Using the following data
df <- data.frame(category = sample(1:3, replace = TRUE, 50),
testgroup = sample(c('A', 'B'), replace = TRUE, 50),
var_1 = rnorm(50),
var_2 = rnorm(50),
var_3 = rnorm(50)
)
I would like to apply a 2-sample-t-test in each category comparing the difference in means between A and B with respect to all 3 variables.
Ideally, the output is generated using the tidyverse and broom packages.
I was struggling too long with the split-apply-combine-approach and I guess there is already a nice solution out there with a few lines of code.
Thanks a lot for your support!
The general rule of thumb is to get the arguments for the desired function (t.test in this case) in side-by-side columns. In your case, we aim to have A and B side by side:
X <- df %>% group_by( category, testgroup ) %>%
summarize( across(starts_with("var"), list) ) %>%
ungroup() %>%
pivot_longer( starts_with("var"), "variable", values_to="values" ) %>%
pivot_wider( names_from="testgroup", values_from="values" )
# # A tibble: 9 x 4
# category variable A B
# <int> <chr> <list> <list>
# 1 1 var_1 <dbl [3]> <dbl [3]>
# 2 1 var_2 <dbl [3]> <dbl [3]>
# 3 1 var_3 <dbl [3]> <dbl [3]>
# 4 2 var_1 <dbl [11]> <dbl [9]>
# 5 2 var_2 <dbl [11]> <dbl [9]>
# ...
We are now well positioned to apply a two-sample t-test and process the results with broom:
X %>% mutate(test = map2(A, B, t.test),
result = map(test, broom::tidy) ) %>%
unnest( result )
# # A tibble: 9 x 15
# category variable A B test estimate estimate1 estimate2 statistic
# <int> <chr> <lis> <lis> <lis> <dbl> <dbl> <dbl> <dbl>
# 1 1 var_1 <dbl… <dbl… <hte… 1.07 0.400 -0.665 1.08
# 2 1 var_2 <dbl… <dbl… <hte… -0.376 0.350 0.726 -0.415
# 3 1 var_3 <dbl… <dbl… <hte… -0.701 -0.102 0.599 -0.434
# 4 2 var_1 <dbl… <dbl… <hte… -0.276 -0.335 -0.0587 -0.531
# 5 2 var_2 <dbl… <dbl… <hte… 0.727 0.689 -0.0374 1.74
# ...