Brm_multiple warning with fitted()

Interfaces brms
1

I am using mice and brm_multiple for multiple imputation of a predictor variable in a simple multilevel regression model. After the model has run and converged well, I am then using fitted() and predict() on this model with an argument to “newdata” specifying the level of predictions. When doing so I receive the warning “Using only the first imputed data set. Please interpret the results with caution until a more principled approach has been implemented.”

I am surprised by this warning given that I am supplying an argument to “newdata”. I would expect the warning if that were not the case and fitted values were being generated at the level of the original predictors, but am wondering if in this case it doesn’t actually apply. But perhaps the warning is alluding to something about how the multiple models are being used (the “combine” parameter in brm_multiple is set to TRUE).

  • Operating System: MacOS 10.13.6
  • brms Version: 2.10.0
2

Very likely this warning is nothing to worry about but I would still appreciate a minimal reproducible example for check it out.

3

Hi Paul, thanks for the help. Here’s a MWE:

library(brms)
library(mice)
data(iris)

add NA’s in predictor (10% of sample)

iris$Petal.Length[sample(1:nrow(iris), size = 0.1*nrow(iris), replace = F)] <- NA
iris_imp <- mice(iris, m=3, print=F)

mod <- brm_multiple(Sepal.Length~Petal.Length + (1|Species), data = iris_imp, chains=4, iter=2000)
pdat <- data.frame(Sepal.Length = seq(min(iris$Sepal.Length), max(iris$Sepal.Length), length.out = 50), Petal.Length = seq(min(iris$Petal.Length, na.rm=T), max(iris$Petal.Length, na.rm=T), length.out = 50))

line that produces error

fitted(mod, newdata=pdat, re_formula=NA, scale=“response”)

4

The warning was indeed a false positive and should no longer appear when newdata is passed in the dev version of brms from github.

5

Great, thanks for confirming!