Bayesian repeated measures ANOVA
Hello,
I am new to jasp and to bayesian analysis and after reading tutorials and blogs I am still not sure I am correctly implementing and reporting a bayesian repeated measure anova. I would very much appreciate your help and feedback. Two groups of participants (controls vs patients) performed the same experiment which had two conditions (A and B).
With the classical analysis I found a significant interaction between condition and group, but I would like to report also bayesian analysis alongside with the classical one.
In jasp I have selected bayesian rm anova and entered the two conditions in the repeated measures cells and group as between subject factor.
I have also selected Bayes factor: BF10; Order: Compare to best model; Effects: Across all models.
I get the following output, from which it seems like the model including the interaction term is the best model. Now, what is it more appropriate to report? I would like to report the bias factor for the interaction which I belive is 6.602 (listed in BFincl). Am I right?
Then, if I understood correctly, another useful information to report is how better the best model (i.e. the one including interaction) is explaining the data compared to the other model (i.e. the one without interaction).
To get this information I have added condition and group to the null model and looked at the BF incl.
Now I am reporting these analyses, together with the classical one, like this:
... There was a significant interaction between condition and group (F(xx)=xx , p=xxx, BF=6.602). Bayesian Repeated Measures ANOVA showed that there was positive evidence for including the interaction between condition and group as predictor of thresholds. Specifically, the observed data were approximately 2.2 times more likely under the model where thresholds were predicted by the interaction between group and condition, compared to the model where these were predicted by the two main factors only.
Is this way of reporting the results acceptable? In other words is it correct to use BFincl reported in the first picture as bayes factor for the interaction and to quantify how much including the interaction improves the prediction of the data compared to a model that do not include it? Is this latter information usually reported?
Thanks in advance for any help.
Best,
Comments
A few quick remarks:
1. "Specifically, the observed data were approximately 2.2 times more likely under the model where thresholds were predicted by the interaction between group and condition, compared to the model where these were predicted by the two main factors only."
Yes, but perhaps make sure to convey that what you are comparing is the model with the interaction *and the two main effects* versus the model with only the two main effects.
2. About the inclusion BF. First, when you report it, I would make it very clear that it is in fact the *inclusion* BF, so it is across all models. The reason why it is 6.6 is that there are several other models who are doing relatively poorly. The two-main-effect model is doing relatively well,. so if you are comparing it against that specifically, the BF will be less compelling.
E.J.