tests/testthat/test-z_test.R

In BFF: Bayes Factor Functions

test_that("two-sample: basic functionality", {

  ### test fixed omega -- non-local prior z-test (two-sided)
  fit <- z_test_BFF(
    z_stat = 1.5,
    alternative = "two.sided",
    n1 = 50,
    n2 = 50,
    omega = 0.5)

  # check that the BF and omega is consistent
  testthat::expect_equal(fit$log_bf, -0.27839, tolerance = 1e-5)
  testthat::expect_equal(fit$omega,  0.50)

  # test S3 methods
  testthat::expect_equal(
    testthat::capture_output_lines(fit, print = TRUE, width = 100),
    c(
      "\tBayesian non-local two-sample z test"  ,
      ""                                        ,
      "log Bayes factor = -0.28"                 ,
      "omega = 0.50 (Cohen's d)"                ,
      "alternative = two.sided"
    )
  )
  testthat::expect_error(plot(fit), "Bayes factor function can be plotted only if a specific omega/tau2 is not user set")

  # TODO: fix posterior plots
  # - I fixed the arguments not being properly passed
  # - however, the posterior distribution does not integrate to 1
  # (I remember that I raised this issue when I was in US, and Saptati was working on fixing it)

  # # this is how the functions should work
  # posterior_plot(fit)
  # posterior_plot(fit, prior = TRUE)
  #
  # # this highlights the issue (run `devtools::load_all()` first)
  # tau2 <- get_two_sample_tau2(n1 = fit$input$n1, n2 = fit$input$n2, w = fit$omega, r = fit$r)
  #
  # # does not integrate to 1
  # integrate(
  #   f = function(x) .z_test.posterior(
  #     z_stat = fit$input$z_stat, tau2 = tau2, r = fit$r, effect_size = x,
  #     n = fit$input$n, n1 = fit$input$n1, n2 = fit$input$n2, one_sample = fit$one_sample, one_sided = fit$alternative != "two.sided"),
  #   lower = -Inf,
  #   upper = Inf
  # )
  #
  # # prior seems to work just fine (i.e., integrates to one)
  # integrate(
  #   f = function(x) .z_test.prior(
  #     tau2 = tau2, r = fit$r, effect_size = x,
  #     n = fit$input$n, n1 = fit$input$n1, n2 = fit$input$n2, one_sample = fit$one_sample, one_sided = fit$alternative != "two.sided"),
  #   lower = -Inf,
  #   upper = Inf
  # )
  # <\TODO>

  # vdiffr::expect_doppelganger("t_test-two_sample-two_sided-posterior",           posterior_plot(fit))
  # vdiffr::expect_doppelganger("t_test-two_sample-two_sided-posterior_and_prior", posterior_plot(fit, prior = TRUE))


  ### test fixed omega -- non-local prior z-test (one-sided)
  fit <- z_test_BFF(
    z_stat = 1.5,
    alternative = "greater",
    n1 = 50,
    n2 = 50,
    omega = 0.5)

  # check that the BF and omega is consistent
  testthat::expect_equal(fit$log_bf, 0.29278, tolerance = 1e-4)
  testthat::expect_equal(fit$omega,  0.50)

  # test S3 methods
  testthat::expect_equal(
    testthat::capture_output_lines(fit, print = TRUE, width = 100),
    c(
      "\tBayesian non-local two-sample z test"  ,
      ""                                        ,
      "log Bayes factor = 0.29"                 ,
      "omega = 0.50 (Cohen's d)"                ,
      "alternative = greater"
    )
  )
  # vdiffr::expect_doppelganger("z_test-two_sample-one_sided-posterior",           posterior_plot(fit))
  # vdiffr::expect_doppelganger("z_test-two_sample-one_sided-posterior_and_prior", posterior_plot(fit, prior = TRUE))


  ### test unspecified omega -- BFF (one-sided; also change n1/n2)
  fit <- z_test_BFF(
    z_stat = 2.5,
    alternative = "two.sided",
    n1 = 25,
    n2 = 75)

  # check that the BF and omega is consistent
  testthat::expect_equal(fit$log_bf, 2.07857, tolerance = 1e-5)
  testthat::expect_equal(fit$omega,  0.45)

  # test S3 methods
  testthat::expect_equal(
    testthat::capture_output_lines(fit, print = TRUE, width = 100),
    c(
      "\tBayesian non-local two-sample z test"  ,
      ""                                        ,
      "maximized log Bayes factor = 2.08"       ,
      "maximized omega = 0.45 (Cohen's d)"      ,
      "alternative = two.sided"
    )
  )
  # vdiffr::expect_doppelganger("t_test_BFF-two_sample-two_sided-BFF",                 plot(fit))
  # vdiffr::expect_doppelganger("t_test_BFF-two_sample-two_sided-posterior",           posterior_plot(fit))
  # vdiffr::expect_doppelganger("t_test_BFF-two_sample-two_sided-posterior_and_prior", posterior_plot(fit, prior = TRUE, color = c("red", "blue"), linetype = c(3,5),
  #                                                                                                   linewidth = c(2, 1), x_limit = c(-2, 2)))


  # check that the data.frame plot output also works
  # no_plot_plot <- posterior_plot(fit, plot = FALSE, prior = TRUE)
  # testthat::expect_true(is.data.frame(no_plot_plot))
  # testthat::expect_equal(colnames(no_plot_plot), c("x", "prior", "posterior"))

  ### test unspecified omega -- BFF (one-sided)
  fit <- z_test_BFF(
    z_stat = 1.5,
    alternative = "less",
    n1 = 50,
    n2 = 50)

  # check that the BF and omega is consistent
  testthat::expect_equal(fit$log_bf, 0.00, tolerance = 1e-5)
  testthat::expect_equal(fit$omega,  0.00)

  # test S3 methods
  testthat::expect_equal(
    testthat::capture_output_lines(summary(fit), print = TRUE, width = 100),
    c(
      "\tBayesian non-local two-sample z test",
      ""                                        ,
      "maximized log Bayes factor = 0.00"       ,
      "maximized omega = 0.00 (Cohen's d)"      ,
      "alternative = less"
    )
  )
  # vdiffr::expect_doppelganger("z_test_BFF-two_sample-one_sided-BFF", plot(fit))
  # testthat::expect_error(posterior_plot(fit), "There is no non-local prior distribution")
})