R 中 MICE 包的被动插补语法

Question

Suppose we have a numeric variable age , which is sometimes missing. In using it to predict other variables, we want to allow for non-linearity, so we create age_factor . We should impute age_factor passively: https://www.gerkovink.com/miceVignettes/Passive_Post_processing/Passive_imputation_post_processing.html

But how would the syntax work ? Something like:

method["age_factor"] <- "~ 18_34 if age <= 34,
                           35_49 if age <= 49... "

Thank you !

Answer 1

Why do you can't do it like that:

age_factor <- function(x){
     if (x <= 34) {
         y <- "18_34"
     } else {
         y <- "35_49"
     }
return(y)    
}

age_factor(32)
[1] "18_34"

Answer 2

Using syntax which is close to what you have written:

method["age_factor"] <- "~ ifelse(is.na(age), NA,  
   ifelse(age < 20, 'under 20', 
   ifelse(age < 25, '20 to 24', 
   ifelse(age < 35, '25 to 34',
   '35 and over'))))"

NB I have included NA in the possible results because imputations can sometimes return NA. (Also, you need to enclose your factor level names in single quotes, because the expression as a whole is enclosed in double quotes, otherwise R will think your code for age_factor ends prematurely.)

However--I would avoid ifelse() in this instance, preferring cutr::smart_cut() for its simplicity and resistance to hard-to-spot semantic errors:

require(cutr)
cuts <- c(-Inf, 20, 25, 35, Inf)
imp$age_factor <- with(imp, smart_cut(age, cuts, 
    labels = c('under 20', '20 to 24', '25 to 34', '35 and over'), output = "factor"))

NB the argument output = "factor" will give you an unordered factor; you can set this to other values eg ordered factor, integer, etc.

Happy imputing! :-)