Hi everyone. I study Psychology and plan to do some Bayesian contrast analyses (e.g. a 2x2 interaction contrast within a larger 2x3 design).
I’ve read through some documentation of bayestestR and brms as well as some insightful forum posts. I’ve come across the workflow of using brms models in combination with emmeans and bayesfactor_parameters to calculate BayesFactors for specific contrasts. However, since (as far as I know) the priors one specifies in the brms function are applied to the model parameters, this means that for cases like testing a 2x2 interaction within a larger design, it becomes quite a handful to specify the prior in such a way that it applies correctly to the contrast of interest. (At least I am currently not sure, how I could approach this.) Additionally, there seems to be the need to use orthonormal factor coding in larger designs?
Anyway, after reading about these things and trying them out in R, I am now asking myself this: Wouldn’t it be a much easier approach to simply add the contrast of interest as a variable to the dataframe and then directly use it as a predictor in brms? Then one could use bayesfactor_parameters to calculate the BF for this parameter, right? It seems to me that this would both make it easier to specify the prior and fix the problem regarding factor coding in larger designs while producing the same result as the “traditional approach” using emmeans.
It seems to me like a rather failsafe alternative, however, I’ve not seen anyone employ or discuss this. Therefore, it would be great to get some input from some more experienced people. I’m looking forward to your responses.
Thanks a lot in advance!