How to use the spread function properly in tidyr

ayk picture ayk · Jan 8, 2016 · Viewed 15.2k times · Source

How do I change the following table from:

Type    Name    Answer     n
TypeA   Apple   Yes        5
TypeA   Apple   No        10
TypeA   Apple   DK         8
TypeA   Apple   NA        20
TypeA   Orange  Yes        6
TypeA   Orange  No        11
TypeA   Orange  DK         8
TypeA   Orange  NA        23

Change to:

Type    Name    Yes   No   DK   NA  
TypeA   Apple   5     10   8    20
TypeA   Orange  6     11   8    23

I used the following codes to get the first table.

df_1 <- 
  df %>% 
  group_by(Type, Name, Answer) %>% 
  tally()  

Then I tried to use the spread command to get to the 2nd table, but I got the following error message:

"Error: All columns must be named"

df_2 <- spread(df_1, Answer)

Answer

Nicholas G Reich picture Nicholas G Reich · Jan 15, 2016

Following on the comment from ayk, I'm providing an example. It looks to me like when you have a data_frame with a column of either a factor or character class that has values of NA, this cannot be spread without either removing them or re-classifying the data. This is specific to a data_frame (note the dplyr class with the underscore in the name), as this works in my example when you have values of NA in a data.frame. For example, a slightly modified version of the example above:

Here is the dataframe

library(dplyr)
library(tidyr)
df_1 <- data_frame(Type = c("TypeA", "TypeA", "TypeB", "TypeB"),
                   Answer = c("Yes", "No", NA, "No"),
                   n = 1:4)
df_1

Which gives a data_frame that looks like this

Source: local data frame [4 x 3]

   Type Answer     n
  (chr)  (chr) (int)
1 TypeA    Yes     1
2 TypeA     No     2
3 TypeB     NA     3
4 TypeB     No     4

Then, when we try to tidy it, we get an error message:

df_1 %>% spread(key=Answer, value=n)
Error: All columns must be named

But if we remove the NA's then it 'works':

df_1 %>%
    filter(!is.na(Answer)) %>%
    spread(key=Answer, value=n)
Source: local data frame [2 x 3]

   Type    No   Yes
  (chr) (int) (int)
1 TypeA     2     1
2 TypeB     4    NA

However, removing the NAs may not give you the desired result: i.e. you might want those to be included in your tidied table. You could modify the data directly to change the NAs to a more descriptive value. Alternatively, you could change your data to a data.frame and then it spreads just fine:

as.data.frame(df_1) %>% spread(key=Answer, value=n)
   Type No Yes NA
1 TypeA  2   1 NA
2 TypeB  4  NA  3