I’m preparing a tutorial script for a random slopes model with a gaussian outcome. With a binary x variable, the brm code is fairly straightforward as far as such models go:

`model <- brm ( y ~ x + (1+x|group_id), data = dat, chains = 2, cores = 2, iter = 3000, warmup = 1000)`

One objective for users of the tutorial is to use the respective fixed effects and the random effects (both intercepts and slopes, when necessary) to calculate predictions on the response scale. My inclination is to have them use fixef() and ranef() to extract the posterior means that would allow them to do that.

However, I’m seeing that brms has a coef() function that purports to do all of those calculations: https://rdrr.io/cran/brms/man/coef.brmsfit.html

When I compare the extractions from coef() with what I calculate from fixef() and ranef(), everything looks mostly identical. There are discrepancies, though, mostly starting a few digits to the right of the decimal. My inclination is to attribute this to rounding error, but before recommending the use of coef(), I figured that it would be ideal to confirm its functionality.

Can someone with more knowledge of brms confirm what happens under the hood with coef()? Thanks.

- Operating System: Mac OS
- brms Version: 2.0.1