CODEWITHSUNDEEP
rdplyr
kaz7
kaz7

Reputation: 13

Selecting n random groups from grouped data in R

I have grouped data composed of students clustered within 160 schools. I would like to take a random sample of 30 schools from that dataset. I hard-coded a solution (see below), but is there a wrapper function or quicker way to do this in R? Kind of like sample_n() or top_n(), but those return n observations per group, whereas I want 100% of the observations from n groups.

# First, some example data. Each row represents one student in a given school, and that student's favourite fruit.

df <- tribble(
    ~school_id, ~favourite_fruit,
    #----------#---------------
    1, "apple",
    1, "banana",
    2, "kiwi",
    2, "tomato",
    3, "strawberry",
    3, "cherry",
    4, "orange",
    4, "lime"
)

# My hard-coded solution

school_vector <- df %>% 
    group_by(school_id) %>% 
    select(school_id) %>% 
    count() %>% 
    ungroup() %>% 
    select(school_id) %>% 
    sample_n(2)

df_subset <- df %>% 
    filter(school_id %in% school_vector$school_id) %>% 
    as_tibble()

Upvotes: 1

Views: 75

Answers (1)

IceCreamToucan
IceCreamToucan

Reputation: 28705

You can create a sample of school_ids within filter and use that with your current %in% logic

df %>% 
  filter(school_id %in% sample(unique(school_id), 2))
# # A tibble: 4 x 2
#   school_id favourite_fruit
#       <dbl> <chr>          
# 1         3 strawberry     
# 2         3 cherry         
# 3         4 orange         
# 4         4 lime

As a function:

group_samp <- function(df, group_var, n){
  df %>% 
    filter({{group_var}} %in% sample(unique({{group_var}}), n))
}

df %>% 
  group_samp(school_id, 2)
# # A tibble: 4 x 2
#   school_id favourite_fruit
#       <dbl> <chr>          
# 1         1 apple          
# 2         1 banana         
# 3         2 kiwi           
# 4         2 tomato

Upvotes: 4

Related Questions