adding y | weights(weight) ~ ... to brm_formula

[yonisidi]

Currently I don't think it is possible to add y | weights(weight) ~ ... in the brm_formula as it is constructed right now. I tried adding it through ... but it seemed to be ignored. Would an explicit formal need to be defined?

Here is a discussion about weights for observations in normal brms formula syntax.

https://discourse.mc-stan.org/t/weights-in-brm/4278/14

[wlandau]

Yes, both brm_formula() and brm_data() would need new arguments and infrastructure for weights.

Is this a common enough use case to need direct support in brms.mmrm? If the limitation affects only a small number of cases, there is a quick workaround:

library(brms.mmrm)
data <- brm_data(
  data = brm_simulate_simple()$data,
  outcome = "response",
  group = "group",
  time = "time",
  patient = "patient",
  reference_group = "group_1",
  reference_time = "time_1"
)
formula <- brm_formula(data)
text <- paste(formula[[1]][[2]], "| brms::weights(weight)")
formula[[1]][[2]] <- parse(text = text)[[1]]
formula
#> response | brms::weights(weight) ~ group + group:time + time + unstr(time = time, gr = patient) 
#> sigma ~ 0 + time

^{Created on 2025-07-30 with reprex v2.1.1}

[wlandau]

I also wonder if it would be feasible to combine this with the existing ‘y | mi() ~’ syntax for multiple imputation during model fitting.

[yonisidi]

It is pretty common for us to use obs weights in the models.

could there be a cleaner solution for all the non-standard formula pieces that are embedded into brms, so we don't have to do one offs? it seems like there could be all kinds of things we wouldn't think people would need, but would get reqs for.

[U1004542]

Yes, both brm_formula() and brm_data() would need new arguments and infrastructure for weights.

Is this a common enough use case to need direct support in brms.mmrm? If the limitation affects only a small number of cases, there is a quick workaround:

library(brms.mmrm)
data <- brm_data(
data = brm_simulate_simple()$data,
outcome = "response",
group = "group",
time = "time",
patient = "patient",
reference_group = "group_1",
reference_time = "time_1"
)
formula <- brm_formula(data)
text <- paste(formula[[1]][[2]], "| brms::weights(weight)")
formula[[1]][[2]] <- parse(text = text)[[1]]
formula
#> response | brms::weights(weight) ~ group + group:time + time + unstr(time = time, gr = patient)
#> sigma ~ 0 + time
Created on 2025-07-30 with reprex v2.1.1

Thanks for the solution. I have tried implementing it. However, even if the "weight" variable is present in the dataset, brm_model function doesn't seem to recognize this weight. Of course, I can pass this through "stanvar" writing a new piece of stan code, but wanted to check if there is any other more efficient way. Thanks in advance for your help,

[wlandau]

could there be a cleaner solution for all the non-standard formula pieces that are embedded into brms, so we don't have to do one offs? it seems like there could be all kinds of things we wouldn't think people would need, but would get reqs for.

I'm not sure. Injecting arbitrary additional response information breaks the encapsulated brms_formula() interface. And to your point, I would like to avoid adding more one-offs because there are a lot of response modifiers. I started implementing #136 (now closed) before I realized there were so many.

I have tried implementing it. However, even if the "weight" variable is present in the dataset, brm_model function doesn't seem to recognize this weight.

I just learned that the brms function is actually brms::resp_weights(), which brms understands if you just write weights() without namespacing the call.