Howdy, Stranger!

It looks like you're new here. If you want to get involved, click one of these buttons!

Supported by

P(M|data) vs BF- repeated measures ANOVA Bayesian analysis

Hello,

I am interested to examine the effect of a three-way interaction with two within-participants variables (2 levels each) and one between-participants variable (2 levels).

I used JASP to examine it with a Bayesian analysis. If I understand correctly, to test the effect of the three-way interaction I need to divide the BF10 of the model with the interaction by the BF10 of the model without the interaction (i.e., the model that includes all the possible effects except for the three-way interaction). Alternatively, I can divide the equivalent B(M|data) indexes of these two models.

The problem is that the two ways give different results (BF10=2 ; P(M|data)=3), and now the question is- which one should I count on?

To complex things further, as a sanity check, I examined the same dataset with the R package "BayesFactor", and there I got a third result (BF10=2.78). Does anyone know what are the (meaningful) differences between the JASP and the BayesFactor algorithms?


Thanks a lot!

Lior

Comments

  • Hi Lior,

    Bayes factors are the ratio of P(Data|M) (the likelihood), not P(M|Data) (the posterior probability of the model).

    Hope that helps!

  • Also, if you know you need the main effects (because you are interested in the interaction) you can go to the "Model" tab and add them to the null model. This will simplify the presentation of the results.

    Cheers,

    E.J.

  • Thank you all!

Sign In or Register to comment.