khusmann / interlacer

An R package for working with labelled missingness
http://kylehusmann.com/interlacer/
Apache License 2.0
1 stars 1 forks source link

interlacer

Project Status: WIP – Initial development is in progress, but there
has not yet been a stable, usable release suitable for the
public. R-CMD-check codecov

When a value is missing in your data, sometimes you want to know why it is missing. Many textual tabular data sources will encode missing reasons as special values interlaced with the regular values in a column (e.g. N/A, REFUSED, -99, etc.). Unfortunately, the missing reasons are lost when these values are all converted into a single NA type. Working with missing reasons in R traditionally requires loading variables as character vectors and doing a bunch of string comparisons and type conversions to make sense of them.

interlacer provides functions that load variables from interlaced data sources into a special interlaced column type that holds values and NA reasons in separate channels of the same variable. In most contexts, you can treat interlaced columns as if they were regular values: if you take the mean of an interlaced column, for example, you get the mean of its values, without its missing reasons interfering in the computation.

Unlike a regular column, however, the missing reasons are still available. This means you can still filter data frames on variables by specific missing reasons, or generate summary statistics with breakdowns by missing reason. In other words, you no longer have to constantly manually include / exclude missing reasons in computations by filtering them with awkward string comparisons or type conversions… everything just works!

In addition to the introduction in vignette("interlacer") be sure to also check out:

⚠️ ⚠️ ⚠️ WARNING ⚠️ ⚠️ ⚠️

This library is currently in its experimental stages, so be aware that its interface is quite likely to change in the future. In the meantime, please try it out and let me know what you think!

Installation

The easiest way to get interlacer is to install via devtools:

install.packages("devtools") # If devtools is not already installed

devtools::install_github("khusmann/interlacer")

Usage

To use interlacer, load it into your current R session:

library(interlacer, warn.conflicts = FALSE)

interlacer supports the following file formats with these read_interlaced_*() functions, which extend the readr::read_*() family of functions:

As a quick demo, consider the following example file bundled with interlacer:

library(dplyr, warn.conflicts = FALSE)
library(readr)

read_file(interlacer_example("colors.csv")) |>
  cat()
#> person_id,age,favorite_color
#> 1,20,BLUE
#> 2,REFUSED,BLUE
#> 3,21,REFUSED
#> 4,30,OMITTED
#> 5,1,N/A
#> 6,41,RED
#> 7,50,OMITTED
#> 8,30,YELLOW
#> 9,REFUSED,REFUSED
#> 10,OMITTED,RED
#> 11,10,REFUSED

In this csv file, values are interlaced with three possible missing reasons: REFUSED, OMITTED, and N/A.

With readr, loading these data would result in a data frame where all missing reasons are replaced with NA:

read_csv(
  interlacer_example("colors.csv"),
  na = c("REFUSED", "OMITTED", "N/A")
)
#> # A tibble: 11 × 3
#>    person_id   age favorite_color
#>        <dbl> <dbl> <chr>         
#>  1         1    20 BLUE          
#>  2         2    NA BLUE          
#>  3         3    21 <NA>          
#>  4         4    30 <NA>          
#>  5         5     1 <NA>          
#>  6         6    41 RED           
#>  7         7    50 <NA>          
#>  8         8    30 YELLOW        
#>  9         9    NA <NA>          
#> 10        10    NA RED           
#> 11        11    10 <NA>

With interlacer, missing reasons are preserved:

(ex <- read_interlaced_csv(
  interlacer_example("colors.csv"),
  na = c("REFUSED", "OMITTED", "N/A")
))
#> # A tibble: 11 × 3
#>    person_id       age favorite_color
#>    <dbl,fct> <dbl,fct> <chr,fct>     
#>  1         1        20 BLUE          
#>  2         2 <REFUSED> BLUE          
#>  3         3        21 <REFUSED>     
#>  4         4        30 <OMITTED>     
#>  5         5         1 <N/A>         
#>  6         6        41 RED           
#>  7         7        50 <OMITTED>     
#>  8         8        30 YELLOW        
#>  9         9 <REFUSED> <REFUSED>     
#> 10        10 <OMITTED> RED           
#> 11        11        10 <REFUSED>

As you can see, in the printout above each column is defined by two types: a type for values, and a type for missing reasons. The age column, for example, has type double for its values, and type factor for its missing reasons:

ex$age
#> <interlaced<dbl, fct>[11]>
#>  [1] 20        <REFUSED> 21        30        1         41        50       
#>  [8] 30        <REFUSED> <OMITTED> 10       
#> NA levels: REFUSED OMITTED N/A

Computations automatically operate on values:

mean(ex$age, na.rm = TRUE)
#> [1] 25.375

But the missing reasons are still there! To indicate a value should be treated as a missing reason instead of a regular value, you can use the na() function. The following, for example, will filter the data set for all individuals that REFUSED to give their favorite color:

ex |>
  filter(favorite_color == na("REFUSED"))
#> # A tibble: 3 × 3
#>   person_id       age favorite_color
#>   <dbl,fct> <dbl,fct> <chr,fct>     
#> 1         3        21 <REFUSED>     
#> 2         9 <REFUSED> <REFUSED>     
#> 3        11        10 <REFUSED>

And here’s a pipeline that will compute a breakdown of the mean age of respondents for each favorite color, with separate categories for each missing reason:

ex |>
  summarize(
    mean_age = mean(age, na.rm = TRUE),
    n = n(),
    .by = favorite_color
  ) |>
  arrange(favorite_color)
#> # A tibble: 6 × 3
#>   favorite_color mean_age     n
#>   <chr,fct>         <dbl> <int>
#> 1 BLUE               20       2
#> 2 RED                41       2
#> 3 YELLOW             30       1
#> 4 <REFUSED>          15.5     3
#> 5 <OMITTED>          40       2
#> 6 <N/A>               1       1

But this just scratches the surface of what can be done with interlacer… check out vignette("interlacer") for a more complete overview!

Known Issues

  1. Some base functions, like base::ifelse(), drop the missing reason channel on interlaced types, converting them into regular vectors

For example:

ex |>
  mutate(
    favorite_color = ifelse(age < 18, na("REDACTED"), favorite_color)
  )
#> # A tibble: 11 × 3
#>    person_id       age favorite_color
#>    <dbl,fct> <dbl,fct> <chr>         
#>  1         1        20 BLUE          
#>  2         2 <REFUSED> <NA>          
#>  3         3        21 <NA>          
#>  4         4        30 <NA>          
#>  5         5         1 <NA>          
#>  6         6        41 RED           
#>  7         7        50 <NA>          
#>  8         8        30 YELLOW        
#>  9         9 <REFUSED> <NA>          
#> 10        10 <OMITTED> <NA>          
#> 11        11        10 <NA>

This is due to a limitation of R. If you run into this, use the tidyverse equivalent of the function. Tidyverse functions are designed to more correctly handle type conversions. In this example, we can use dplyr::if_else():

ex |>
  mutate(
    favorite_color = if_else(
      age < 18,
      na("REDACTED_UNDERAGE"),
      favorite_color,
      missing = na("REDACTED_MISSING_AGE")
    )
  )
#> # A tibble: 11 × 3
#>    person_id       age favorite_color        
#>    <dbl,fct> <dbl,fct> <chr,fct>             
#>  1         1        20 BLUE                  
#>  2         2 <REFUSED> <REDACTED_MISSING_AGE>
#>  3         3        21 <REFUSED>             
#>  4         4        30 <OMITTED>             
#>  5         5         1 <REDACTED_UNDERAGE>   
#>  6         6        41 RED                   
#>  7         7        50 <OMITTED>             
#>  8         8        30 YELLOW                
#>  9         9 <REFUSED> <REDACTED_MISSING_AGE>
#> 10        10 <OMITTED> <REDACTED_MISSING_AGE>
#> 11        11        10 <REDACTED_UNDERAGE>
  1. Performance with large data sets

You may notice that on large datasets interlacer runs significantly slower than readr / vroom. Although interlacer uses vroom under the hood to load delimited data, it is not able to take advantage of many of its optimizations because vroom does not currently support column-level missing values. As soon as vroom supports column-level missing values, I will be able to remedy this!

Related work

interlacer was inspired by the haven, labelled, and declared packages. These packages provide similar functionality to interlacer, but are more focused on providing compatibility with missing reason data imported from SPSS, SAS, and Stata. interlacer has slightly different aims:

  1. Be fully generic: Add a missing value channel to any vector type.
  2. Provide functions for reading / writing interlaced CSV files (not just SPSS / SAS / Stata files)
  3. Provide a functional API that integrates well into tidy pipelines

Future versions of interlacer will provide functions to convert to and from these other packages’ types.

For a more detailed discussion, see vignette("other-approaches").

Acknowledgements

The development of this software was supported, in whole or in part, by the Institute of Education Sciences, U.S. Department of Education, through Grant R305A170047 to The Pennsylvania State University. The opinions expressed are those of the authors and do not represent the views of the Institute or the U.S. Department of Education.