Converting a dataframe from "wide" format to "long" format in R

Question

I have the following dataframe:

 df = data.frame(A_1 = c(1,2,3), A_2 = c(4,5,6), A_3 = c(7,8,9), B_1 = c(10, 11, 12), B_2 = c(13, 14, 15), B_3 = c(16, 17, 18))

 #> df
 #  A_1 A_2 A_3 B_1 B_2 B_3
 #1   1   4   7  10  13  16
 #2   2   5   8  11  14  17
 #3   3   6   9  12  15  18

The column names contain both a letter and a number. The letter refers to a specific variable (e.g A is a factor, B is a factor), while the numbers in the column names, refer to individuals. In other words, each individual has values for A and B: A_1 and B_1 are columns for Individual 1, and A_2, B_2 are columns for Individual 2, etc.

I would like to achieve the following result: note that all the "A" columns are merge into one "A" column, and the same goes for the "B" columns, etc. :

   A  B
 # 1 10
 # 2 11
 # 3 12
 # 4 13
 # 5 14
 # 6 15
 # 7 16
 # 8 17
 # 9 18

Is there any easy way to achieve that? Please note that my real dataframe contains more than 20 distinct letter columns (A, B, C, ...), each letter having three subcolumns (e.g: A_1, A_2, A_3).

Thanks!!

Answer 1

This is known as "reshaping" your data from a "wide" format to a "long" format. In base R, one tool is reshape, but you'll need an "id" variable first:

reshape(df, direction = "long", varying = names(df), sep = "_")
#     time A  B id
# 1.1    1 1 10  1
# 2.1    1 2 11  2
# 3.1    1 3 12  3
# 1.2    2 4 13  1
# 2.2    2 5 14  2
# 3.2    2 6 15  3
# 1.3    3 7 16  1
# 2.3    3 8 17  2
# 3.3    3 9 18  3

You can drop the other columns if required.

---

For fun, here's another approach, using the "reshape2" package (start with your original sample data):

library(reshape2)
dfL <- melt(as.matrix(df))
dfL <- cbind(dfL, colsplit(dfL$Var2, "_", c("Factor", "Individual")))
dcast(dfL, Individual + Var1 ~ Factor, value.var="value")
#   Individual Var1 A  B
# 1          1    1 1 10
# 2          1    2 2 11
# 3          1    3 3 12
# 4          2    1 4 13
# 5          2    2 5 14
# 6          2    3 6 15
# 7          3    1 7 16
# 8          3    2 8 17
# 9          3    3 9 18

---

If you live on the bleeding edge, "data.table" version 1.8.11 has now implemented "melt" and "dcast". I haven't played much with it yet, but it is pretty straightforward too. Again, as with all the solutions I've provided so far, an "id" is needed.

library(reshape2)
library(data.table)
packageVersion("data.table") ## Must be at least 1.8.11 to work
# [1] ‘1.8.11’

DT <- data.table(cbind(id = sequence(nrow(df)), df))
DTL <- melt(DT, id.vars="id")
DTL[, c("Fac", "Ind") := colsplit(variable, "_", c("Fac", "Ind"))]
dcast.data.table(DTL, Ind + id ~ Fac)
#    Ind id A  B
# 1:   1  1 1 10
# 2:   1  2 2 11
# 3:   1  3 3 12
# 4:   2  1 4 13
# 5:   2  2 5 14
# 6:   2  3 6 15
# 7:   3  1 7 16
# 8:   3  2 8 17
# 9:   3  3 9 18

---

Update

Another option is to use merged.stack from my "splitstackshape" package. It works nicely if you also use as.data.table(df, keep.rownames = TRUE), which would create the equivalent of the data.table(cbind(id = sequence(nrow(df)), df)) step in the "data.table" approach.

library(splitstackshape)
merged.stack(as.data.table(df, keep.rownames = TRUE), 
             var.stubs = c("A", "B"), sep = "_")
#    rn .time_1 A  B
# 1:  1       1 1 10
# 2:  1       2 4 13
# 3:  1       3 7 16
# 4:  2       1 2 11
# 5:  2       2 5 14
# 6:  2       3 8 17
# 7:  3       1 3 12
# 8:  3       2 6 15
# 9:  3       3 9 18

And for fairness/completeness, here's an approach with "tidyr" + "dplyr".

library(tidyr)
library(dplyr)
df %>%
  gather(var, value, A_1:B_3) %>%
  separate(var, c("var", "time")) %>%
  group_by(var, time) %>%
  mutate(grp = sequence(n())) %>%
  ungroup() %>%
  spread(var, value)
# Source: local data frame [9 x 4]
# 
#   time grp A  B
# 1    1   1 1 10
# 2    1   2 2 11
# 3    1   3 3 12
# 4    2   1 4 13
# 5    2   2 5 14
# 6    2   3 6 15
# 7    3   1 7 16
# 8    3   2 8 17
# 9    3   3 9 18