Lightweight Data Structure for Recoding Categorical Data without Factors.

rcoder

rcoder outlines a lightweight data structure that captures categorical codings and easily converts them to other implementations.

Usage

rcoder’s main functions are coding() and code(). code() maps a character-valued key to some value to represent a categorical level.

library(rcoder)
library(magrittr)

code("Yes", 1)
#> <Code>
#> label: 'Yes'
#> value: 1

code() objects can hold an arbitrary amount of metadata. Useful pieces of information to know are descriptions of what levels represent (in case the label doesn’t have enough information) and whether or not the value represents a missing value.

code(
  "No response",
  -99,
  description = "Surveyed individual ignored question when asked",
  missing = TRUE
)
#> <Code>
#> label: 'No response'
#> value: -99
#> description: 'Surveyed individual ignored question when asked'
#> Represents a missing value

coding() is a collection of code() objects that represents the full set of values for a categorical variable.

coding(
  code("Don't know", 0, missing = TRUE),
  code("Never", 1),
  code("Rarely", 2),
  code("Sometimes", 3),
  code("Frequently", 4),
  code("Always", 5),
  code("No response", -99, missing = TRUE),
  code("Refused", -88, missing = TRUE),
  code("Absent", -77, missing = TRUE)
)
#> <Coding>
#> # A tibble: 9 × 5
#>   link        label       value description missing
#>   <chr>       <chr>       <dbl> <chr>       <lgl>  
#> 1 Don't know  Don't know      0 Don't know  TRUE   
#> 2 Never       Never           1 Never       FALSE  
#> 3 Rarely      Rarely          2 Rarely      FALSE  
#> 4 Sometimes   Sometimes       3 Sometimes   FALSE  
#> 5 Frequently  Frequently      4 Frequently  FALSE  
#> 6 Always      Always          5 Always      FALSE  
#> 7 No response No response   -99 No response TRUE   
#> 8 Refused     Refused       -88 Refused     TRUE   
#> 9 Absent      Absent        -77 Absent      TRUE

coding() objects are designed to be an intermediate representation of categorical data so that they can be converted into different representations on the fly, e.g. ODK XLSForm choices and STATA/SPSS columns via haven.

coding(
  code("Never", 1),
  code("Rarely", 2),
  code("Sometimes", 3),
  code("Frequently", 4),
  code("Always", 5),
  .label = "frequency"
) %>%
  coding_to_odk()
#> # A tibble: 5 × 3
#>   list_name  name label     
#>   <chr>     <dbl> <chr>     
#> 1 frequency     1 Never     
#> 2 frequency     2 Rarely    
#> 3 frequency     3 Sometimes 
#> 4 frequency     4 Frequently
#> 5 frequency     5 Always

To facilitate recoding, coding() objects link to one another through the code() labels. If multiple values are collapsed into one, use the links_from parameter to identify which values are combined into one.

original_coding <- coding(
  code("No", 0L),
  code("Yes", 1L),
  code("No response", -99L, missing = TRUE),
  code("Refused", -88L, missing = TRUE),
  code("Absent", -77L, missing = TRUE)
)

new_coding <- coding(
  code("No", 0L),
  code("Yes", 1L),
  code("Missing", NA, links_from = c("No response", "Refused", "Absent"))
)

new_coding %>%
  link_codings(original_coding)
#>          link label_to value_to     label_1 value_1
#> 1      Absent  Missing       NA      Absent     -77
#> 2          No       No        0          No       0
#> 3 No response  Missing       NA No response     -99
#> 4     Refused  Missing       NA     Refused     -88
#> 5         Yes      Yes        1         Yes       1

These linked codings can be converted into a function that accepts a vector and returns a recoded vector.

new_coding %>%
  link_codings(original_coding) %>%
  make_recode_query()
#> function (x) 
#> {
#>     attrs <- attributes(x)
#>     x <- dplyr::case_when(x == -77L ~ NA_integer_, x == 0L ~ 
#>         0L, x == -99L ~ NA_integer_, x == -88L ~ NA_integer_, 
#>         x == 1L ~ 1L)
#>     attributes(x) <- attrs
#>     x
#> }
#> <environment: 0x7faab3ea3298>

Installation

Install rcoder from CRAN with:

install.packages("rcoder")

To get the latest stable changes that may not be on CRAN, install from the nyuglobalties r-universe with:

install.packages("rcoder", repos = "https://nyuglobalties.r-universe.dev")

Finally, get the development version straight from GitHub:

# install.packages("remotes")
remotes::install_github("nyuglobalties/rcoder")