as.mids() changes derived logical variables after complete(..., "long") round-trip

[irenerlover]

I may have found an issue with as.mids() when converting edited long-format imputed data back into a mids object.

I imputed two binary variables first, then derived two additional logical variables from the completed long data. After converting the long data back using as.mids(), some derived values appeared to change unexpectedly. In particular, rows with remission_6m_upcr == TRUE and remission_6m_egfr == TRUE should give remission_6m_pr == FALSE, but after the round-trip I observed cases where remission_6m_pr became TRUE.

Reproducible example

library(mice)
library(dplyr)
library(tibble)

set.seed(123)

# Example raw data: only first 2 variables are imputed
dat0 <- tibble(
  id = 1:20,
  remission_6m_upcr = sample(c(TRUE, FALSE, NA), 20, replace = TRUE,
                             prob = c(0.35, 0.45, 0.20)),
  remission_6m_egfr = sample(c(TRUE, FALSE, NA), 20, replace = TRUE,
                             prob = c(0.35, 0.45, 0.20))
)

# Convert logical to 0/1 for mice
dat_imp <- dat0 %>%
  mutate(
    remission_6m_upcr = as.integer(remission_6m_upcr),
    remission_6m_egfr = as.integer(remission_6m_egfr)
  )

# Run mice
ini  <- mice(dat_imp, maxit = 0, printFlag = FALSE)
meth <- ini$method
pred <- ini$predictorMatrix

meth["id"] <- ""
pred[, "id"] <- 0
pred["id", ] <- 0

meth["remission_6m_upcr"] <- "logreg"
meth["remission_6m_egfr"] <- "logreg"

imp <- mice(
  dat_imp,
  m = 5,
  method = meth,
  predictorMatrix = pred,
  seed = 2026,
  printFlag = FALSE
)

# Path A: derive variables after complete(long)
long_A <- complete(imp, action = "long", include = TRUE) %>%
  arrange(.imp, .id) %>%
  mutate(
    remission_6m_upcr = as.logical(remission_6m_upcr),
    remission_6m_egfr = as.logical(remission_6m_egfr),
    remission_6m = case_when(
      is.na(remission_6m_upcr) | is.na(remission_6m_egfr) ~ NA,
      remission_6m_upcr == TRUE & remission_6m_egfr == TRUE ~ TRUE,
      TRUE ~ FALSE
    ),
    remission_6m_pr = case_when(
      is.na(remission_6m_upcr) & is.na(remission_6m_egfr) ~ NA,
      is.na(remission_6m_upcr) & remission_6m_egfr == TRUE ~ TRUE,
      remission_6m_upcr == TRUE & is.na(remission_6m_egfr) ~ TRUE,
      remission_6m_upcr == TRUE & remission_6m_egfr == FALSE ~ TRUE,
      remission_6m_upcr == FALSE & remission_6m_egfr == TRUE ~ TRUE,
      remission_6m_upcr == TRUE & remission_6m_egfr == TRUE ~ FALSE,
      TRUE ~ FALSE
    )
  )

# Convert back to mids
imp_back <- as.mids(long_A)

# Extract again
long_B <- complete(imp_back, action = "long", include = TRUE) %>%
  arrange(.imp, .id)

# Compare
compare_A <- long_A %>%
  select(.imp, .id, id,
         remission_6m_upcr, remission_6m_egfr,
         remission_6m, remission_6m_pr) %>%
  arrange(.imp, .id)

compare_B <- long_B %>%
  select(.imp, .id, id,
         remission_6m_upcr, remission_6m_egfr,
         remission_6m, remission_6m_pr) %>%
  arrange(.imp, .id)

all.equal(compare_A, compare_B)

compare_A %>%
  filter(remission_6m_upcr == TRUE, remission_6m_egfr == TRUE) %>%
  select(.imp, .id, id,
         remission_6m_upcr, remission_6m_egfr,
         remission_6m, remission_6m_pr)

compare_B %>%
  filter(remission_6m_upcr == TRUE, remission_6m_egfr == TRUE) %>%
  select(.imp, .id, id,
         remission_6m_upcr, remission_6m_egfr,
         remission_6m, remission_6m_pr)

Outcome:

In my actual data, some values of remission_6m_pr appeared to switch after converting back with as.mids().

Session info:

> sessionInfo()
R version 4.5.2 (2025-10-31 ucrt)
Platform: x86_64-w64-mingw32/x64
Running under: Windows 11 x64 (build 26200)

Matrix products: default
  LAPACK version 3.12.1

locale:
[1] LC_COLLATE=Chinese (Traditional)_Hong Kong SAR.utf8  LC_CTYPE=Chinese (Traditional)_Hong Kong SAR.utf8   
[3] LC_MONETARY=Chinese (Traditional)_Hong Kong SAR.utf8 LC_NUMERIC=C                                        
[5] LC_TIME=Chinese (Traditional)_Hong Kong SAR.utf8    

time zone: Asia/Hong_Kong
tzcode source: internal

attached base packages:
[1] stats     graphics  grDevices utils     datasets  methods   base     

other attached packages:
[1] tibble_3.3.1 dplyr_1.2.0  mice_3.19.0 

loaded via a namespace (and not attached):
 [1] Matrix_1.7-4      mitml_0.4-5       glmnet_4.1-10     compiler_4.5.2    rpart_4.1.24      tidyselect_1.2.1 
 [7] Rcpp_1.1.1        tidyr_1.3.2       splines_4.5.2     boot_1.3-32       lattice_0.22-7    R6_2.6.1         
[13] generics_0.1.4    shape_1.4.6.1     pan_1.9           rbibutils_2.4.1   iterators_1.0.14  MASS_7.3-65      
[19] backports_1.5.0   nloptr_2.2.1      nnet_7.3-20       minqa_1.2.8       pillar_1.11.1     rlang_1.1.7      
[25] broom_1.0.12      cli_3.6.5         withr_3.0.2       magrittr_2.0.4    Rdpack_2.6.5      jomo_2.7-6       
[31] foreach_1.5.2     grid_4.5.2        rstudioapi_0.18.0 lme4_1.1-38       nlme_3.1-168      lifecycle_1.0.5  
[37] reformulas_0.4.4  vctrs_0.7.1       glue_1.8.0        codetools_0.2-20  survival_3.8-3    purrr_1.2.1      
[43] tools_4.5.2       pkgconfig_2.0.3  

Why is it the case? Thank you!!

Best,
Irene

[stefvanbuuren]

Thanks for the detailed reprex, @irenerlover.

This is not a bug in mice — it is a consequence of how as.mids() reconstructs the mids object.

Two things worth noting first

1. Type mismatch warning

You will see:

    Warning: Type mismatch for variable(s): remission_6m_upcr
    Imputation method logreg is for categorical data.

logreg expects a factor, but remission_6m_upcr and remission_6m_egfr are stored as integer. This works in practice but is fragile. The cleaner approach is to store them as factors from the start:

    dat_imp <- dat0 %>%
      mutate(
        remission_6m_upcr = factor(remission_6m_upcr),
        remission_6m_egfr = factor(remission_6m_egfr)
      )

2. The "13 element mismatches"

[1] "Component \"remission_6m_pr\": 13 element mismatches"

This is the symptom you observed. The cause is explained below.

Why it happens

When as.mids() processes long_A it separates the data into two parts:

• original data (.imp == 0): stored in mids$data
• imputations (.imp >= 1): stored in mids$imp, but only for variables that had missing values in the original

remission_6m_pr has no missing values in the .imp == 0 rows — its values there are all FALSE because remission_6m_upcr and remission_6m_egfr are still NA at that point, and your case_when() falls through to TRUE ~ FALSE. So as.mids() stores no imputations for remission_6m_pr. When complete() reconstructs each imputed dataset it pulls remission_6m_pr from mids$data (the original row), ignoring the values you computed in long_A.

The fix

Compute derived variables after the round-trip, not before:

# Step 1: convert types only — do NOT derive new variables yet
long_A <- complete(imp, action = "long", include = TRUE) %>%
  arrange(.imp, .id) %>%
  mutate(
    remission_6m_upcr = as.logical(remission_6m_upcr),
    remission_6m_egfr = as.logical(remission_6m_egfr)
  )

# Step 2: round-trip
imp_back <- as.mids(long_A)

# Step 3: derive AFTER extracting completed datasets
long_B <- complete(imp_back, action = "long", include = TRUE) %>%
  arrange(.imp, .id) %>%
  mutate(
    remission_6m = case_when(
      is.na(remission_6m_upcr) | is.na(remission_6m_egfr) ~ NA,
      remission_6m_upcr & remission_6m_egfr ~ TRUE,
      TRUE ~ FALSE
    ),
    remission_6m_pr = case_when(
      is.na(remission_6m_upcr) & is.na(remission_6m_egfr) ~ NA,
      remission_6m_upcr & remission_6m_egfr ~ FALSE,
      TRUE ~ FALSE
    )
  )

A simpler alternative

If the only reason for the as.mids() round-trip is to derive these variables, you can skip it entirely and apply the derivation directly to long_A. The round-trip is only needed when you want to use mice pooling functions (with(), pool()) on models that include the new variables.

[irenerlover]

Thanks for the stefvanbuuren detailed answer. I tried to explain what happened in my case in my own words: This seems to be a special case (.imp = 0: NA + TRUE => TRUE; for .imp = 1, NA is filled with TRUE, but the TRUE on the right-hand side does not change at the same time because of the way as.mids() reconstructs the data:'pulls remission_6m_pr from mids$data' ), which appears to trigger this behavior.