dplyr: passing a grouped tibble to a custom function

Question

(The following scenario simplifies my actual situation)
My data comes from villages, and I would like to summarize an outcome variable by a village variable.

> data
   village     A     Z      Y 
     <chr> <int> <int>   <dbl> 
 1       a     1     1   500     
 2       a     1     1   400     
 3       a     1     0   800  
 4       b     1     0   300  
 5       b     1     1   700  

For example, I would like to calculate the mean of Y only using Z==z by villages. In this case, I want to have (500 + 400)/2 = 450 for village "a" and 700 for village "b".

Please note that the actual situation is more complicated and I cannot directly use this answer, but the point is I need to pass a grouped tibble and a global variable (z) to my function.

z <- 1 # z takes 0 or 1
data %>%
    group_by(village) %>% # grouping by village
    summarize(Y_village = Y_hat_village(., z)) # pass a part of tibble and a global variable

Y_hat_village <- function(data_village, z){
    # This function takes a part of tibble (`data_village`) and a variable `z`
    # Calculate the mean for a specific z in a village
    data_z <- data_village %>% filter(Z==get("z"))
    return(mean(data_z$Y))
}

However, I found . passes entire tibble and the code above returns the same values for all groups.

Answer 1

There are a couple things you can simplify. One is in your function: since you're passing in a value z to the function, you don't need to use get("z"). You have a z in the global environment that you pass in; or, more safely, assign your z value to a variable with some other name so you don't run into scoping issues, and pass that in to the function. In this case, I'm calling it z_val.

library(tidyverse)

z_val <- 1

Y_hat_village2 <- function(data, z) {
  data_z <- data %>% filter(Z == z)
  return(mean(data_z$Y))
}

You can make the function call on each group using do, which will get you a list-column, and then unnesting that column. Again note that I'm passing in the variable z_val to the argument z.

df %>%
  group_by(village) %>%
  do(y_hat = Y_hat_village2(., z = z_val)) %>%
  unnest()
#> # A tibble: 2 x 2
#>   village y_hat
#>   <chr>   <dbl>
#> 1 a         450
#> 2 b         700

However, do is being deprecated in favor of purrr::map, which I am still having trouble getting the hang of. In this case, you can group and nest, which gives a column of data frames called data, then map over that column and again supply z = z_val. When you unnest the y_hat column, you still have the original data as a nested column, since you wanted access to the rest of the columns still.

df %>%
  group_by(village) %>%
  nest() %>%
  mutate(y_hat = map(data, ~Y_hat_village2(., z = z_val))) %>%
  unnest(y_hat)
#> # A tibble: 2 x 3
#>   village data             y_hat
#>   <chr>   <list>           <dbl>
#> 1 a       <tibble [3 × 3]>   450
#> 2 b       <tibble [2 × 3]>   700

Just to check that everything works okay, I also passed in z = 0 to check for 1. scoping issues, and 2. that other values of z work.

df %>%
  group_by(village) %>%
  nest() %>%
  mutate(y_hat = map(data, ~Y_hat_village2(., z = 0))) %>%
  unnest(y_hat)
#> # A tibble: 2 x 3
#>   village data             y_hat
#>   <chr>   <list>           <dbl>
#> 1 a       <tibble [3 × 3]>   800
#> 2 b       <tibble [2 × 3]>   300

Answer 2

As an extension/modification to @patL's answer, you can also wrap the tidyverse solution within purrr:map to return a list of two tibbles, one for each z value:

z <- c(0, 1);
map(z, ~df %>% filter(Z == .x) %>% group_by(village) %>% summarise(Y.mean = mean(Y)))
#[[1]]
## A tibble: 2 x 2
#  village Y.mean
#  <fct>    <dbl>
#1 a         800.
#2 b         300.
#
#[[2]]
## A tibble: 2 x 2
#  village Y.mean
#  <fct>    <dbl>
#1 a         450.
#2 b         700.

---

Sample data

df <- read.table(text =
    "  village     A     Z      Y
 1       a     1     1   500
 2       a     1     1   400
 3       a     1     0   800
 4       b     1     0   300
 5       b     1     1   700  ", header = T)