recoding variables
in dplyr
February 8, 2020
I don’t often deal with questionnaire data in R, but Ariana and I have started trying import her qualtrics data into R and to write a script to score her measures. The first step is to recode the variables to make it possible to add up scores on subscales.
load packages
library(tidyverse)
make a little dataframe
df <- data.frame("pp_no" = 1:12,
"sectionA_1" = c("Strongly Agree","Agree", "Disagree","Strongly Disagree"),
"sectionA_2" = c("Strongly Agree","Agree", "Disagree","Strongly Disagree"),
"sectionB_1" = c("Frequently","Sometimes", "Infrequently"),
"sectionB_2" = c("Frequently","Sometimes", "Infrequently"))
Option 1: use mutate() and case_when()
My first intuition is to use case_when(), which I have
written about before. This option creates a new column for the first variable, recoding the response options into numeric values 1-4.
df %>%
mutate(sectionA_1_cw = case_when(sectionA_1 == "Strongly Agree" ~ 4,
sectionA_1 == "Agree" ~ 3,
sectionA_1 == "Disagree" ~ 2,
sectionA_1 == "Strongly Disagree" ~ 1))
## pp_no sectionA_1 sectionA_2 sectionB_1 sectionB_2
## 1 1 Strongly Agree Strongly Agree Frequently Frequently
## 2 2 Agree Agree Sometimes Sometimes
## 3 3 Disagree Disagree Infrequently Infrequently
## 4 4 Strongly Disagree Strongly Disagree Frequently Frequently
## 5 5 Strongly Agree Strongly Agree Sometimes Sometimes
## 6 6 Agree Agree Infrequently Infrequently
## 7 7 Disagree Disagree Frequently Frequently
## 8 8 Strongly Disagree Strongly Disagree Sometimes Sometimes
## 9 9 Strongly Agree Strongly Agree Infrequently Infrequently
## 10 10 Agree Agree Frequently Frequently
## 11 11 Disagree Disagree Sometimes Sometimes
## 12 12 Strongly Disagree Strongly Disagree Infrequently Infrequently
## sectionA_1_cw
## 1 4
## 2 3
## 3 2
## 4 1
## 5 4
## 6 3
## 7 2
## 8 1
## 9 4
## 10 3
## 11 2
## 12 1
Option 2a: use mutate() and recode()
case_when() can take pretty complicated arguments, and evaluates them in an ordered way, which might be more than you need. Perhaps recode() from dplyr is a simpler option.
df %>%
mutate(sectionA_1_rc = recode(sectionA_1, "Strongly Agree" = "4",
"Agree" = "3",
"Disagree" = "2",
"Strongly Disagree" = "1", .default = "NA"))
## pp_no sectionA_1 sectionA_2 sectionB_1 sectionB_2
## 1 1 Strongly Agree Strongly Agree Frequently Frequently
## 2 2 Agree Agree Sometimes Sometimes
## 3 3 Disagree Disagree Infrequently Infrequently
## 4 4 Strongly Disagree Strongly Disagree Frequently Frequently
## 5 5 Strongly Agree Strongly Agree Sometimes Sometimes
## 6 6 Agree Agree Infrequently Infrequently
## 7 7 Disagree Disagree Frequently Frequently
## 8 8 Strongly Disagree Strongly Disagree Sometimes Sometimes
## 9 9 Strongly Agree Strongly Agree Infrequently Infrequently
## 10 10 Agree Agree Frequently Frequently
## 11 11 Disagree Disagree Sometimes Sometimes
## 12 12 Strongly Disagree Strongly Disagree Infrequently Infrequently
## sectionA_1_rc
## 1 4
## 2 3
## 3 2
## 4 1
## 5 4
## 6 3
## 7 2
## 8 1
## 9 4
## 10 3
## 11 2
## 12 1
Seems like case_when() and recode() work similarly, but by making a new variable for every recoded one, the df will end up twice as wide as it needs to be. There is a solution to that : transmute. If you transmute() rather than mutate(), you can recode the variable and drop the old one in one step. Lets try that.
df %>%
transmute(sectionA_1_rc = recode(sectionA_1, "Strongly Agree" = "4",
"Agree" = "3",
"Disagree" = "2",
"Strongly Disagree" = "1", .default = "NA"))
## sectionA_1_rc
## 1 4
## 2 3
## 3 2
## 4 1
## 5 4
## 6 3
## 7 2
## 8 1
## 9 4
## 10 3
## 11 2
## 12 1
Yikes! that is a bit drastic! It drops all the variables except the one you have recoded. If you are recoding all the variables in one go, that might work. But maybe a better solution would be to mutate and replace the variable you are recoding, rather than naming it something new.
df %>%
mutate(sectionA_1 = recode(sectionA_1, "Strongly Agree" = "4",
"Agree" = "3",
"Disagree" = "2",
"Strongly Disagree" = "1", .default = "NA"))
## pp_no sectionA_1 sectionA_2 sectionB_1 sectionB_2
## 1 1 4 Strongly Agree Frequently Frequently
## 2 2 3 Agree Sometimes Sometimes
## 3 3 2 Disagree Infrequently Infrequently
## 4 4 1 Strongly Disagree Frequently Frequently
## 5 5 4 Strongly Agree Sometimes Sometimes
## 6 6 3 Agree Infrequently Infrequently
## 7 7 2 Disagree Frequently Frequently
## 8 8 1 Strongly Disagree Sometimes Sometimes
## 9 9 4 Strongly Agree Infrequently Infrequently
## 10 10 3 Agree Frequently Frequently
## 11 11 2 Disagree Sometimes Sometimes
## 12 12 1 Strongly Disagree Infrequently Infrequently
dplyr “scoped” verbs (_all, _at, _if)
The questions for each scale tend to have the same response options, so it should be possible to mutate all the variables that share features in one go. This is where scoped verbs in dplyr come in.
I am not the first person to write about these. I found blog posts by Rebecca Barter and Suzan Baert really useful. Thanks @RLadies!
Mutate options
-
mutate_all()affects every variable -
mutate_at()affects variables selected with a character vector orvars() -
mutate_if()affects variables selected with a predicate function
Option 2b: use mutate_at() and recode()
In this situation, I think mutate_at() is going to be most useful. To select particular variables, use the vars() function and starts_with(), ends_with(), or contains() to specify which variables to apply the function to. Here the function we want to apply to each variable starting with section_A_ is recode(), so we specify that using ~ .Then set up the recode mappings, and specify that “NA” should be used as default if a match can’t be made.
When I look at this code, it reads… mutate only the variables starting with section A_, apply the recode function to all of them, recoding Strongly Agree as 4 etc etc, putting NA in if you can’t make a match.
df %>%
mutate_at(vars(starts_with("sectionA_")), ~ recode(., "Strongly Agree" = "4",
"Agree" = "3",
"Disagree" = "2",
"Strongly Disagree" = "1", .default = "NA"))
## pp_no sectionA_1 sectionA_2 sectionB_1 sectionB_2
## 1 1 4 4 Frequently Frequently
## 2 2 3 3 Sometimes Sometimes
## 3 3 2 2 Infrequently Infrequently
## 4 4 1 1 Frequently Frequently
## 5 5 4 4 Sometimes Sometimes
## 6 6 3 3 Infrequently Infrequently
## 7 7 2 2 Frequently Frequently
## 8 8 1 1 Sometimes Sometimes
## 9 9 4 4 Infrequently Infrequently
## 10 10 3 3 Frequently Frequently
## 11 11 2 2 Sometimes Sometimes
## 12 12 1 1 Infrequently Infrequently
Then you can %>% another series of recoding for other variables in the df.
df %>%
mutate_at(vars(starts_with("sectionA_")), ~ recode(., "Strongly Agree" = "4",
"Agree" = "3",
"Disagree" = "2",
"Strongly Disagree" = "1", .default = "NA")) %>%
mutate_at(vars(starts_with("sectionB_")), ~ recode(., "Frequently" = "3",
"Sometimes" = "2",
"Infrequently" = "0",
.default = "NA"))
## pp_no sectionA_1 sectionA_2 sectionB_1 sectionB_2
## 1 1 4 4 3 3
## 2 2 3 3 2 2
## 3 3 2 2 0 0
## 4 4 1 1 3 3
## 5 5 4 4 2 2
## 6 6 3 3 0 0
## 7 7 2 2 3 3
## 8 8 1 1 2 2
## 9 9 4 4 0 0
## 10 10 3 3 3 3
## 11 11 2 2 2 2
## 12 12 1 1 0 0
Done!