# Brms formula for correlational structure of distributional model

Hi all,
I’d like to estimate the effect of two predictors, `pred1` and `pred2` on mean and sigma of a normal distribution. In addition, I’d like to model:
a) the correlations between random-effect intercept and slopes (and between slopes), only for the mean
b) the correlation between random intercept for mean with random intercept for sigma
c) uncorrelated random intercept for sigma with random slopes for mean

That is, I’d like the formula to be something like this, which does not work because of the double specification of `(1|subject)`:

``````bf(response ~ (1|SID|subject) + (1+pred1+pred2|subject) + pred1 + pred2,
sigma ~ (1|SID|subject) + pred1 + pred2)
``````

And not this, because it doesn’t fulfil c) above:

``````bf(response ~ (1+pred1+pred2|SID|subject) + pred1 + pred2,
sigma ~ (1|SID|subject) + pred1 + pred2)
``````

Is this possible with brms?

Thank you!
João

• Operating System: Linux 5.3.0 (Ubuntu 18.04.1)
• brms Version: 2.12.0

Using the ID syntax, you can ensure that a whole block of random effects is modeled as correlated. What you cannot do is have a block in which some are correlated and some are not unless they divide in two subblocks in which all random effects in one block are either all correlated or all uncorrelated.