Imputation based on factor level

Question

I am trying to perform imputation with MICE on the following dataframe.

marketValue <- c(NA, 234234, NA, 243243, NA, NA, 234523, NA, 232427, 112214)
bathrooms <- c(3,3,2,3,5,4,1,5,6,3)
garageSqFt <- c(400, 385, 454, 534, 210, NA, 342, 423, 535, NA)
totalSqFT <- c(NA, NA, 1231, 2232, 4564, 2122, 4324, 4342, 1299, 4355)
units <- c(1, 1, 1, 1, 1, 1, 1.5, NA, 2, 5)
subDivId <- c("112", "111", "111", "111", "112", "111", "112", "112", "111", 
"112")
data <- data.frame(marketValue, bathrooms, garageSqFt, totalSqFT, units, 
subDivId)

In the actual data frame there are about 1300 subDivId factor levels and I would like to create new data frames (each of which have rows all with the same subDivId) then impute within each data frame. My attempt

splitSubDiv <- split(data, data$subDivId) 
for (neighborhood in splitSubDiv){
    testData <- mice(neighborhood, m=5, maxit = 5) 
    impData <- complete(testData, 5) 
    str(impData) }

This doesn't seem to work. No imputation is being done, things are still NA. What am I doing Wrong?

What is your expected results? Give a brief logic on how to get there — Onyambu, Feb 14 '18 at 05:33

score 1 · Accepted Answer · answered Feb 14 '18 at 17:26

1

In your attempt you are not assigning your imputed data frames to objects, so the results are not preserved.

You can try this approach using lapply instead of a for loop. You'll end up with a list of imputed data frames.

splitSubDivImp <- lapply(splitSubDiv, function(neighborhood) {
  testData <- mice(neighborhood, m = 5, maxit = 5) 
  impData <- complete(testData, 5)
})

splitSubDivImp

answered Feb 14 '18 at 17:26

Juan Bosco

1,420
5
19
23

2

This will run `mice` five times, using only the last imputation and wasting the 4 others. It would be more efficient to just run `testData <- mice(...)` and then `impData <- lapply(1:5, function(i) complete(testData,i))`. – SimonG Feb 21 '18 at 08:29

Imputation based on factor level

1 Answers1