Skip to contents

Prove a recode with a before/after cross-tabulation

Source: R/categorical_guard.R
verify_recode.Rd

Checks, after the fact, that a recode did what you said it did. It cross-tabulates the original values against the recoded ones and stops with an error unless each original category went to exactly one new category, that category is the one in your declared mapping, no rows were lost or added, and no value became missing (or was filled in). Use it directly after any recode, including recodes done by other code or other software, to catch swapped groups before a model runs.

Usage

verify_recode(original, recoded, declared)

Arguments

original, recoded

Atomic vectors of the same length: the values before and after the recode, row for row.

declared

Named character vector, old = new: the mapping the analyst says was applied. Old values not listed are expected to be unchanged.

Value

Invisibly, when every check passes, the cross-tabulation as a data frame with columns original, recoded (factors) and Freq (integer), including the zero cells of the full table. Otherwise an error.

Details

Factors are compared through their labels. The checks, in order:

  1. original and recoded are atomic and the same length;

  2. is.na(original) equals is.na(recoded) element by element;

  3. in the table of non-missing pairs, no original value has more than one recoded value (the recode is a function of the label);

  4. each original value's recoded value equals declared[old]; an original value that is not a name of declared must be unchanged (identity).

The first failed check is an error naming the offending categories.

See also

guard_recode, recode_manifest.

Examples

before <- c("W", "B", "W", NA)
after <- c("White", "Black", "White", NA)
tab <- verify_recode(before, after, c(W = "White", B = "Black"))
tab[tab$Freq > 0, ]
#>   original recoded Freq
#> 1        B   Black    1
#> 4        W   White    2

# a value not in the declared mapping must be unchanged
verify_recode(c("W", "X"), c("White", "X"), c(W = "White"))

# swapped groups are caught
try(verify_recode(c("W", "B"), c("Black", "White"), c(W = "White", B = "Black")))
#> Error : verify_recode: ‘B’ was mapped to ‘White’ but the declared mapping says ‘Black’. THIS is how groups get swapped; fix the recode before any model runs.

# a value that became NA is caught
try(verify_recode(c("W", "B"), c("White", NA), c(W = "White", B = "Black")))
#> Error : verify_recode: missingness changed during the recode (values silently became NA, or NAs were filled).