Aside from using a bound on a fixed effect parameter (i.e ub=0), is there a relatively simple way of constraining a parameter to be negative in brms (maybe something analogous to a lognormal prior for a parameter that is known to be positive)? Or is there a way to put a prior on the negated parameter? It’s only one parameter in my model, so I’m trying to avoid using the somewhat convoluted solution shown in issue 86.
Not sure about the answer to your original question, but could you make it positive and then multiply the corresponding covariates by -1?
1 Like