tests/testthat/test-seed.R

In s3alfisc/fwildclusterboot: Fast Wild Cluster Bootstrap Inference for Linear Models

test_that("seed works for OLS", {
  skip_on_cran()
  skip_if_not(
    find_proglang("julia"),
    message = "skip test as julia installation not found."
  )

  set.seed(12312)
  dqrng::dqset.seed(9786)

  #' @srrstats {G5.0} *Where applicable or practicable, tests should use
  #' standard data sets with known properties (for example, the
  #' [NIST Standard Reference Datasets](https://www.itl.nist.gov/div898/strd/),
  #'  or data sets provided by other widely-used R packages).* All tests are based
  #'  on random data sets.

  data1 <<-
    fwildclusterboot:::create_data(
      N = 5000,
      N_G1 = 40,
      icc1 = 0.01,
      N_G2 = 10,
      icc2 = 0.01,
      numb_fe1 = 10,
      numb_fe2 = 10,
      seed = 1
    )
  lm_fit <-
    lm(proposition_vote ~ treatment + ideology1 + log_income + Q1_immigration,
      data = data1
    )


  for (engine in c("R", "R-lean", "WildBootTests.jl")) {
    # Case 1: seed set, no internal seeds
    set.seed(123)
    dqrng::dqset.seed(123)

    boot_lm_s1 <- suppressMessages(
      boottest(
        object = lm_fit,
        clustid = "group_id1",
        B = 999,
        param = "treatment",
        type = "rademacher",
        conf_int = FALSE,
        engine = engine
      )
    )

    boot_lm_s2 <- suppressMessages(
      boottest(
        object = lm_fit,
        clustid = "group_id1",
        B = 999,
        param = "treatment",
        type = "rademacher",
        conf_int = FALSE,
        engine = engine
      )
    )

    expect_true(boot_lm_s1$p_val != boot_lm_s2$p_val)

    # Case 5: different starting seeds

    set.seed(9)
    dqrng::dqset.seed(9)
    boot_lm_s1 <- suppressMessages(
      boottest(
        object = lm_fit,
        clustid = "group_id1",
        B = 999,
        param = "treatment",
        type = "rademacher",
        conf_int = FALSE,
        engine = engine
      )
    )

    set.seed(2)
    dqrng::dqset.seed(2)
    boot_lm_s2 <- suppressMessages(
      boottest(
        object = lm_fit,
        clustid = "group_id1",
        B = 999,
        param = "treatment",
        type = "rademacher",
        conf_int = FALSE,
        engine = engine
      )
    )

    expect_true(boot_lm_s1$p_val != boot_lm_s2$p_val)
  }
})