合并估算和非估算数据

Question

I have a question about merging datasets after multiple imputation. I have created an example to explain my problem:

id <- c(1,2,3,4,5,6,7,8,9,10)
age <- c(60,NA,90,55,60,61,77,67,88,90)
bmi <- c(30,NA,NA,23,24,NA,27,23,26,21)
time <- c(62,88,85,NA,68,62,89,62,70,99)
dat <- data.frame(id, age, bmi, time)
dat

id <- c(1,2,3,4,5,6,7,8,9,10)
m1 <- c(60,78,90,55,60,61,77,67,88,90)
m2 <- c(30,44,35,23,24,22,27,23,26,21)
m3 <- c(62,88,85,78,68,62,89,62,70,99)
dat2 <- data.frame(id, m1, m2, m3)
dat2

I have two datasets, dat and dat2. The dataset dat contains missing variables, so I use multiple imputation to impute this dataset (package MICE):

library(mice)
impdat <- mice(dat, maxit = 0)
methdat <- impdat$method
preddat <- impdat$predictorMatrix
preddat["id",] <- 0
preddat[,"id"] <- 0
impdat <- mice(dat, method = methdat, predictorMatrix = preddat, seed =         
2018, maxit = 10, m = 5)

Now I want to merge the imputed dataset impdat with the dataset dat2. But that is were my problem arises. I tried the following:

completedat <- complete(impdat, include = T, action = 'long')
finaldat <- merge(completedat, dat2, by = "id")

finaldat <- as.mids(finaldat)
  Error in `[<-.data.frame`(`*tmp*`, j, value = c(61, 88)) : replacement has 2 rows, data has 1  

However, this gives me an error message. The merging is successful, because the dataframe completedat is what I want. The problem is that I cannot transform it back to a mids object.

I know I can add the variables from dat2 one by one. That does work:

completedat <- complete(impdat, include = T, action = 'long')
completedat$m1 <- dat2$m1
finaldat2 <- as.mids(completedat)

In this example, this is okay, because dat2 only has 4 variables. In my real data, I have approximately 200 variables that I want to add to my multiple imputed dataset, so I hope there is an easier way to add all those variables to my imputed dataset. Can somebody help me?

Answer 1

Wouldn't cbind work provided that you want to combine imputed and non-imputed data?

id <- c(1,2,3,4,5,6,7,8,9,10)
age <- c(60,NA,90,55,60,61,77,67,88,90)
bmi <- c(30,NA,NA,23,24,NA,27,23,26,21)
time <- c(62,88,85,NA,68,62,89,62,70,99)
dat <- data.frame(id, age, bmi, time)
dat

id <- c(1,2,3,4,5,6,7,8,9,10)
m1 <- c(60,78,90,55,60,61,77,67,88,90)
m2 <- c(30,44,35,23,24,22,27,23,26,21)
m3 <- c(62,88,85,78,68,62,89,62,70,99)
dat2 <- data.frame(id, m1, m2, m3)
dat2

# install.packages("mice")
library(mice)
impdat <- mice(dat, 
               seed = 2018, 
               maxit = 10, 
               m = 5)
impdat
# Class: mids
# Number of multiple imputations:  5 
# Imputation methods:
#   id   age   bmi  time 
# "" "pmm" "pmm" "pmm" 
# PredictorMatrix:
#   id age bmi time
# id    0   1   1    1
# age   1   0   1    1
# bmi   1   1   0    1
# time  1   1   1    0

impdat = complete(impdat)
impdat

# id age bmi time
# 1   1  60  30   62
# 2   2  60  24   88
# 3   3  90  24   85
# 4   4  55  23   89
# 5   5  60  24   68
# 6   6  61  24   62
# 7   7  77  27   89
# 8   8  67  23   62
# 9   9  88  26   70
# 10 10  90  21   99

final_data = cbind(impdat, dat2)
final_data
# id age bmi time id m1 m2 m3
# 1   1  60  30   62  1 60 30 62
# 2   2  60  24   88  2 78 44 88
# 3   3  90  24   85  3 90 35 85
# 4   4  55  23   89  4 55 23 78
# 5   5  60  24   68  5 60 24 68
# 6   6  61  24   62  6 61 22 62
# 7   7  77  27   89  7 77 27 89
# 8   8  67  23   62  8 67 23 62
# 9   9  88  26   70  9 88 26 70
# 10 10  90  21   99 10 90 21 99

Answer 2

I experienced this same issue. In my case, I had a different number of observations between my imputed and non-imputed data sets. To fix this, after I merged the data, I then re-coded the variable .id . The mice package outputs .id when you call mice and complete(..., action = 'long') . This is different from your data frame variable id but they should correspond to each other by the following code.

library(dplyr)
# recode .id based on value of id
mydata <- mutate(mydata, .id = as.numeric(as.factor(id)))
# this step is important according to the mice manual
mydata <- mydata[order(mydata$.imp, mydata$.id),]

The as.mids function worked for me when I applied this recode and I hope it works for you, too.