Transform frequencies in factor levels into percentages

Question

Consider the following data frame:

x <-c(rep (c ("s1", "s2", "s3"),each=5 ))
y <- c(rep(c("a", "b", "c", "d", "e"), 3) )               
z<-c(1:15)    

x_name <- "dimensions"
y_name <- "aspects"
z_name<-"value"
df <- data.frame(x,y,z)
names(df) <- c(x_name,y_name, z_name)

How I can calculate and add a new column of percentages for each factor level? In this case percentage of 'value' for each 'dimension' across 'aspects' so that I have something like this:

To illustrate: 7 in percentage represents the percentage of value 1 taken from the total of values under aspects corresponding to dimension s1 (15 in this case) etc.

I searched the forum, but the answers in creating percentages go only across one column and not two. I finally produced this column, by first calculating in a separate df total value in aspects across dimensions, using aggregate. Then I merged the initial df with the new one with merge and created a percentage column by dividing the value column to total value. But I feel that this is very clumsy. Is there a simpler way to do this?

Answer 1

In base R you can use ave to do these sorts of grouped calculations:

df$percentage <- ave(df$value, df$dimensions, FUN=function(x) 100*x/sum(x))
df
#    dimensions aspects value percentage
# 1          s1       a     1   6.666667
# 2          s1       b     2  13.333333
# 3          s1       c     3  20.000000
# 4          s1       d     4  26.666667
# 5          s1       e     5  33.333333
# 6          s2       a     6  15.000000
# 7          s2       b     7  17.500000
# 8          s2       c     8  20.000000
# 9          s2       d     9  22.500000
# 10         s2       e    10  25.000000
# 11         s3       a    11  16.923077
# 12         s3       b    12  18.461538
# 13         s3       c    13  20.000000
# 14         s3       d    14  21.538462
# 15         s3       e    15  23.076923

In dplyr you could use group_by and mutate:

library(dplyr)
df %>% group_by(dimensions) %>% mutate(percentage=100*value/sum(value))
# Source: local data frame [15 x 4]
# Groups: dimensions [3]
# 
#    dimensions aspects value percentage
#        (fctr)  (fctr) (int)      (dbl)
# 1          s1       a     1   6.666667
# 2          s1       b     2  13.333333
# 3          s1       c     3  20.000000
# 4          s1       d     4  26.666667
# 5          s1       e     5  33.333333
# 6          s2       a     6  15.000000
# 7          s2       b     7  17.500000
# 8          s2       c     8  20.000000
# 9          s2       d     9  22.500000
# 10         s2       e    10  25.000000
# 11         s3       a    11  16.923077
# 12         s3       b    12  18.461538
# 13         s3       c    13  20.000000
# 14         s3       d    14  21.538462
# 15         s3       e    15  23.076923

Any sort of rounding desired could be performed by wrapping the percentage calculation in the round function and passing the desired precision.

Answer 2

You can use round and a fast data.table approach:

library(data.table)
setDT(df)[,percentage:=round(100*value/sum(value)), dimensions][]

#   dimensions aspects value percentage
# 1:         s1       a     1          7
# 2:         s1       b     2         13
# 3:         s1       c     3         20
# 4:         s1       d     4         27
# 5:         s1       e     5         33
# 6:         s2       a     6         15
# 7:         s2       b     7         18
# 8:         s2       c     8         20
# 9:         s2       d     9         22
#10:         s2       e    10         25
#11:         s3       a    11         17
#12:         s3       b    12         18
#13:         s3       c    13         20
#14:         s3       d    14         22
#15:         s3       e    15         23