tests/testthat/test-compute_indirect_effect_for.R

In JSmediation: Mediation Analysis Using Joint Significance

test_that("Mod arg accepts single numeric value", {
  moderated_mediation_model <-
    ho_et_al %>%
    dplyr::mutate(condition_c =
                    build_contrast(condition,
                                   "High discrimination",
                                   "Low discrimination"),
                  linkedfate_c =
                    scale(linkedfate, scale = FALSE),
                  sdo_c =
                    scale(sdo, scale = FALSE)) %>%
    mdt_moderated(
      condition_c,
      hypodescent,
      linkedfate_c,
      sdo_c
    )

  expect_error(
    moderated_mediation_model %>% compute_indirect_effect_for(Mod = "foo")
  )

  expect_error(
    moderated_mediation_model %>% compute_indirect_effect_for(Mod = c(1, 2))
  )

  expect_error(
    moderated_mediation_model %>% compute_indirect_effect_for(Mod = 0),
    NA
  )
})

test_that("JSmediation approach is consistent with the {mediation} approach (Mod = 1)", {

  skip_if(! requireNamespace("mediation", quietly = TRUE))

  # note that because the mediation::mediate is computationnaly intensive, we
  # compute the conditionnal indirect effect for a single moderator value.



  # set seed
  withr::local_seed(123)
  # data set
  dataset <-
    ho_et_al %>%
    dplyr::mutate(condition_c = build_contrast(condition,
                                               "Low discrimination",
                                               "High discrimination")) %>%
    dplyr::mutate(dplyr::across(c(linkedfate, sdo),
                                ~ as.numeric(scale(.))))

  # mediation approach
  model_1 <- lm(linkedfate ~ condition_c * sdo, dataset)
  model_2 <- lm(hypodescent ~ (condition_c + linkedfate) * sdo, dataset)

  mediation_model <-
    mediation::mediate(model_1, model_2,
                       covariates = list(sdo = 1),
                       boot = TRUE,
                       boot.ci.type = "perc",
                       sims = 5000,
                       treat    = "condition_c",
                       mediator = "linkedfate")

  JSmediation_model <-
    dataset %>%
    mdt_moderated(DV = hypodescent,
                  IV = condition_c,
                  M = linkedfate,
                  Mod = sdo) %>%
    compute_indirect_effect_for(Mod = 1)

  #checks
  JSmediation_estimate <- purrr::chuck(JSmediation_model, "estimate")
  mediation_estimate   <- purrr::chuck(mediation_model, "d0")

  expect_equal(JSmediation_estimate, mediation_estimate, tolerance = 5e-2)
})

# test_that("JSmediation approach is consistent with the {processR} approach (Mod = -1, 0, 1)", {
#   # set seed
#   withr::local_seed("123")
#
#   skip_if(! requireNamespace("processR", quietly = TRUE))
#
#   # data set
#   dataset <-
#     ho_et_al %>%
#     dplyr::mutate(condition_c = build_contrast(condition,
#                                                "Low discrimination",
#                                                "High discrimination")) %>%
#     dplyr::mutate(dplyr::across(c(linkedfate, sdo),
#                                 ~ as.numeric(scale(.))))
#
#   # processR approach
#   # gen eqn (model 59)
#   moderated_mediation_eqn <-
#     processR::tripleEquation(X = "condition_c",
#                              M = "linkedfate",
#                              Y = "hypodescent",
#                              moderator = list(name = "sdo",
#                                               site = list(c("a", "b", "c"))))
#
#   moderated_mediation_fit <-
#     withr::with_options(list(warn = -1),
#                         lavaan::sem(moderated_mediation_eqn,
#                                     dataset))
#
#   moderated_mediation_indirect_indices <-
#     processR::modmedSummary(moderated_mediation_fit)
#
#   # JSmediation approach
#   JSmediation_model <-
#     dataset %>%
#     mdt_moderated(condition_c,
#                   hypodescent,
#                   linkedfate,
#                   sdo)
#
#   # check
#   processR_estimates <-
#     purrr::chuck(moderated_mediation_indirect_indices, "indirect")
#
#   JSmediation_estimate <-
#     purrr::chuck(moderated_mediation_indirect_indices, "values") %>%
#     purrr::map_dbl(~ JSmediation_model %>%
#                      compute_indirect_effect_for(.x) %>%
#                      purrr::chuck("estimate"))
#
#   expect_equal(processR_estimates, JSmediation_estimate,
#                tolerance = 5e-2)
# })