Return multiple values based on duplicate max values in another column

Question

I want to summarise a df by group using multiple functions. Replication data:

library(dplyr)
df1 <- data.frame(a=c('a', 'a', 'b', 'b', 'c', 'c'), b=c(1,NA,3,2,2,1), c=c(1,3,5,5,2,4))

One of these is a custom function that asks for the value of df1$b when max(df1$c) in each group (df1$a). When the result is NA, it should return the value for df1$b for the second-highest value of df1$c. The following works:

namax <- function(x,y) ifelse(is.na(y[x==max(x)] & length(x)>1),
                              y[x==sort(x,partial=length(x)-1)[length(x)-1]], y[x==max(x)])

I then try to summarise df1 using:

df2 <- df1 %>%
  dplyr::group_by(a) %>%
  summarise(meanc = mean(c),
            maxc = namax(c,b))

Which returns the following, because for df$a == 'b' the max value of df1$c occurs twice for different values of df1$b.

Error: Column 'maxc' must be length 1 (a summary value), not 2

Is there an elegant solution through which dplyr returns both values, while simultaneously executing the other call to summarise() (e.g. by adding do() to the call to group_by)? In my applied case I am trying to run several different calls to summarise, aside from the one using the namax function.

Sotos · Accepted Answer · 2020-01-24 10:29:04Z

You can put the values in a list, i.e.

library(dplyr)

 df1 %>%
     group_by(a) %>%
     summarise(meanc = mean(c),
               maxc = list(namax(c, b)))

# A tibble: 3 x 3
#  a     meanc maxc     
#  <fct> <dbl> <list>   
#1 a         2 <dbl [1]>
#2 b         5 <dbl [2]>
#3 c         3 <dbl [1]>

You can use unnest() to expand,

df1 %>%
     group_by(a) %>%
     summarise(meanc = mean(c),
               maxc = list(namax(c, b))) %>% 
     unnest()

# A tibble: 4 x 3
#  a     meanc  maxc
#  <fct> <dbl> <dbl>
#1 a         2     1
#2 b         5     3
#3 b         5     2
#4 c         3     1

Collectives™ on Stack Overflow

Return multiple values based on duplicate max values in another column

1 Answer 1

Comments

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Related