4

In my dataset respondents are grouped together and there is data available about their age. I want all the people in the same group to have the value of the oldest person in that group.

So my example data looks like this.

      df <- data.frame(groups = c(1,1,1,2,2,2,3,3,3), 
                       age = c(12, 23, 34, 13, 24, 35, 13, 25, 36), 
                       value = c(1, 2, 3, 4, 5, 6, 7, 8, 9))

> df
  groups age value
1      1  12     1
2      1  23     2
3      1  34     3
4      2  13     4
5      2  24     5
6      2  35     6
7      3  13     7
8      3  25     8
9      3  36     9

And I want it to look this this

> df
  groups age value new_value
1      1  12     1         3
2      1  23     2         3
3      1  34     3         3
4      2  13     4         6
5      2  24     5         6
6      2  35     6         6
7      3  13     7         9
8      3  25     8         9
9      3  36     9         9

Any idea how to do this with dplyr?

I have tried something like this, but it doesn't work

df %>% 
        group_by(groups) %>% 
        mutate(new_value = df$value[which.max(df$age)])
r2evans
  • 141,215
  • 6
  • 77
  • 149
Victor Hartman
  • 45
  • 4

1 Answers1

7

Up front, "never" (okay, almost never) use df$ within a dplyr pipe. In this case, df$value[which.max(df$age)] is referencing the original data each time, not the grouped data. Inside each group in this dataset, value is length 3 whereas df$value is length 9.

The only times I feel it is appropriate to use df$ (referencing the original value of the current dataset) inside a pipe is when it is required to look at pre-pipeline data, in absence of any grouping, reordering, or new variables created outside of the currently-saved (pre-pipeline) version of df.

dplyr

library(dplyr)
df %>%
  group_by(groups) %>%
  mutate(new_value = value[which.max(age)]) %>%
  ungroup()
# # A tibble: 9 x 4
#   groups   age value new_value
#    <dbl> <dbl> <dbl>     <dbl>
# 1      1    12     1         3
# 2      1    23     2         3
# 3      1    34     3         3
# 4      2    13     4         6
# 5      2    24     5         6
# 6      2    35     6         6
# 7      3    13     7         9
# 8      3    25     8         9
# 9      3    36     9         9

data.table

library(data.table)
DT <- as.data.table(df)
DT[, new_value := value[which.max(age)], by = .(groups)]

base R

df$new_value <- ave(seq_len(nrow(df)), df$groups,
                    FUN = function(i) df$value[i][which.max(df$age[i])])
df
#   groups age value new_value
# 1      1  12     1         3
# 2      1  23     2         3
# 3      1  34     3         3
# 4      2  13     4         6
# 5      2  24     5         6
# 6      2  35     6         6
# 7      3  13     7         9
# 8      3  25     8         9
# 9      3  36     9         9

The base R approach seems to be the least-elegant-looking solution. I believe that ave is the best approach, but it has many limitations, first being that it only works on one value-object (value) in the absence of others (we need to know age).

r2evans
  • 141,215
  • 6
  • 77
  • 149
  • Wow that was quick, thanks! Looks like I was really close with my own solution... The dplyr method works, but is really slow. Data.table is much quicker. – Victor Hartman Nov 30 '21 at 13:37
  • 1
    For many things, you will likely find that `data.table` is much faster than similar base or `dplyr` solutions; I tend to prefer `data.table` for its speed and referential semantics, but its learning curve is non-trivial. See my top-note edit about why your approach was close but flawed. Glad it helped. – r2evans Nov 30 '21 at 13:42