I understand the prior concept in Bayesian, which is cool, but their turning into code is too hard. As much as I know, publishing without the priors isn't a good practice. I have quite large dataset, n = 5000, thus very mildly informative priors would be more than fine. Can anyone help?
I have a hurdle_lognormal model as follows:
fit = brm(bf(received_treatment_hours ~ p1 + p2 + p3_fct + p4 + p5_fct + p6 + p7 + (1 | region), hu ~ p1 + p2 + p3_fct + p4 + p5_fct + p6 + p7 + (1 | region)), data = df, family = hurdle_lognormal(), cores = 3, chains = 3, prior = prior)
received_treatment_hours is a zero-inflated (50%) outcome variable, containing also extreme values (outliers). Ranging from 0-100, most patients received about 5 hours.
This is the prior summary
prior_summary(fit)
How to code a general b prior for lognormal and binomial part of the model? Let's assume that more >30 hours in received treatment hours are unlikely between different predictor levels. The proportion of zeros ranged quite a bit between different predictor levels (from 10-90%) - how to code this?
What can be defined with those sd, intercept and sigma priors?
Code I wrote myself for lognormal part of the model:
prior = c(prior(student(3, 0, 15), class=b), #allows extreme values and 2xSD = 2x15 = 30 hours
prior(student_t(3, 0, 15), class= sd, group = county)) #prior for hierarchical part of the model, allows extreme values and 2xSD = 2x15 = 30 hours
