Combine multiple response categories into one indicator-style one

This is a function to speed up a common task of combining multiple responses (e.g. "Excellent", "Good") into a single one (e.g. "Excellent / Good"). It's useful for converting questions in their original format into one where we can use a single number to show the percentage of people who rated something excellent or good. Two additional steps are optional but done by default: non-answers are dropped with sub_nonanswers(), and the data is filtered for only the named aggregated responses. Both of these steps can be disabled independently.

Usage

combine_response(
  data,
  categories,
  response = response,
  value = value,
  filter_responses = TRUE,
  drop_nonanswers = TRUE,
  nons = c("Don't know", "Refused"),
  ...
)

Arguments

data

A data frame with groups maintained.

categories

A named list of vectors. Each list item should be a character vector of response categories to lump together, and each vector's name should be the new name for the aggregated response. This follows the ... argument of forcats::fct_collapse(), as that is where it is passed directly.

response

Bare column name of where responses are found, including those considered to be non-answers. Default: response

value

Bare column name of values, Default: value

filter_responses

Logical: if TRUE (the default), only responses matching the names of the categories argument will be kept.

drop_nonanswers

Logical: if TRUE (the default), will drop nonanswers with sub_nonanswers() after aggregating.

nons

Character vector of responses to be removed. Default: c("Don't know", "Refused")

...

Arguments passed on to sub_nonanswers

factor_response: Logical: if TRUE (default), returns response variable as a factor. This is likely a more useful way to handle response categories once non-answers have been removed.
rescale: Logical: if TRUE, values will be scaled based on their total. If FALSE (the default), values are scaled based on an assumption that all responses add to 1. In some cases, crosstabs with heavy rounding might not add up to 1 when they should, so rescaling helps handle that.

Value

A data frame with groups maintained. It will have fewer rows, depending on arguments passed for filtering and dropping nonanswers.

Details

A few gotchas are possible with this function:

Unless you're using a data frame with only one location, category, and group, you'll likely want to use this on a grouped data frame. This function maintains all groups of the data frame passed in, but it is outside the purview of this function to create any groups. You should do that in advance yourself (see examples).
Removing non-answers and filtering happen independently but are informed by each other. If you decide to filter the data but not drop non-answers, the values of nons will be included as response categories to keep. This is to avoid a situation where more data is dropped than you might intend.

Examples

  xt <- system.file("extdata/test_xtab2021.xlsx", package = "dcws") |>
      read_xtabs(process = TRUE, year = 2021)
#> ℹ xtab2df is being called on the data with the following parameters:
#> → year = 2021
#> → col = x1
#> → code_pattern = `NULL` (filling in default pattern)
#> → verbose = TRUE
# how responsive is local govt?
  local_govt <- dplyr::filter(xt, code == "Q4A", category == "Age")
  unique(local_govt$response)
#> [1] "Excellent"                                 
#> [2] "Good"                                      
#> [3] "Fair"                                      
#> [4] "Poor"                                      
#> [5] "Don't know enough about it in order to say"
#> [6] "Refused"                                   
# combine excellent & good, drop non-answers, filter
# note that this question has different wording for "Don't know" response
  local_govt |>
     dplyr::group_by(category, group) |>
     combine_response(
         list(excellent_good = c("Excellent", "Good")),
         nons = c("Don't know enough about it in order to say", "Refused")
     )
#> # A tibble: 4 × 4
#> # Groups:   category, group [4]
#>   category group      response       value
#>   <fct>    <fct>      <fct>          <dbl>
#> 1 Age      Ages 18-34 excellent_good 0.253
#> 2 Age      Ages 35-49 excellent_good 0.176
#> 3 Age      Ages 50-64 excellent_good 0.319
#> 4 Age      Ages 65+   excellent_good 0.560
# combine but no filtering or dropping
local_govt |>
     dplyr::group_by(category, group) |>
     combine_response(
         list(excellent_good = c("Excellent", "Good")),
         nons = c("Don't know enough about it in order to say", "Refused"),
         filter_responses = FALSE,
         drop_nonanswers = FALSE
     )
#> # A tibble: 20 × 4
#> # Groups:   category, group [4]
#>    category group      response                                   value
#>    <fct>    <fct>      <fct>                                      <dbl>
#>  1 Age      Ages 18-34 Don't know enough about it in order to say  0.17
#>  2 Age      Ages 18-34 excellent_good                              0.21
#>  3 Age      Ages 18-34 Fair                                        0.34
#>  4 Age      Ages 18-34 Poor                                        0.27
#>  5 Age      Ages 18-34 Refused                                     0   
#>  6 Age      Ages 35-49 Don't know enough about it in order to say  0.15
#>  7 Age      Ages 35-49 excellent_good                              0.15
#>  8 Age      Ages 35-49 Fair                                        0.27
#>  9 Age      Ages 35-49 Poor                                        0.42
#> 10 Age      Ages 35-49 Refused                                     0   
#> 11 Age      Ages 50-64 Don't know enough about it in order to say  0.06
#> 12 Age      Ages 50-64 excellent_good                              0.3 
#> 13 Age      Ages 50-64 Fair                                        0.35
#> 14 Age      Ages 50-64 Poor                                        0.28
#> 15 Age      Ages 50-64 Refused                                     0   
#> 16 Age      Ages 65+   Don't know enough about it in order to say  0.16
#> 17 Age      Ages 65+   excellent_good                              0.47
#> 18 Age      Ages 65+   Fair                                        0.2 
#> 19 Age      Ages 65+   Poor                                        0.18
#> 20 Age      Ages 65+   Refused                                     0   

# multiple combined categories--useful for Likert questions
area_change <- dplyr::filter(xt, code == "Q2", category == "Gender")
unique(area_change$response)
#> [1] "Much better"     "Somewhat better" "About the same"  "Somewhat worse" 
#> [5] "Much worse"      "Don't know"      "Refused"        
# using default filtering, this drops middle category ("About the same")
area_change |>
  dplyr::group_by(category, group) |>
  combine_response(
     list(
         `Getting better` = c("Much better", "Somewhat better"),
         `Getting worse` = c("Much worse", "Somewhat worse")
     ))
#> # A tibble: 4 × 4
#> # Groups:   category, group [2]
#>   category group  response       value
#>   <fct>    <fct>  <fct>          <dbl>
#> 1 Gender   Male   Getting better 0.455
#> 2 Gender   Male   Getting worse  0.222
#> 3 Gender   Female Getting better 0.305
#> 4 Gender   Female Getting worse  0.337
# instead turn off filtering but keep non-answer dropping
# unfortunately now responses are out of order...
area_change |>
  dplyr::group_by(category, group) |>
  combine_response(
     list(
         `Getting better` = c("Much better", "Somewhat better"),
         `Getting worse` = c("Much worse", "Somewhat worse")
     ), filter_responses = FALSE)
#> # A tibble: 6 × 4
#> # Groups:   category, group [2]
#>   category group  response       value
#>   <fct>    <fct>  <fct>          <dbl>
#> 1 Gender   Male   About the same 0.323
#> 2 Gender   Male   Getting better 0.455
#> 3 Gender   Male   Getting worse  0.222
#> 4 Gender   Female About the same 0.368
#> 5 Gender   Female Getting better 0.305
#> 6 Gender   Female Getting worse  0.337
# ...unless response is already a factor, in which case levels stick
area_change |>
  dplyr::group_by(category, group) |>
  dplyr::mutate(response = forcats::as_factor(response)) |>
  combine_response(
     list(
         `Getting better` = c("Much better", "Somewhat better"),
         `Getting worse` = c("Much worse", "Somewhat worse")
     ), filter_responses = FALSE)
#> # A tibble: 6 × 4
#> # Groups:   category, group [2]
#>   category group  response       value
#>   <fct>    <fct>  <fct>          <dbl>
#> 1 Gender   Male   Getting better 0.455
#> 2 Gender   Male   About the same 0.323
#> 3 Gender   Male   Getting worse  0.222
#> 4 Gender   Female Getting better 0.305
#> 5 Gender   Female About the same 0.368
#> 6 Gender   Female Getting worse  0.337