t-stat by group using tidyr::nest and purrr::map

tstat.R

library(dplyr)
library(ggplot2)
library(tidyr)
library(purrr)
library(broom)
library(forcats)

d <- data.frame(
  value = rnorm(24),
  type = factor(rep(c("A","B"), c(5,3))),
  group = rep(1:3, each=8),
  group_var = rep(rnorm(3), each=8)
)

d %>%
  ggplot(aes(type, value)) +
  geom_boxplot() +
  facet_wrap(~ group)

d %>%
  mutate(type = fct_relevel(type, "B", "A")) %>%
  nest(data = c(type, value)) %>%
  mutate(ttest = map(data, ~t.test(value ~ type, data=.)),
         tstat = map(ttest, tidy)) %>%
  unnest(tstat) %>%
  select(group, group_var, statistic)

# manually (note that t-test usually does A - B, but we want B - A)
d %>%
  filter(group == 1) %>%
  t.test(value ~ type, data=.)

fct_relevel here is just to deal with the fact that t.test computes A - B but we usually want the reverse comparison

nest takes the columns we specify and converts them into a nested data.frame called data

after nest, the operations are within group, and map takes the data and applies the function in the second argument to make a new column in the outer data.frame. the columns in the outer data.frame can be complex objects like the result of t.test

tidy cleans up fitted objects and turns them into tables

unnest them propagates the information back to the outer data.frame