mice() changes the predictorMatrix

[markseeto]

Thanks for the mice package. I'm using version 3.6.0.

The example below shows mice() changing the predictorMatrix. It appears to be related to the similarity of the variable names, because it doesn't happen if one of the variable names is changed.

Am I misunderstanding something? Thanks.

library(mice)

set.seed(1)

# Repeated measurements of 'y' on each subject.
# 'score' and 'score.previous' are constant within each subject.

d <- data.frame(subject = 1:10,
                score = rnorm(10),
                score.previous = rnorm(10))
d[1:2, "score"] <- NA

DataA <- d[rep(1:10, each = 3), ]  # 3 observations per subject

DataA$y <- rnorm(nrow(DataA))
DataA[4:5, "y"] <- NA

iniA <- mice(DataA, maxit=0)

methA <- iniA$method
methA["score"] <- "2lonly.pmm"
methA["y"] <- "2l.pan"

predMatA <- iniA$predictorMatrix
predMatA[c("score", "score.previous", "y"), "subject"] <- -2

predMatA

##                subject score score.previous y
## subject              0     1              1 1
## score               -2     0              1 1
## score.previous      -2     1              0 1
## y                   -2     1              1 0

miceA <- mice(data = DataA, m = 5, maxit = 10,
              method = methA, predictorMatrix = predMatA)

miceA$predictorMatrix

##                subject score score.previous y
## subject              0     1              1 1
## score               -2     0              0 1
## score.previous      -2     1              0 1
## y                   -2     1              1 0

# Note that miceA$predictorMatrix["score", "score.previous"] is 0.

# Next, use the same data but change the name 'score.previous' to 'sp'.

DataB <- DataA

names(DataB)[names(DataB) == "score.previous"] <- "sp"

predMatB <- predMatA
rownames(predMatB) <- colnames(predMatB) <- names(DataB)

methB <- methA

names(methB) <- names(DataB)

miceB <- mice(data = DataB, m = 5, maxit = 10,
              method = methB, predictorMatrix = predMatB)

miceB$predictorMatrix

##         subject score sp y
## subject       0     1  1 1
## score        -2     0  1 1
## sp           -2     1  0 1
## y            -2     1  1 0

# miceB$predictorMatrix["score", "sp"] is 1.

[stefvanbuuren]

Thanks. That wasn't right indeed.

Now fixed.

[markseeto]

Thanks Stef.