I have data from an online survey where respondents go through a loop of questions 1-3 times. The survey software (Qualtrics) records this data in multiple columns—that is, Q3.2 in the survey will have columns Q3.2.1.
, Q3.2.2.
, and Q3.2.3.
:
df <- data.frame(
id = 1:10,
time = as.Date('2009-01-01') + 0:9,
Q3.2.1. = rnorm(10, 0, 1),
Q3.2.2. = rnorm(10, 0, 1),
Q3.2.3. = rnorm(10, 0, 1),
Q3.3.1. = rnorm(10, 0, 1),
Q3.3.2. = rnorm(10, 0, 1),
Q3.3.3. = rnorm(10, 0, 1)
)
# Sample data
id time Q3.2.1. Q3.2.2. Q3.2.3. Q3.3.1. Q3.3.2. Q3.3.3.
1 1 2009-01-01 -0.2059165 -0.29177677 -0.7107192 1.52718069 -0.4484351 -1.21550600
2 2 2009-01-02 -0.1981136 -1.19813815 1.1750200 -0.40380049 -1.8376094 1.03588482
3 3 2009-01-03 0.3514795 -0.27425539 1.1171712 -1.02641801 -2.0646661 -0.35353058
...
I want to combine all the QN.N* columns into tidy individual QN.N columns, ultimately ending up with something like this:
id time loop_number Q3.2 Q3.3
1 1 2009-01-01 1 -0.20591649 1.52718069
2 2 2009-01-02 1 -0.19811357 -0.40380049
3 3 2009-01-03 1 0.35147949 -1.02641801
...
11 1 2009-01-01 2 -0.29177677 -0.4484351
12 2 2009-01-02 2 -1.19813815 -1.8376094
13 3 2009-01-03 2 -0.27425539 -2.0646661
...
21 1 2009-01-01 3 -0.71071921 -1.21550600
22 2 2009-01-02 3 1.17501999 1.03588482
23 3 2009-01-03 3 1.11717121 -0.35353058
...
The tidyr
library has the gather()
function, which works great for combining one set of columns:
library(dplyr)
library(tidyr)
library(stringr)
df %>% gather(loop_number, Q3.2, starts_with("Q3.2")) %>%
mutate(loop_number = str_sub(loop_number,-2,-2)) %>%
select(id, time, loop_number, Q3.2)
id time loop_number Q3.2
1 1 2009-01-01 1 -0.20591649
2 2 2009-01-02 1 -0.19811357
3 3 2009-01-03 1 0.35147949
...
29 9 2009-01-09 3 -0.58581232
30 10 2009-01-10 3 -2.33393981
The resultant data frame has 30 rows, as expected (10 individuals, 3 loops each). However, gathering a second set of columns does not work correctly—it successfully makes the two combined columns Q3.2
and Q3.3
, but ends up with 90 rows instead of 30 (all combinations of 10 individuals, 3 loops of Q3.2, and 3 loops of Q3.3; the combinations will increase substantially for each group of columns in the actual data):
df %>% gather(loop_number, Q3.2, starts_with("Q3.2")) %>%
gather(loop_number, Q3.3, starts_with("Q3.3")) %>%
mutate(loop_number = str_sub(loop_number,-2,-2))
id time loop_number Q3.2 Q3.3
1 1 2009-01-01 1 -0.20591649 1.52718069
2 2 2009-01-02 1 -0.19811357 -0.40380049
3 3 2009-01-03 1 0.35147949 -1.02641801
...
89 9 2009-01-09 3 -0.58581232 -0.13187024
90 10 2009-01-10 3 -2.33393981 -0.48502131
Is there a way to use multiple calls to gather()
like this, combining small subsets of columns like this while maintaining the correct number of rows?
This approach seems pretty natural to me:
df %>%
gather(key, value, -id, -time) %>%
extract(key, c("question", "loop_number"), "(Q.\\..)\\.(.)") %>%
spread(question, value)
First gather all question columns, use extract()
to separate into question
and loop_number
, then spread()
question back into the columns.
#> id time loop_number Q3.2 Q3.3
#> 1 1 2009-01-01 1 0.142259203 -0.35842736
#> 2 1 2009-01-01 2 0.061034802 0.79354061
#> 3 1 2009-01-01 3 -0.525686204 -0.67456611
#> 4 2 2009-01-02 1 -1.044461185 -1.19662936
#> 5 2 2009-01-02 2 0.393808163 0.42384717