Hello all,

I’m trying to fit a distributional model in brms. To start, I just want to fit a model where the mean varies by group. So I simulated some simple data using the following code with fixed values:

Int <- c(-0.5, 0, 0.5)

B <- 0.01

N <- 75

Y <- rnorm(N, mean=0, sd=0.5)

X <- runif(N, min=4, max=6.5)

Res <- 0.01

Y <- c()

CG <- c()

for (i in 1:N) {

CG[i] <- sample(Int, 1)

Y[i] <- B * X[i] + CG[i] + rnorm(1, 0, Res)

}

testdf <- data.frame(Y = Y, X = X, CG = CG)

Then I am trying to fit a simple model as shown here:

bprior <- prior(student_t(5, 0, 5), class = b) +

prior(normal(0, 0.5), class = Intercept) +

prior(cauchy(0, 2.5), class = sd)

fit <- brm( Y ~ X + ( 1 | CG ), data = testdf, prior = bprior, family = gaussian(), control = list(adapt_delta = 0.85, max_treedepth = 10) )

I thought this model matched the generating simulation, but I am getting persistent warnings about post-warmup divergences (no matter how I tweak the adapt_delta). Using brms’ default priors also yields the persistent divergences. Thus I assume the model I specified *does not* match the simulation above, and so I was hoping for some advice on what’s awry.

Thank you,

–Jon