Perform t tests within dplyr groups using summarise_all [R]

user_11825717 0
Viewed 122

Let's say I want to compare the price of apples and oranges in each country in two different currencies: USA and BTC.

USA ~ fruit for each country
BTC ~ fruit for each country

library(tidyverse)

prices <- tibble(
  country = c(rep("USA", 6), rep("Spain", 6), rep("Korea", 6)),
  fruit = rep(c("apples", "apples", "apples", "oranges", "oranges", "oranges"), 3),
  price_USA = rnorm(18),
  price_BTC = rnorm(18)
)

prices %>% 
  group_by(country) %>% 
  summarise(
    pval_USA = t.test(price_USA ~ fruit)$p.value
    pval_BTC = t.test(price_BTC ~ fruit)$p.value
  )

Now let's say there are many columns and I want to use summarise_all instead of naming each column. Is there a way to perform a t-test within each group (country) and on each column (price_USA, price_BTC) using the dplyr::summarise_all function? The approaches I've tried so far have been giving me errors.

prices %>% 
  group_by(country) %>% 
  summarise_at(
    c("price_USA", "price_BTC"),
    function(x) {t.test(x ~ .$fruit)$p.value}
  )
> Error in model.frame.default(formula = x ~ .$fruit) : 
  variable lengths differ (found for '.$fruit')
tidyr dplyr r
1 Answers

You can do this by reshaping your data from wide to long format. Here's a solution using dplyr:

library(tidyverse)

prices <- tibble(
  country = c(rep("USA", 6), rep("Spain", 6), rep("Korea", 6)),
  fruit = rep(c("apples", "apples", "apples", "oranges", "oranges", "oranges"), 3),
  price_USA = rnorm(18),
  price_BTC = rnorm(18)
)

prices %>% 
  pivot_longer(cols = starts_with("price"), names_to = "name",
               values_to = "price", names_prefix = "price_") %>%
  group_by(country, name) %>%
  summarise(pval = t.test(price ~ fruit)$p.value)
#> # A tibble: 6 x 3
#> # Groups:   country [3]
#>   country name   pval
#>   <chr>   <chr> <dbl>
#> 1 Korea   BTC   0.458
#> 2 Korea   USA   0.721
#> 3 Spain   BTC   0.732
#> 4 Spain   USA   0.526
#> 5 USA     BTC   0.916
#> 6 USA     USA   0.679
user_131575360
Related