tests/testthat/test-seq_anova_error_rates.R

In sprtt: Sequential Probability Ratio Tests Toolbox

#* @testing seq_anova

test_that("Check error rates", {
  # load packages
  library(testthis)
  library(dplyr)

  # skip the tests in special situations
  # skip_on_cran() # skip on CRAN checks: necessary!
  # skip_on_ci() # skip on continuous platforms like GitHub Actions
  skip("seq_anova simulation: not written yet") # skip test

  # test setup: simulate the data
  # testthat::source_dir(
  #   "testdata-raw/",
  #   pattern = "error_rates_anova\\.[rR]$",
  #   env = test_env(),
  #   chdir = TRUE,
  #   wrap = TRUE
  # )

  all_data <- testthis::read_testdata("error_rates_anova_simulation")

  effect_data <- all_data %>%
    dplyr::filter(f_simulated != 0)
  null_effect_data <- all_data %>%
    dplyr::filter(f_simulated == 0)

  # test data reduction: seq_anova must be more efficient than fixed anova
  # expect less than
  expect_true(
    sum(effect_data$sample_size_seq, na.rm = TRUE) /
    sum(effect_data$sample_size_fixed, na.rm = TRUE) < 1
  )
  # at least in 80% of the cases the seq_anova is more efficient
  percent_smaller_samples <- mean(effect_data$sample_smaller)
  expect_true(percent_smaller_samples > 0.80)

  alpha_error_rate <- mean(null_effect_data$error)
  beta_error_rate <- mean(effect_data$error)

  expect_true(alpha_error_rate < 0.05)
  expect_true(beta_error_rate < 0.05)

})