extractManipData() does not work correctly after applying globalRecode()

[MuellerRoman]

Dear sdcMicro maintainers

I applied globalRecode() to the numeric variable age in order to generalize it into age categories. However, after calling extractManipData() I noticed that the age variable is still numeric rather than the expected factor.

I suspect the reason lies in aux_functions.R, line 471:

## quick and dirty: ensure that keyVars are factors:
  if (!is.null(k) && !ignoreKeyVars) {
    for (i in seq_len(ncol(k))) {
      cc <- class(origKeys[[colnames(k)[i]]])[1]
      vname <- colnames(k)[i]
      v_p <- o[[vname]]
      if (cc != class(v_p)[1]) {
        if (cc == "integer") {
          o[[vname]] <- as.integer(v_p)
        }
        if (cc == "character") {
          o[[vname]] <- as.character(v_p)
        }
        if (cc == "numeric") {
          o[[vname]] <- as.numeric(v_p)
        }
        if (cc == "logical") {
          o[[vname]] <- as.logical(v_p)
        }
        if (cc == "ordered") {
          o[[vname]] <- as.ordered(v_p)
        }
      }
    }
  }

Because the code section above restores the variable class to that of the original dataset, factorized numeric variables are converted back to numeric.

Note:

• The section above was last modified in extractManipData fails when only 1 keyVar is set #321
• In order to apply globalRecode() to a numeric variable, it must first be defined as a key variable (i.e., as categorical) within createSdcObj(). This is not directly related to the issue above, but it may contribute to the confusion.

Here is a reproducible example:

library(sdcMicro)

# create test data ----

set.seed(123)  # for reproducibility
n <- 50

# age 18–65
age <- round(runif(n, min = 18, max = 65))

# income
income <- round(rnorm(n, 60000, 20))

# occupation categories
occupation <- sample(c("student", "employed", "unemployed"),
                     size = n, replace = TRUE,
                     prob = c(0.3, 0.5, 0.2))

# gender categories
gender <- sample(c("male", "female", "nonbinary"),
                 size = n, replace = TRUE,
                 prob = c(0.45, 0.45, 0.10))

# depression score
depression_score <- rnorm(n, mean = 10, sd = 5)

# combine
df <- data.frame(
  age,
  income,
  occupation,
  gender,
  depression_score
)

# Apply globalRecode() to numeric variable "age" ----
sdcCat <- createSdcObj(dat = df,
                       keyVars = c("gender", "occupation", "age"))

sdcCat <- globalRecode(sdcCat,
                       column = "age",
                       breaks = seq(10, 80, 10)) 

# extract dataset
df_recoded <- extractManipData(sdcCat)

# after calling extractManipData(), "age" remains numeric rather than a factor
is.numeric(df_recoded$age)
is.factor(df_recoded$age)

# when extracting "age" directly from the sdcObject, it's correctly represented as factor
is.factor(sdcCat@manipKeyVars$age)```

[bernhard-da]

hi @MuellerRoman thx for reporting, this is indeed a bit tricky.

From sdcMicro's perspective, the variable (age) is initially numeric (as it was when calling createSDCObj()). This is also why it is returned as numeric when applying extractManipData() which I think is correct.

Quick workaround to fix your issue would be to call the globalRecode() function just before setting up the problem instance;

# recode before creating sdc-object
df <- globalRecode(
  obj = df,
  column = "age",
  breaks = seq(10, 80, 10)
) 

# create object
sdcCat <- createSdcObj(
  dat = df,   
  keyVars = c("gender", "occupation", "age")
)

class(extractManipData(sdcCat)$age)
# [1] "factor"

[bernhard-da]

closing because I do not think this is a "real" bug.