compared to models that do not have that interaction. For the case at hand, the model with the 3-way interaction is compared to the models without it (but with the 2-way interactions). Somehow this model does very poorly (it is not even in the top-10, so the truncated table does not show it). This is somewhat peculiar and should not be the main take-away.
4. Overall, the data show support for the main effects, but "absence of evidence" for the interactions.

EJ

SamL

Hi EJ,

Thanks for your response.

As you are saying there is an absence of effects for the interactions, does this mean you're suggesting I should select the "Across all models" option?

The BFincl is 41.36 for "across matched models" option, and 1.47 for "across all models".

 I am still unclear if one of these options more comparable with a frequentist RM ANOVA than the other?

Thank you.