G-priors in brms

Dear brms community,
I am trying to convey g-prior specification for the variance components in an hierarchical linear model (Y~A * B+(A * B|Sub)) with brms but I can’t find any guidance about. All I found is this thread and this this discussion on github. Does anyone have any insights about it?


What exactly would you like insight on? It looks like the problem was solved on the github page with the example provided being:

<define csq and V in R>
prior <- set_prior("multi_normal(0, sigma^2 * csq * V)")
stanvars <- stan_var(csq) + stan_var(V)
fit <- brm(y ~ x, data, prior = prior, stan_vars = stanvars)

It seems like all you need to do is define csq and V as objects in R (e.g., csq <- ...), specify the prior(s) as shown, and then declare csq and V as Stan variables that get passed to the brm(...) call

Yes well, I thought this solution is relevant only for multivariate models. Good to know I got it wrong then, thank you :)

I see. To be fair, I’ve never used a g-prior before, so I don’t know the intricacies here. At least in the example from github, it looks like the model is not multivariate (e.g., the formula is just y ~ x)

1 Like