Recoding Values
Introduction
Recoding values is one of the most common tasks a researcher needs to do before data analysis. For me, often I need to prepare my data in R first before using it for advanced statistical analyses in Mplus. In this case, it is important to recode missing values to a specific extreme value (e.g., -999) since it will be more efficient for Mplus to recognize and handle missing values. In this post, I will demonstrate a way (Oh yeah! This is the beauty of R： All roads lead to Rome.) to handle the recording task using case_when function in Tidyverse. There are different ways to get this job done, but I feel that case_when makes the most sense to me. Let’s get started.
#load required packages
library(tidyverse)
Let’s create a hypothetical dataset!
data<-data.frame(cost1 = c(2, 3, 4, 2, NA),
cost2 = c(1, 2, 3, 4, 5),
cost3 = c(NA, NA, 3, 2, 1)
)
data
## cost1 cost2 cost3
## 1 2 1 NA
## 2 3 2 NA
## 3 4 3 3
## 4 2 4 2
## 5 NA 5 1
Task 1: Recode Missing Values to -999
Let’s first try a slow way to recode the missing values to -999:
#slow way: deal with one variable at a time
data %>%
mutate(
cost1 = case_when(
is.na(cost1) ~ -999,
TRUE ~ as.numeric(cost1)),
cost3 = case_when(
is.na(cost3) ~ -999,
TRUE ~ as.numeric(cost3))
)
## cost1 cost2 cost3
## 1 2 1 -999
## 2 3 2 -999
## 3 4 3 3
## 4 2 4 2
## 5 -999 5 1
is.na(cost1) ~ -999 asks R to evaluate if a specific case is missing information in cost1. If a case is missing, then (~) it will be coded as -999. TRUE ~ as.numeric(cost1) means that if a specific case is not a missing value in cost1 (the fact that a case doesn’t satisfy the first condtion is TRUE), then it will not be changed and will remain the same numeric value as it is. This approach is ok if you are dealing with a couple of variables. However, if you are to recode several variables at the same time, then, this approach is not efficient. We need a faster way.
#quicker way: deal with multiple variables at a time
data %>%
mutate(across(cost1:cost3, ~ case_when(
is.na(.) ~ -999,
TRUE ~ as.numeric(.)
)))
## cost1 cost2 cost3
## 1 2 1 -999
## 2 3 2 -999
## 3 4 3 3
## 4 2 4 2
## 5 -999 5 1
As it is shown here, all we need is across function and specify what variables we are going to recode. Dot (.) refers to the selected variables (i.e., cost1, cost2, cost3). This block of codes ask R to recode missing values across cost1 to cost3 as -999 and for cases that contain information, they will remain the same numeric value as they are.
Task 2: Recode Values for Reverse Scoring Items
It is very common to encounter reverse scoring items in educational research. case_when can also be used to recode these items with ease. Suppose that cost1-cost3 are all measured on a 5-point Likert scale and cost 3 is a reverse scoring item. We woud like to recode this item so that 5 will be 1, and 1 will be 5.
#check our data again
data
## cost1 cost2 cost3
## 1 2 1 NA
## 2 3 2 NA
## 3 4 3 3
## 4 2 4 2
## 5 NA 5 1
#recode one reverse scoring item
data %>%
mutate(cost3 = case_when(
cost3 == 5 ~ 1,
cost3 == 4 ~ 2,
cost3 == 3 ~ 3,
cost3 == 2 ~ 4,
cost3 == 1 ~ 5
))
## cost1 cost2 cost3
## 1 2 1 NA
## 2 3 2 NA
## 3 4 3 3
## 4 2 4 4
## 5 NA 5 5
How about multiple items? Again, across is your friend:
#recorde multiple reverse scoring items at the same time
data %>%
mutate(across(cost1:cost3, ~ case_when(
. == 5 ~ 1,
. == 4 ~ 2,
. == 3 ~ 3,
. == 2 ~ 4,
. == 1 ~ 5
)))
## cost1 cost2 cost3
## 1 4 5 NA
## 2 3 4 NA
## 3 2 3 3
## 4 4 2 4
## 5 NA 1 5
I guess you’ve found that although these are decent solutions, one major drawback is that they simply replace the old values with the new ones. Sometimes, what you want is to create new variables for recorded items to avoid confusion. Fortunately, case_when is also capable of handling that.
#recorde multiple reverse scoring items at the same time and create new variables for recoded items
data %>%
mutate(across(cost1:cost3, ~ case_when(
. == 5 ~ 1,
. == 4 ~ 2,
. == 3 ~ 3,
. == 2 ~ 4,
. == 1 ~ 5
),
.names = "{.col}_r")
)
## cost1 cost2 cost3 cost1_r cost2_r cost3_r
## 1 2 1 NA 4 5 NA
## 2 3 2 NA 3 4 NA
## 3 4 3 3 2 3 3
## 4 2 4 2 4 2 4
## 5 NA 5 1 NA 1 5
By adding .names argument, you will be able to specify the output names. In this case, I append _r to the old variables.
Task 3: Recode Values based on Multiple Conditions
Last, let’s work on a more complex task that requires you to recode values based on multiple conditions. For instance, if we want to create a new dichotomous variable, costHigh (0 and 1), that is based on specific values of cost1, cost2, and cost3. What should we do?
#let's create a new dataset
data2<-data.frame(
cost1 = c(2, 3, 4, 2, 3),
cost2 = c(1, 2, 3, 4, 5),
cost3 = c(2, 4, 3, 2, 4)
)
data2
## cost1 cost2 cost3
## 1 2 1 2
## 2 3 2 4
## 3 4 3 3
## 4 2 4 2
## 5 3 5 4
#create a new variable called costHigh that is based on the values of cost 1, cost2, and cost 3.
#If all the values of cost1-cost3 are equal to or lower than 2, then costHigh will be 0.
#If at least one of the values of cost1-cost3 is higher than 2, then costHigh will be 1.
data2 %>%
mutate(costHigh = case_when(
cost1 <= 2 & cost2 <= 2 & cost3 <= 2 ~ 0,
TRUE ~ 1))
## cost1 cost2 cost3 costHigh
## 1 2 1 2 0
## 2 3 2 4 1
## 3 4 3 3 1
## 4 2 4 2 1
## 5 3 5 4 1
