Interpreting Bayesian ANOVA

AKMneuro

If the null model outperforms all others, but the analysis of effects across matched samples provides evidence for the inclusion of an interaction or main effect, how should this be reported?
I would also love if anyone has published examples of how to appropriately write a results section using Bayesian ANOVA?
Thanks in advance
Andrew

sebastiaan

Hi Andrew,

If the null model outperforms all others, but the analysis of effects across matched samples provides evidence for the inclusion of an interaction or main effect, how should this be reported?

If there are no main effects, but only an interaction (i.e. a full cross-over interaction), then you can get the situation where the null model outperforms the model with the interaction, even though the interaction by itself is supported as indicated by the Inclusion BF based on matched models. This is because the model with the interaction also contains the main effects, and these can drag the model's performance down.

However, if the null model outperforms the model with only the main effect, then the inclusion BF should not favor the main effect either. Or at least that would be unusual.

So before answering your questions, could you post the model tables here so we can see what's going on exactly?

Cheers!
Sebastiaan