Clean way to calculate both group and overall statistics - r

Question

I would like like to calculate the median not only for different groups of my data, but also the median over all groups and store the result in a single data.frame. While accomplishing each of these tasks separately is easy, I have not found a clean way to do both at the same time.
Right now, what I'm doing is calculate both statistics separately; then join the results; then tidy the data if necessary. Here's an example of what this may look like if I wanted to know the median delay per day and per month:
library(dplyr)
library(hflights)
data(hflights)
# Calculate both statistics separately
per_day <- hflights %>%
group_by(Year, Month, DayofMonth) %>%
summarise(Delay = mean(ArrDelay, na.rm = TRUE)) %>%
mutate(Interval = "Daily")
per_month <- hflights %>%
group_by(Year, Month) %>%
summarise(Delay = mean(ArrDelay, na.rm = TRUE)) %>%
mutate(Interval = "Monthly", DayofMonth = NA)
# Join into a single data.frame
my_summary <- full_join(per_day, per_month,
by = c("Year", "Month", "DayofMonth", "Interval", "Delay"))
my_summary
# Source: local data frame [377 x 5]
# Groups: Year, Month
#
# Year Month DayofMonth Delay Interval
# 1 2011 1 1 10.067642 Daily
# 2 2011 1 2 10.509745 Daily
# 3 2011 1 3 6.038627 Daily
# 4 2011 1 4 7.970740 Daily
# 5 2011 1 5 4.172650 Daily
# 6 2011 1 6 6.069909 Daily
# 7 2011 1 7 3.907295 Daily
# 8 2011 1 8 3.070140 Daily
# 9 2011 1 9 17.254325 Daily
# 10 2011 1 10 11.040388 Daily
# .. ... ... ... ... ...
Are there better ways to do this?
(Note that in many cases one could easily progressively roll up summaries as pointed out in the Introduction to dplyr. However, this doesn't work for statistics like median, mean etc.)

Answer 1

As a one-off table. This is fairly straightforward in data.table:
require(data.table)
setDT(hflights)[,{
mo_del <- mean(ArrDelay,na.rm=TRUE)
.SD[,.(DailyDelay = mean(ArrDelay,na.rm=TRUE),MonthlyDelay = mo_del),by=DayofMonth]
},by=.(Year,Month)]
# Year Month DayofMonth DailyDelay MonthlyDelay
# 1: 2011 1 1 10.0676417 4.926065
# 2: 2011 1 2 10.5097451 4.926065
# 3: 2011 1 3 6.0386266 4.926065
# 4: 2011 1 4 7.9707401 4.926065
# 5: 2011 1 5 4.1726496 4.926065
# ---
# 361: 2011 12 14 1.0293610 5.013244
# 362: 2011 12 17 -0.1049822 5.013244
# 363: 2011 12 24 -4.1457490 5.013244
# 364: 2011 12 25 -2.2976827 5.013244
# 365: 2011 12 31 46.4846491 5.013244
How it works. The basic syntax is DT[i,j,by].
With by=.(Year,Month), all operations in j are done per "by group."
We can nest another "by group" using the data.table of the current Subset of Data, .SD.
To return columns in j we use .(colname1=col1,colname2=col2,...).
Creating new variables. Alternately, we could create new variables in hflights using := in j.
hflights[,DailyDelay := mean(ArrDelay,na.rm=TRUE),.(Year,Month,DayofMonth)]
hflights[,MonthlyDelay := mean(ArrDelay,na.rm=TRUE),.(Year,Month)]
Then we can view the summary table:
hflights[,.GRP,.(Year,Month,DayofMonth,DailyDelay,MonthlyDelay)]
# Year Month DayofMonth DailyDelay MonthlyDelay .GRP
# 1: 2011 1 1 10.0676417 4.926065 1
# 2: 2011 1 2 10.5097451 4.926065 2
# 3: 2011 1 3 6.0386266 4.926065 3
# 4: 2011 1 4 7.9707401 4.926065 4
# 5: 2011 1 5 4.1726496 4.926065 5
# ---
# 361: 2011 12 14 1.0293610 5.013244 361
# 362: 2011 12 17 -0.1049822 5.013244 362
# 363: 2011 12 24 -4.1457490 5.013244 363
# 364: 2011 12 25 -2.2976827 5.013244 364
# 365: 2011 12 31 46.4846491 5.013244 365
(Something needed to be put in j here, so I used the "by group" code, .GRP.)