Placement of stancode submitted via stanvars in the "if (!prior_only) { }" section

Interfaces brms
1

Hello,

In brms one can use the stanvar function to add code to a model.

For instance, if one specfies

dt = data.frame(x = rnorm(100), y = rnorm(100), z = rnorm(100))
dt$ysample(100,5)] = NA

bform = 
  bf(y ~ mi(x)) + 
  bf(x | mi() ~ z) + 
  set_rescor(FALSE)

my_stanvars = 
  stanvar(
    scode = "Yl_x[Jmi_x] = Ymi_x;",
    block = "model", position = "end")

make_stancode(
  bform,
  stanvars = my_stanvars,
  data = dt)

The the stancode Yl_x[Jmi_x] = Ymi_x; will be put at the very end of the model block:

model {
  if (!prior_only) {
    ...
    vector[N_x] Yl_x = Y_x;
    ...
    Yl_x[Jmi_x] = Ymi_x;
    ...
  }
  ...
  Yl_x[Jmi_x] = Ymi_x;
}

This leads to an “out of scope” error because Yl_x is used outside of if (!prior_only) {...} .

My question is if there is a way to use stanvar for the model block such that the stan code is written inside of if (!prior_only) {...}.

Thanks in advance!
Guido

So far I am just adding more stan code to re-create variables that were already created indside of if (!prior_only) {...} so that I can use them, but this is cumbersome and allows for errors.

The code above is only to illustrate the problem. The use case is a bit more complicated. In a nutshell, I want to evaluate the likelihood of y twice. Once for an outcome model with a restricted set of predictors (e.g. y \sim x) and once for an imputation model for a broader set of variables (e.g. y \sim x + z).

2

@wds15 provided the solution here.

One can set the block-flag to “likelihood” in the more recent brms version(s):

my_stanvars = 
  stanvar(
    scode = "Yl_x[Jmi_x] = Ymi_x;",
    block = "likelihood", position = "end")