CODEWITHSUNDEEP
rtidyrtidyverse
Joy
Joy

Reputation: 769

tidyr::gather na.rm with missing data

Let's say I have multiple columns in a data frame that measure the same concept, but in different methods (e.g. there are multiple kinds of IQ tests, and students could have any one of them, or none at all). I want to combine the various methods into a single column (obvious use case for tidyr).

If the data is something like this:

mydata <- data.frame(ID = 55:64, 
                 age = c(12, 12, 14, 11, 20, 10, 13, 15, 18, 17),
                 Test1 = c(100, 90, 88, 115, NA, NA, NA, NA, NA, NA),
                 Test2 = c(NA, NA, NA, NA, 100, 120, NA, NA, NA, NA),
                 Test3 = c( NA, NA, NA, NA, NA, NA, 110, NA, 85, 150))

I would naturally want to perform something like this (note that I use na.rm = TRUE in order to not have the many many NA's in my data set get their own rows):

library(tidyr)
tests <- gather(mydata, key=IQSource, value=IQValue, c(Test1, Test2, Test3), na.rm = TRUE)
tests

Giving me:

ID age IQSource IQValue 1 55 12 Test1 100 2 56 12 Test1 90 3 57 14 Test1 88 4 58 11 Test1 115 15 59 20 Test2 100 16 60 10 Test2 120 27 61 13 Test3 110 29 63 18 Test3 85 30 64 17 Test3 150

The problem is that I have a student (ID=62) that doesn't have any IQ scores in any of the three, and I don't want to lose her other data (the data in the ID and age columns).

Is there a way to distinguish, in tidyr, that yes, I want to remove NA's where I do have data in at least one column I'm gathering, yet at the same time want to prevent data loss when all of the columns to gather are NA?)

Upvotes: 3

Views: 1953

Answers (3)

eipi10
eipi10

Reputation: 93851

If students can each have only one IQ test...

library(tidyverse)

mydata %>%
  gather(key=IQSource, value=IQValue, Test1:Test3) %>%
  group_by(ID) %>%
  arrange(IQValue) %>%
  slice(1)

      ID   age IQSource IQValue
 1    55    12    Test1     100
 2    56    12    Test1      90
 3    57    14    Test1      88
 4    58    11    Test1     115
 5    59    20    Test2     100
 6    60    10    Test2     120
 7    61    13    Test3     110
 8    62    15    Test1      NA
 9    63    18    Test3      85
10    64    17    Test3     150

If students can each have multiple IQ tests...

mydata %>%
  # Add an ID with multiple IQ tests
  bind_rows(data.frame(ID=65, age=13, Test1=100, Test2=100, Test3=NA)) %>%
  gather(key=IQSource, value=IQValue, Test1:Test3) %>%
  group_by(ID) %>%
  filter(!is.na(IQValue) | all(is.na(IQValue))) %>%
  filter(all(!is.na(IQValue)) | !duplicated(IQValue)) %>%
  arrange(ID, IQSource)

      ID   age IQSource IQValue
 1    55    12    Test1     100
 2    56    12    Test1      90
 3    57    14    Test1      88
 4    58    11    Test1     115
 5    59    20    Test2     100
 6    60    10    Test2     120
 7    61    13    Test3     110
 8    62    15    Test1      NA
 9    63    18    Test3      85
10    64    17    Test3     150
11    65    13    Test1     100
12    65    13    Test2     100

Upvotes: 1

Thomas K
Thomas K

Reputation: 3311

I did'nt find a direct solution, but you could right_join back the original data.frame and then deselect all columns which you don't need. 

library(tidyr)
library(dplyr)

mydata %>% 
  gather(key, val, Test1:Test3, na.rm = T) %>%
  right_join(mydata) %>% 
  select(-contains("Test"))
#> Joining, by = c("ID", "age")
#>    ID age   key val
#> 1  55  12 Test1 100
#> 2  56  12 Test1  90
#> 3  57  14 Test1  88
#> 4  58  11 Test1 115
#> 5  59  20 Test2 100
#> 6  60  10 Test2 120
#> 7  61  13 Test3 110
#> 8  62  15  <NA>  NA
#> 9  63  18 Test3  85
#> 10 64  17 Test3 150

Alternatively, you could of course first create a data.frame with all the variables you want to keep and then join it:

id_data <- select(mydata, ID, age)

mydata %>% 
  gather(key, val, Test1:Test3, na.rm = T) %>%
  right_join(id_data)

Upvotes: 3

svenhalvorson
svenhalvorson

Reputation: 1080

I think this will do the trick for you:

    # make another data frame which has just ID and whether or not they missed all 3 tests
    missing = mydata %>% 
      mutate(allNA = is.na(Test1) & is.na(Test2) & is.na(Test3)) %>%
      select(ID, allNA)

    # Gather and keep NAs  
    tests <- gather(mydata, key=IQSource, value=IQValue, c(Test1, Test2, Test3), na.rm = FALSE)

    # Keep the rows that have a IQValue or missed all tests
    tests = left_join(tests, missing) %>% 
      filter(!is.na(IQValue) | allNA)
    # Remove duplicated rows of individuals who missed all exams
    tests = tests[!is.na(tests$IQValue) | !duplicated(tests[["ID"]]), ]

Upvotes: 1

Related Questions