Blur Example"

In deident: Persistent Data Anonymization Pipeline

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

One way to reduce identifiability of a data set is by converting a categorical variable to have a more aggregated taxonomy (i.e. a many-to-one mapping). Here we refer to such a method as a 'blur' as it causes features to be joined together in such a way to hide the underlying information.

As an example, consider the ShiftsWorked data:

library(deident)
head(ShiftsWorked)

A simple 'blur' might be to change the taxonomy of 'Shift' e.g. combine 'Day' and 'Night' into a new group 'Working' and ignore the 'Rest' shifts. To do this we define the values we wish to change as a vector, build a pipeline and apply it to the data:

shift_blur <- c("Day" = "Working", "Night" = "Working")
blur_pipe <- ShiftsWorked |>
  add_blur(Shift, blur=shift_blur)

apply_deident(ShiftsWorked, blur_pipe)

The category_blur utility

Applying the blur is relatively simple, but constructing it can be complex. Consider the starwars data set supplied by dplyr:

starwars <- dplyr::starwars
head(starwars)

And notably the species variable:

table(starwars$species)

Imagine we wanted to reduce identifiability by aggregating the data into Human vs Non-Human. We could code the vector by hand, but human error can lead to mistakes. To aid in designing complex blurs we supply the category_blur utility which uses regex to define the groups.

human_blur <- category_blur(
  starwars$species,
  "NotHuman" = "^(?!Human)" # Doesn't start with "Human"
)

And the vector returned can be passed into a new pipeline as before.

species_pipe <- starwars |>
  add_blur(species, blur=human_blur)

new_starwars <- apply_deident(starwars, species_pipe)

table(new_starwars$species)

deident documentation built on April 3, 2025, 6:14 p.m.