Summarise dataframe with correlation of variables based on multiple groups-CodePudding

I'm working with a dataset that has two level grouping. Here I give an example:

set.seed(123)
example=data.frame(
  id = c(rep(1,20),rep(2,20)),  # grouping 
  Grp = rep(c(rep('A',10),rep('B',10)),2),  # grouping
  target = c(rep(1:10,2), rep(20:11,2)),
  var1 = sample(1:100,40,replace=TRUE),
  var2 =sample(1:100,40,replace=TRUE)
)

In this case, the data is grouped by id and by Grp. I want to calculate the correlation of target with var1 and var2. However, I don't know which is the most efficient way to apply this using a tidy approach and based on the groups.

I tried with dplyr approach. Like using:

example %>% group_by(id,Grp) %>% 
  summarise(cor(target,c(var1,var2)))  # length error

Or even creating a custom function and applying it. But this last only summarise all the data without grouping:

corr_analisis_e = function(df){
  return( cor(df[,'target'] , df[,c('var1','var2')]) )
  
}

example %>% group_by(id,Grp) %>%  corr_analisis_e() # get all the data at once

As output, I would expect to get something like a matrix or dataframe of 4 rows and 2 columns, where each row is a group (id and Grp) and columns are var1 and var2. Every value the result from the cor() method.

CodePudding user response：

example %>% 
  group_by(id, Grp) %>% 
  summarise(across(c(var1, var2), ~ cor(.x, target)))

# A tibble: 4 x 4
# Groups:   id [2]
     id Grp     var1   var2
  <dbl> <chr>  <dbl>  <dbl>
1     1 A     -0.400  0.532
2     1 B     -0.133 -0.187
3     2 A     -0.655 -0.103
4     2 B     -0.580  0.445

The grouping columns can then be removed with %>% ungroup %>% select(var1:var2).