I have a data set that not every subject’s observations were observed at the exact same time points, but I want to turn it in to a data set that every one’s observations were observed at the exact same time points (so that I can use it in SAS proc traj).
For example, suppose I have dataset "m":
id <- c(1,1,1,1,2,2,3,3,3)
age <- c(2,3,4,5,3,6,2,5,8)
IQ <- c(3,4,5,4,6,5,3,8,10)
m <- data.frame(id,age,IQ)
> m
id age IQ
1 1 2 3
2 1 3 4
3 1 4 5
4 1 5 4
5 2 3 6
6 2 6 5
7 3 2 3
8 3 5 8
9 3 8 10
> unique(age)
[1] 2 3 4 5 6 8
I want to turn m to m2. But I can only do that manually.
id2 <- c(1,1,1,1,1,1,2,2,2,2,2,2,3,3,3,3,3,3)
age2 <- c(2,3,4,5,6,8,2,3,4,5,6,8,2,3,4,5,6,8)
IQ2 <- c(3,4,5,4,NA,NA,6,5,NA,NA,NA,NA,3,8,10,NA,NA,NA)
m2 <- data.frame(id2,age2,IQ2)
m2
> m2
id2 age2 IQ2
1 1 2 3
2 1 3 4
3 1 4 5
4 1 5 4
5 1 6 NA
6 1 8 NA
7 2 2 6
8 2 3 5
9 2 4 NA
10 2 5 NA
11 2 6 NA
12 2 8 NA
13 3 2 3
14 3 3 8
15 3 4 10
16 3 5 NA
17 3 6 NA
18 3 8 NA
Does anyone know a smarter way to do this?
Using tidyr, this is a one liner. You use the complete function, which creates rows with each combination of the columns passed to it, filling the rest of the rows with NA:
library(tidyr)
complete(m, id, age)
Source: local data frame [18 x 3]
id age IQ
(dbl) (dbl) (dbl)
1 1 2 3
2 1 3 4
3 1 4 5
4 1 5 4
5 1 6 NA
6 1 8 NA
7 2 2 NA
8 2 3 6
9 2 4 NA
10 2 5 NA
11 2 6 5
12 2 8 NA
13 3 2 3
14 3 3 NA
15 3 4 NA
16 3 5 8
17 3 6 NA
18 3 8 10
We could do this using data.table. We convert the data.frame to data.table (setDT(m)), set the key columns (setkey) and join with the cross-join of unique elements of 'id' and 'age'
library(data.table)
setkey(setDT(m), id, age)[CJ(unique(id), unique(age))]
# id age IQ
# 1: 1 2 3
# 2: 1 3 4
# 3: 1 4 5
# 4: 1 5 4
# 5: 1 6 NA
# 6: 1 8 NA
# 7: 2 2 NA
# 8: 2 3 6
# 9: 2 4 NA
#10: 2 5 NA
#11: 2 6 5
#12: 2 8 NA
#13: 3 2 3
#14: 3 3 NA
#15: 3 4 NA
#16: 3 5 8
#17: 3 6 NA
#18: 3 8 10
In the devel version i.e. v1.9.5, we can use unique=TRUE within CJ (from @Frank's comment)
setDT(m, key=c('id', 'age'))[CJ(id, age, unique=TRUE)]
Benchmarks
set.seed(24)
m1 <- data.frame(id=rep(1:10000, each=10), age=sample(2:400, 10000*10,
replace=TRUE), IQ=rnorm(10000*10))
system.time(res1 <- complete(m1, id, age))
# user system elapsed
#18.888 0.000 16.258
system.time({ DT <- as.data.table(m1)
res2 <- setkey(DT, id, age)[CJ(unique(id), unique(age))]})
# user system elapsed
# 0.000 0.000 0.279
library(microbenchmark)
jeremy <- function() complete(m1, id, age)
akrun <- function() {DT <- as.data.table(m1)
setkey(DT, id, age)[CJ(unique(id), unique(age))]}
microbenchmark(jeremy(), akrun(), times=20L, unit='relative')
#Unit: relative
# expr min lq mean median uq max neval cld
#jeremy() 24.95042 30.84234 17.138 23.09175 12.16891 8.305394 20 b
# akrun() 1.00000 1.00000 1.000 1.00000 1.00000 1.000000 20 a
来源：https://stackoverflow.com/questions/32654706/how-to-create-missing-value-for-repeated-measurement-data