Different imputation results for mice.impute.rf based on operating system

[janickweberpals]

First of all thanks for this amazing package and associated documentation, I really learned so much about missing data!

I noticed that I get slightly different results when running imputations that use a random forest model depending the OS/platform. This is even when defining the same seed. I tested the following reprex with the same R (4.3.2) and mice package version (3.17.0) on two different platforms (Linux Ubuntu and MacOS) and I get two slightly different results:

# mice version 3.17.0
library(mice)
 
# impute data
imp <- mice(
  data = nhanes,
  m = 20,
  method = "rf",
  rfPackage = "ranger",
  seed = 1234,
  printFlag = FALSE
  )
 
# make long
imp_long <- complete(imp, action = "long")
 
# show bmi average
mean(imp_long$bmi)

Output on Ubuntu 20.04.6: 26.5236
Output on MacOS 15.2: 26.4744

I suspect it may have to do with the RNG invoked by the downstream random forest package.