Simulating categorical variables

Abigail Stamm

2026-04-22

This vignette covers the basics of creating categorical variables. I would like your feedback on some of the features I am considering adding. For other variable types, please see the Settings Table Design vignette.

The toysurveydata package is designed to be a quick, easy way to create synthetic data that mimic real data structures for the following purposes:

• to test workflows and error handling
• to teach data management

When not to use toysurveydata

This package does not perform any special modeling and does not require existing data. Most functions calculate only one variable at a time and do not take into account the values of or relationships with other variables. This package would not be appropriate for research.

Most of the functions in this package read in a settings table that follows a specific data schema. The settings table should be populated with values and probabilities based on the results of past surveys and research.

Categorical single item selection

Settings dataset

For categorical data where only one item is selected, your table should contain the following settings. All other settings are ignored.

• variable: variable name in final dataset
• type = “select-one”: string that triggers select_categorical_set()
  Only the value in the first row is used.
• options: possible values, with one value for each row
• probability_1: proportions or probabilities of that item being selected
  • Use response probabilities from past surveys.
  • The function uses sample(), so probabilities need not sum to 100.
• miss_pct: percentage of missing values expected (between 0 and 100)
  • If you want no missing values, set “miss_pct” to 0.
  • If there are multiple values for “miss_pct”, the function uses the maximum non-missing value.
  • If all values are missing, “miss_pct” is set to 0.

Response dataset

The response dataset is created by running the select_categorical_set() function, which is a wrapper for the sample_one() function. If your survey contains many single item selection questions, this function allows you to generate the data for all of the questions at once.

Example

Create a simple data frame with each of these settings. This data frame could contain multiple different variables. I am showing only one variable here for simplicity.

gender_opts <- data.frame(
  variable = "gender", 
  type = "select-one", 
  options = c("Female", "Male", "Other", "Prefer not to say"),
  probability_1 = c(50, 50, 5, 5),
  miss_pct = 10 
)

After your settings file is created, read it into select_categorical_set().

gender_resp <- select_categorical_set(d = gender_opts, count = 20)

Categorical multiple item selection

Settings dataset

For categorical data where multiple items can be selected, your table should contain the following settings. All other settings are ignored.

• variable: variable name used to identify options to use in the function
• type = “select-many”: string that triggers select_many_set()
• options: possible values, with one value for each row
  • These will become the variable names in the final dataset, so they should be valid R object names. (Imagine question labels in survey software.)
  • The string “none” is special. The function checks if any strings in the options contain it. If so, any additional selections for the same record are removed.
• labels: value labels for the options (can duplicate “options”)
  • For numeric values instead of strings, replace the strings with numbers. The Boolean value TRUE for all options also works.
• probability_1: proportions or probabilities of that item being selected
  • Use response probabilities from past surveys.
  • The function uses sample(), so probabilities need not sum to 100.
• probability_2: proportions or probabilities (can duplicate “probability_1”)
  • I added two settings for probabilities for two reasons:
    1. to remove very common values, to prevent all records from being assigned to them.
    2. to remove “none” as a possible value for additional selections.
• max_opts: maximum allowed responses
  • A record can be assigned between 1 and “max_opts” values.
  • If there are multiple values for “max_opts”, the function takes the maximum non-missing value.
• miss_pct: percentage of missing values expected (between 0 and 100)
  • If you want no missing values, set “miss_pct” to 0.
  • If there are multiple values for “miss_pct”, the function uses the maximum non-missing value.
  • If all values for “miss_pct” are missing, “miss_pct” is set to 0.

Future multiple item selection features under consideration

1. Add an option to skip the step that removes additional values when a record is assigned “none” to provide messier data for teaching data validation and cleaning.
2. Add an option to force records that were not assigned “none” to be assigned exactly “max_opts” values.
3. Add an option to set unchecked values as something other than missing when at least one value is checked.
4. If “max_opts” is missing, set “max_opts” to the total number of values.

Response dataset

The response dataset is created by running the select_many_set() function, which is a wrapper for the sample_many() function. If your survey contains many multiple item selection questions, this function allows you to generate the data for all of the questions at once.

Example

Create a simple data frame with each of these settings. In this fictional community, most people prefer to visit their primary care doctors for all health needs.

seek_care_opts <- data.frame(
  variable = "seek_care", 
  type = "select-many", 
  options = c("care_doc", "care_clinic", "care_emerg", 
              "care_tele", "care_urgent", "care_other", "care_none"), 
  labels = c("Doctor's office", "Outpatient clinic", "Emergency room", 
             "Telehealth", "Urgent care clinic", "Other", "Do not seek care"),
  probability_1 = c(90, 5, 1, 2, 3, 1, 1), 
  probability_2 = c( 0, 5, 1, 2, 3, 1, 0), 
  max_opts = 3, 
  miss_pct = 0
)

After your settings file is created, read it into select_many_set().

seek_care_resp <- select_many_set(d = seek_care_opts, count = 20)

Combining all data

The functions add “id” columns to all response datasets, so I can join on “id” to combine the datasets. This step is not handled by the package.

resps <- list(gender_resp, seek_care_resp)
full_data <- purrr::reduce(resps, dplyr::left_join, by = "id") |> 
  dplyr::arrange(id)