Calculating mean difference between combinations of dates - r

Question

I would like to calculate the pairwise average and median number of days between multiple date variables.
My raw data df might look as following:
id invitation account_date first_order second_order third_order
1 1/1/2016 1/7/2016 1/20/2016 1/22/2016 NA
2 1/1/2016 1/8/2016 1/22/2016 1/23/2016 1/25/2016
3 1/1/2016 1/5/2016 1/20/2016 2/1/2016 NA
4 1/1/2016 1/2/2016 1/18/2016 2/4/2016 2/6/2016
Given that my data are already properly formatted as dates, it's pretty easy to manually calculate the average and median difference for the combinations of dates by first calculating the pairwise differences, e.g.:
id inv_to_act act_to_first act_to_sec act_to_third
1 6 13 2 NA
2 7 14 1 2
3 4 15 12 NA
4 1 16 17 2
And then using base R: mean(df$act_to_first,na.rm=T).
But, I'd like to compute these calculations on several datasets or subsets of the same dataset, so it's not scalable to do each step over and over again. Plus, I'm pretty sure there must be a melt or plyr solution that I haven't figured out.

Answer 1

You could compute the date differences between each pair of dates by looping through the pairs and using difftime:
combos <- combn(tail(names(df), -1), 2)
diffs <- apply(combos, 2, function(x) {
difftime(df[,x[2]], df[,x[1]], units="days")
})
colnames(diffs) <- paste0(combos[1,], "_TO_", combos[2,])
diffs
# invitation_TO_account_date invitation_TO_first_order invitation_TO_second_order invitation_TO_third_order account_date_TO_first_order
# [1,] 6 19 21 NA 13
# [2,] 7 21 22 24 14
# [3,] 4 19 31 NA 15
# [4,] 1 17 34 36 16
# account_date_TO_second_order account_date_TO_third_order first_order_TO_second_order first_order_TO_third_order second_order_TO_third_order
# [1,] 15 NA 2 NA NA
# [2,] 15 17 1 3 2
# [3,] 27 NA 12 NA NA
# [4,] 33 35 17 19 2
After you do that step, you should be able to easily compute the average of each column:
colMeans(diffs, na.rm=TRUE)
# invitation_TO_account_date invitation_TO_first_order invitation_TO_second_order invitation_TO_third_order account_date_TO_first_order
# 4.5 19.0 27.0 30.0 14.5
# account_date_TO_second_order account_date_TO_third_order first_order_TO_second_order first_order_TO_third_order second_order_TO_third_order
# 22.5 26.0 8.0 11.0 2.0
Once you have these functions, you can put them together in a function and apply that function to any input df:
meanDateRanges <- function(df) {
combos <- combn(tail(names(df), -1), 2)
diffs <- apply(combos, 2, function(x) {
difftime(df[,x[2]], df[,x[1]], units="days")
})
colnames(diffs) <- paste0(combos[1,], "_TO_", combos[2,])
colMeans(diffs, na.rm=TRUE)
}
You could run this function on an input data frame with meanDateRanges(df) or on a list of them with lapply(df.list, meanDateRanges).
Data:
df <- structure(list(id = 1:4, invitation = structure(list(sec = c(0,
0, 0, 0), min = c(0L, 0L, 0L, 0L), hour = c(0L, 0L, 0L, 0L),
mday = c(1L, 1L, 1L, 1L), mon = c(0L, 0L, 0L, 0L), year = c(116L,
116L, 116L, 116L), wday = c(5L, 5L, 5L, 5L), yday = c(0L,
0L, 0L, 0L), isdst = c(0L, 0L, 0L, 0L), zone = c("EST", "EST",
"EST", "EST"), gmtoff = c(NA_integer_, NA_integer_, NA_integer_,
NA_integer_)), .Names = c("sec", "min", "hour", "mday", "mon",
"year", "wday", "yday", "isdst", "zone", "gmtoff"), class = c("POSIXlt",
"POSIXt")), account_date = structure(list(sec = c(0, 0, 0, 0),
min = c(0L, 0L, 0L, 0L), hour = c(0L, 0L, 0L, 0L), mday = c(7L,
8L, 5L, 2L), mon = c(0L, 0L, 0L, 0L), year = c(116L, 116L,
116L, 116L), wday = c(4L, 5L, 2L, 6L), yday = c(6L, 7L, 4L,
1L), isdst = c(0L, 0L, 0L, 0L), zone = c("EST", "EST", "EST",
"EST"), gmtoff = c(NA_integer_, NA_integer_, NA_integer_,
NA_integer_)), .Names = c("sec", "min", "hour", "mday", "mon",
"year", "wday", "yday", "isdst", "zone", "gmtoff"), class = c("POSIXlt",
"POSIXt")), first_order = structure(list(sec = c(0, 0, 0, 0),
min = c(0L, 0L, 0L, 0L), hour = c(0L, 0L, 0L, 0L), mday = c(20L,
22L, 20L, 18L), mon = c(0L, 0L, 0L, 0L), year = c(116L, 116L,
116L, 116L), wday = c(3L, 5L, 3L, 1L), yday = c(19L, 21L,
19L, 17L), isdst = c(0L, 0L, 0L, 0L), zone = c("EST", "EST",
"EST", "EST"), gmtoff = c(NA_integer_, NA_integer_, NA_integer_,
NA_integer_)), .Names = c("sec", "min", "hour", "mday", "mon",
"year", "wday", "yday", "isdst", "zone", "gmtoff"), class = c("POSIXlt",
"POSIXt")), second_order = structure(list(sec = c(0, 0, 0, 0),
min = c(0L, 0L, 0L, 0L), hour = c(0L, 0L, 0L, 0L), mday = c(22L,
23L, 1L, 4L), mon = c(0L, 0L, 1L, 1L), year = c(116L, 116L,
116L, 116L), wday = c(5L, 6L, 1L, 4L), yday = c(21L, 22L,
31L, 34L), isdst = c(0L, 0L, 0L, 0L), zone = c("EST", "EST",
"EST", "EST"), gmtoff = c(NA_integer_, NA_integer_, NA_integer_,
NA_integer_)), .Names = c("sec", "min", "hour", "mday", "mon",
"year", "wday", "yday", "isdst", "zone", "gmtoff"), class = c("POSIXlt",
"POSIXt")), third_order = structure(list(sec = c(NA, 0, NA, 0
), min = c(NA, 0L, NA, 0L), hour = c(NA, 0L, NA, 0L), mday = c(NA,
25L, NA, 6L), mon = c(NA, 0L, NA, 1L), year = c(NA, 116L, NA,
116L), wday = c(NA, 1L, NA, 6L), yday = c(NA, 24L, NA, 36L),
isdst = c(-1L, 0L, -1L, 0L), zone = c("", "EST", "", "EST"
), gmtoff = c(NA_integer_, NA_integer_, NA_integer_, NA_integer_
)), .Names = c("sec", "min", "hour", "mday", "mon", "year",
"wday", "yday", "isdst", "zone", "gmtoff"), class = c("POSIXlt",
"POSIXt"))), .Names = c("id", "invitation", "account_date", "first_order",
"second_order", "third_order"), row.names = c(NA, -4L), class = "data.frame")