r - Retain attributes when using gather from tidyr (attributes are not identical)

Question

I have a data frame that needs to be split into two tables to satisfy Codd's 3rd normal form. In a simple case, the original data frame looks something like this:

library(lubridate)
> (df <- data.frame(hh_id = 1:2,
                   income = c(55000, 94000),
                   bday_01 = ymd(c(20150309, 19890211)),
                   bday_02 = ymd(c(19850911, 20000815)),
                   gender_01 = factor(c("M", "F")),
                   gender_02 = factor(c("F", "F"))))

    hh_id income    bday_01    bday_02 gender_01 gender_02
  1     1  55000 2015-03-09 1985-09-11         M         F
  2     2  94000 1989-02-11 2000-08-15         F         F

When I use the gather function, it warns that the attributes are not identical and loses the factor for gender and the lubridate for bday (or other attributes in the real-world example). Is there a nice tidyr solution to avoid the loss of each column's data type?

library(tidyr)
> (person <- df %>% 
      select(hh_id, bday_01:gender_02) %>% 
      gather(key, value, -hh_id) %>%
      separate(key, c("key", "per_num"), sep = "_") %>%
      spread(key, value))

     hh_id per_num       bday gender
   1     1      01 1425859200      M
   2     1      02  495244800      F
   3     2      01  603158400      F
   4     2      02  966297600      F

   Warning message:
   attributes are not identical across measure variables; they will be dropped

> lapply(person, class)

  $hh_id
  [1] "integer"

  $per_num
  [1] "character"

  $bday
  [1] "character"

  $gender
  [1] "character"

I can imagine a way to do it by gathering each set of variables with the same data type separately and then joining all the tables, but there must be a more elegant solution that I'm missing.

Answer 1

You could just convert your dates to character then convert them back to dates at the end:

(person <- df %>% 
      select(hh_id, bday_01:gender_02) %>% 
      mutate_each(funs(as.character), contains('bday')) %>%
      gather(key, value, -hh_id) %>%
      separate(key, c("key", "per_num"), sep = "_") %>%
      spread(key, value) %>%
      mutate(bday=ymd(bday)))

  hh_id per_num       bday gender
1     1      01 2015-03-09      M
2     1      02 1985-09-11      F
3     2      01 1989-02-11      F
4     2      02 2000-08-15      F

Alternatively, if you use Date instead of POSIXct, you could do something like this:

(person <- df %>% 
      select(hh_id, bday_01:gender_02) %>% 
      gather(per_num1, gender, contains('gender'), convert=TRUE) %>%
      gather(per_num2, bday, contains('bday'), convert=TRUE) %>%
      mutate(bday=as.Date(bday)) %>%
      mutate_each(funs(str_extract(., '\d+')), per_num1, per_num2) %>%
      filter(per_num1 == per_num2) %>%
      rename(per_num=per_num1) %>%
      select(-per_num2))

Edit

The warning you're seeing:

Warning: attributes are not identical across measure variables; they will be dropped

arises from gathering the gender columns, which are factors and have different level vectors (see str(df)). If you were to convert the gender columns to character or if you were to synchronize their levels with something like,

df <- mutate(df, gender_02 = factor(gender_02, levels=levels(gender_01)))

then you will see that the warning goes away when you execute

person <- df %>% 
        select(hh_id, bday_01:gender_02) %>% 
        gather(key, value, contains('gender'))