I have a very general question about using the hypothesis() function or other Bayes Factor functions to compare two point hypotheses with regard to a brm model. Say I have
y ~ a + x + x : a
and two somewhat complex hypotheses, both of which contain point estimates:
hypothesis 1 is that a > 0 & x:a = 0
hypothesis 2 is that a > 0 & x:a - .5 * Intercept = 0
is it possible to use hypothesis() to obtain a Bayes factor for h1/h2? Is it possible at all? How might one go about it?
It is possible in general but not via hypothesis. In general, you can fit the two models corresponding to these hypothesis, by hard cording the restrictions into the model (and setting appropriate priors on the supported space) and then comparing the two models via bridgesampling. For your hypotheses, doing this in brms will likely not be possible and you need to use Stan directly.
I never replied to this, but thank you. We did follow your advice and implement the models in Stan directly (and that actually allowed us to circumvent the need for combined hypotheses tests altogether).