I am trying to understand if I should (ought to) specify an informed prior with
R for a simple linear mixed effect model like:
DV ~ F1 * F2 + (1|Participant)
F2 are both two-level factors. I am only interested to test if one level of
F2 is significantly higher than another, either as the main effect only or in interaction (e.g., a significantly stronger/steeper difference). In effect, this would be a case of one-sided testing, which is, arguably, more “meaningful” to test in the Bayesian framework. For the rest of the model, I do not have any specific prior expectations.
How should I define my informative prior with
Have you used the get_prior command to see if the default priors make sense? Typically I’ll run with the default priors in brms if they make sense given my data and what I am trying to do. After that that I make a second model that has more informed priors.
Thanks Ara! Would you mind sharing a simple example with specified informed priors? Tnx
So we could look at a simplified version of your model. So like DV ~ F1 or DV ~ F2 whichever you think is more interesting. I’d see what priors brms sets for you. Then we can ask in what ways could either F1 or F2 impact (not the technical term :) ) DV.
Like in my line of work I would like at something like this wetland_plant_species01_productivty ~ depth_to_groundwater + (1|research_site) . So my prior for depth_to_groundwater would be something like normal(0.5, 0.5) since I expect that variable to have a slight impact on the productivity. And (1|research_site) would be a student_t of some sort, likely centered on zero (some research sites are similar) but allowing for some extra mass in the tails because extreme values could happen.