tests/testthat/test-eta_squared.R

In effectsize: Indices of Effect Size

test_that("alternative = NULL", {
  m <- aov(mpg ~ factor(cyl) + hp, mtcars)
  expect_equal(
    eta_squared(m),
    eta_squared(m, alternative = NULL)
  )
})

# anova() -----------------------------------------------------------------
test_that("anova()", {
  # Make minimal ANOVA table
  mod <- anova(lm(mpg ~ cyl + hp, mtcars))

  mod1 <- mod
  mod1$DenDF <- mod1$Df[nrow(mod1)]
  mod1 <- mod1[-nrow(mod1), ]

  expect_error(eta_squared(mod1), regexp = NA)
  expect_equal(
    eta_squared(mod1)[, -1],
    F_to_eta2(mod1[["F value"]], mod1$Df, mod1$DenDF),
    ignore_attr = TRUE
  )
  expect_warning(eta_squared(mod1, partial = FALSE), "partial")
  expect_warning(eta_squared(mod1, generalized = TRUE), "Generalized")

  mod2 <- mod1
  mod2$`F value` <- NULL
  expect_error(eta_squared(mod2), "does not")
})

# aov ---------------------------------------------------------------------
test_that("aov", {
  df <- iris
  df$Sepal.Big <- ifelse(df$Sepal.Width >= 3, "Yes", "No")

  fit <- aov(Sepal.Length ~ Species * Sepal.Big, df)

  # eta
  expect_equal(eta_squared(fit, partial = FALSE)$Eta2,
    c(0.618, 0.046, 0.000),
    tolerance = 0.01
  )
  expect_equal(eta_squared(fit, partial = TRUE)$Eta2_partial,
    c(0.649, 0.121, 0.001),
    tolerance = 0.01
  )

  # omega
  expect_equal(omega_squared(fit, partial = FALSE)$Omega2,
    c(0.612, 0.043, 0),
    tolerance = 0.01
  )
  expect_equal(omega_squared(fit, partial = TRUE)$Omega2_partial,
    c(0.638, 0.112, 0),
    tolerance = 0.01
  )

  # epsilon
  expect_equal(epsilon_squared(fit, partial = FALSE)$Epsilon2,
    c(0.614, 0.044, 0),
    tolerance = 0.001
  )
  expect_equal(epsilon_squared(fit, partial = TRUE)$Epsilon2_partial,
    c(0.644, 0.115, 0),
    tolerance = 0.01
  )

  # Cohen's f/f2
  expect_equal(cohens_f_squared(fit, partial = FALSE)$Cohens_f2,
    c(1.623, 0.049, 0.000),
    tolerance = 0.001
  )
  expect_equal(cohens_f_squared(fit, partial = TRUE)$Cohens_f2_partial,
    c(1.850, 0.139, 0.001),
    tolerance = 0.001
  )
  expect_equal(cohens_f(fit, partial = FALSE)$Cohens_f,
    c(1.273, 0.220, 0.021),
    tolerance = 0.01
  )
  expect_equal(cohens_f(fit, partial = TRUE)$Cohens_f_partial,
    c(1.360, 0.373, 0.036),
    tolerance = 0.001
  )
  expect_equal(cohens_f(fit, squared = TRUE), cohens_f_squared(fit))
  expect_equal(cohens_f_squared(fit, squared = FALSE), cohens_f(fit))



  #### One way-between
  expect_message(eta_squared(aov(mpg ~ factor(gear), mtcars)))
  expect_message(eta_squared(aov(mpg ~ factor(gear) + am, mtcars)), regexp = NA)

  #### Alternative
  m <<- aov(mpg ~ factor(gear) + am, mtcars)
  et1 <- eta_squared(m)
  et2 <- eta_squared(m, ci = 0.9, alternative = "two.sided")
  expect_equal(et1$CI_low, et2$CI_low)

  ### parameters:
  expect_equal(
    eta_squared(parameters::model_parameters(m)),
    eta_squared(m)
  )
})


# aovlist -----------------------------------------------------------------
test_that("aovlist", {
  skip_on_cran()
  df <- iris
  df$Sepal.Big <- ifelse(df$Sepal.Width >= 3, "Yes", "No")

  model <<- aov(Sepal.Length ~ Sepal.Big + Error(Species), data = df)

  res <- eta_squared(model, partial = TRUE)
  expect_true(all(c("Group", "Parameter") %in% colnames(res)))
  expect_equal(res$Eta2_partial, c(0.4472423, 0.1217329), tolerance = 0.001)
  expect_equal(eta_squared(model, partial = FALSE)$Eta2,
    c(0.27671136, 0.04641607),
    tolerance = 0.001
  )

  res <- omega_squared(model, partial = TRUE)
  expect_true(all(c("Group", "Parameter") %in% colnames(res)))
  expect_equal(res$Omega2_partial, c(0, 0.04141846), tolerance = 0.001)
  expect_equal(omega_squared(model, partial = FALSE)$Omega2,
    c(0, 0.03287821),
    tolerance = 0.001
  )

  res <- epsilon_squared(model, partial = TRUE)
  expect_true(all(c("Group", "Parameter") %in% colnames(res)))
  expect_equal(res$Epsilon2_partial, c(0, 0.1157174), tolerance = 0.001)
  expect_equal(epsilon_squared(model, partial = FALSE)$Epsilon2,
    c(0, 0.04412238),
    tolerance = 0.001
  )


  expect_equal(
    eta_squared(parameters::model_parameters(model)),
    eta_squared(model)
  )


  skip_if_not_installed("afex")
  # non-partial Eta2 should be the same for aov and aovlist
  data(obk.long, package = "afex")
  suppressMessages({
    model <- afex::aov_car(value ~ treatment * gender + Error(id / (phase * hour)),
      data = obk.long, observed = "gender",
      include_aov = TRUE
    )
  })

  model2 <- aov(value ~ treatment * gender * phase * hour,
    data = model$data$long,
    contrasts = list(
      treatment = contr.sum,
      gender = contr.sum,
      phase = contr.sum,
      hour = contr.sum
    )
  )

  a1 <- eta_squared(model2, partial = FALSE)
  a2 <- eta_squared(model$aov, partial = FALSE)

  rownames(a1) <- a1$Parameter
  rownames(a2) <- a2$Parameter

  expect_equal(
    a1[a1$Parameter, "Eta2"],
    a2[a1$Parameter, "Eta2"]
  )
})

# mlm / anova table -------------------------------------------------------
test_that("mlm / anova table", {
  data("mtcars")
  mtcars$am_f <- factor(mtcars$am)
  mtcars$cyl_f <- factor(mtcars$cyl)

  mod <- lm(cbind(mpg, qsec) ~ am_f * cyl_f, data = mtcars)
  m1 <- lm(mpg ~ am_f * cyl_f, data = mtcars)
  m2 <- lm(qsec ~ am_f * cyl_f, data = mtcars)

  expect_equal(
    eta_squared(mod)$Eta2_partial[1:3],
    eta_squared(m1)$Eta2_partial
  )

  expect_equal(
    eta_squared(mod)$Eta2_partial[4:6],
    eta_squared(m2)$Eta2_partial
  )

  expect_equal(
    eta_squared(mod, partial = FALSE)$Eta2[1:3],
    eta_squared(m1, partial = FALSE)$Eta2
  )

  expect_equal(
    eta_squared(mod, partial = FALSE)$Eta2[4:6],
    eta_squared(m2, partial = FALSE)$Eta2
  )

  # MANOVA table
  mod <- manova(cbind(mpg, qsec) ~ am_f * cyl_f, data = mtcars)
  expect_identical(nrow(eta_squared(mod)), 3L)

  # Row order
  fit <- lm(cbind(mpg, disp, hp) ~ factor(cyl), data = mtcars)
  out <- eta_squared(fit, partial = FALSE, ci = NULL)
  expect_equal(as.character(out$Response), c("mpg", "disp", "hp"))
})


# Cohen's f - R2 change ---------------------------------------------------
test_that("Cohen's f - R2 change", {
  data(hardlyworking)
  m1 <- lm(salary ~ xtra_hours, data = hardlyworking)
  m2 <- lm(salary ~ xtra_hours + n_comps, data = hardlyworking)

  fsD <- cohens_f_squared(m1, model2 = m2)[, 1:4]
  fs <- cohens_f_squared(m2)[-1, -1] # this ONLY works because of the default type-I errors!!!!
  rownames(fsD) <- rownames(fs) <- 1
  expect_equal(fsD, fs, tolerance = 0.01, ignore_attr = TRUE)

  fsD <- cohens_f_squared(m1, model2 = m2)
  R2_1 <- performance::r2(m1)[[1]]
  R2_2 <- performance::r2(m2)[[1]]
  expect_equal(
    fsD$Cohens_f2_partial,
    unname((R2_2 - R2_1) / (1 - R2_2))
  )
})

# generalized Eta -------------------------------------------------------------
test_that("generalized | between", {
  skip_if_not_installed("afex")
  skip_if_not_installed("car")

  data(obk.long, package = "afex")
  suppressMessages(suppressWarnings(
    m <- afex::aov_car(value ~ treatment * gender + Error(id),
      data = obk.long, observed = "gender",
      include_aov = TRUE
    )
  ))

  Aov <- car::Anova(m$aov, type = 3)

  expect_equal(
    anova(m, es = "ges", observed = NULL)$ges,
    eta_squared(Aov, generalized = TRUE, verbose = FALSE)$Eta2_generalized
  )


  expect_equal(
    anova(m, es = "ges", observed = "gender")$ges,
    eta_squared(Aov, generalized = "gender", verbose = FALSE)$Eta2_generalized
  )

  # in a completely between design, with all measured,
  # all are equal to total
  expect_equal(
    eta_squared(Aov, generalized = c("gender", "treatment"), verbose = FALSE)[[2]],
    eta_squared(Aov, partial = FALSE, verbose = FALSE)[[2]]
  )
})


test_that("generalized | within-mixed", {
  skip_if_not_installed("afex")
  data(obk.long, package = "afex")

  # estimate mixed ANOVA on the full design:
  suppressMessages(
    m <- afex::aov_car(value ~ treatment * gender + Error(id / (phase * hour)),
      data = obk.long, observed = "gender",
      include_aov = TRUE
    )
  )


  ef <- eta_squared(m$aov, generalized = "gender")
  af <- anova(m, es = "ges", observed = "gender")
  expect_equal(ef$Eta2_generalized,
    c(
      0.211, 0.083, 0.186, 0.193, 0.099,
      0.002, 0.015, 0.132, 0.001, 0.004,
      0.011, 0.016, 0.008, 0.01, 0.02
    ),
    tolerance = 0.05
  )
  expect_equal(ef$Eta2_generalized,
    af$ges,
    tolerance = 0.1
  )


  ef <- eta_squared(m$aov, generalized = TRUE)
  af <- anova(m, es = "ges", observed = NULL)
  expect_equal(ef$Eta2_generalized,
    c(
      0.286, 0.111, 0.218, 0.264, 0.142,
      0.004, 0.021, 0.185, 0.002, 0.005,
      0.016, 0.023, 0.013, 0.014, 0.029
    ),
    tolerance = 0.05
  )
  expect_equal(ef$Eta2_generalized,
    af$ges,
    tolerance = 0.1
  )
})



# rm-omega ----------------------------------------------------------------
test_that("omega", {
  skip_if_not_installed("afex")
  # cross validated with MOTE
  data(obk.long, package = "afex")

  suppressMessages(suppressWarnings(
    m <- afex::aov_car(value ~ treatment * gender + Error(id / (phase)),
      data = obk.long, observed = "gender",
      include_aov = TRUE
    )
  ))


  ef <- omega_squared(m, partial = TRUE, alternative = "two")
  expect_equal(ef$Omega2_partial,
    c(0.3115, 0.1814, 0.2221, 0.2637, 0.1512, 0, 0),
    tolerance = 0.01
  )
  expect_equal(ef$CI_low,
    c(0, 0, 0, 0, 0, 0, 0),
    tolerance = 0.01
  )

  expect_equal(ef$CI_high,
    c(0.626, 0.553, 0.557, 0.518, 0.355, 0, 0),
    tolerance = 0.01
  )
})


# failed CIs --------------------------------------------------------------

test_that("failed CIs", {
  model <- aov(wt ~ cyl + Error(gear), data = mtcars)

  expect_warning(eta_squared(model), regexp = "CIs")
  expect_warning(eta <- eta_squared(model, verbose = FALSE), regexp = NA)
  expect_identical(nrow(eta), 2L)
  expect_equal(eta[1, "Eta2_partial"], 1)

  expect_warning(eta_squared(model, partial = FALSE), regexp = "CIs")
  expect_warning(eta <- eta_squared(model, partial = FALSE, verbose = FALSE), regexp = NA)
  expect_identical(nrow(eta), 2L)
  expect_equal(eta[1, "Eta2"], 0.34, tolerance = 0.01)
})


# Include intercept -------------------------------------------------------
test_that("include_intercept | car", {
  skip_on_cran()
  skip_if_not_installed("car")

  m <- lm(mpg ~ factor(cyl) * factor(am), data = mtcars)
  AOV <- car::Anova(m, type = 3)

  res0 <- eta_squared(AOV, verbose = FALSE)
  res1 <- eta_squared(AOV, include_intercept = TRUE, verbose = FALSE)
  expect_identical(nrow(res0), 3L)
  expect_identical(nrow(res1), nrow(res0) + 1L)
  expect_identical(res1[[1]][1], "(Intercept)")
  expect_equal(res1[[2]][1], 0.8680899, tolerance = 0.01)

  res0 <- epsilon_squared(AOV, verbose = FALSE)
  res1 <- epsilon_squared(AOV, include_intercept = TRUE, verbose = FALSE)
  expect_identical(nrow(res0), 3L)
  expect_identical(nrow(res1), nrow(res0) + 1L)
  expect_equal(res1[[1]][1], "(Intercept)")


  res0 <- omega_squared(AOV, verbose = FALSE)
  res1 <- omega_squared(AOV, include_intercept = TRUE, verbose = FALSE)
  expect_identical(nrow(res0), 3L)
  expect_identical(nrow(res1), nrow(res0) + 1L)
  expect_identical(res1[[1]][1], "(Intercept)")

  # generalized
  res1 <- eta_squared(AOV, generalized = "cyl", include_intercept = TRUE, verbose = FALSE)
  expect_equal(res1[[1]][1], "(Intercept)")
  expect_equal(res1[[2]][1], 0.784483, tolerance = 0.01)
})


test_that("include_intercept | afex", {
  skip_if_not_installed("afex")
  data(obk.long, package = "afex")

  suppressWarnings(suppressMessages(
    a <- afex::aov_car(value ~ treatment * gender + Error(id),
      include_aov = TRUE,
      data = obk.long
    )
  ))

  resE0 <- eta_squared(a, verbose = FALSE)
  resA0 <- anova(a, es = "pes")
  expect_identical(nrow(resE0), 3L)
  expect_identical(nrow(resE0), nrow(resA0))


  resE1 <- eta_squared(a, include_intercept = TRUE, verbose = FALSE)
  resA1 <- anova(a, es = "pes", intercept = TRUE)
  expect_identical(nrow(resE1), nrow(resE0) + 1L)
  expect_identical(nrow(resE1), nrow(resA1))

  skip_if_not_installed("car")
  resE1 <- eta_squared(car::Anova(a$aov, type = 3), include_intercept = TRUE, generalized = "gender", verbose = FALSE)
  resA1 <- anova(a, es = "ges", intercept = TRUE, observed = "gender")
  expect_equal(resE1[[2]][1], 0.9386555, tolerance = 0.01)
  expect_equal(resE1[[2]][1], resA1[[5]][1], tolerance = 0.01)
})

# Special cases --------------------------------------------------------------

## afex --------------------------------------------------------------------
test_that("afex | within-mixed", {
  skip_if_not_installed("afex")

  data(obk.long, package = "afex")

  suppressMessages(mod <- afex::aov_ez("id", "value", obk.long,
    between = c("treatment", "gender"),
    within = c("phase", "hour"),
    observed = "gender"
  ))

  x <- eta_squared(mod, generalized = TRUE)
  a <- anova(mod, observed = "gender")
  expect_equal(a$ges, x$Eta2_generalized)

  x <- eta_squared(mod)
  a <- anova(mod, es = "pes")
  expect_equal(a$pes, x$Eta2_partial)


  x <- eta_squared(mod, include_intercept = TRUE)
  a <- anova(mod, es = "pes", intercept = TRUE)
  expect_equal(a$pes, x$Eta2_partial)

  # see issue #389
  data <- data.frame(
    subject = c(
      1L, 2L, 1L, 2L, 1L, 2L, 1L, 2L, 1L,
      2L, 1L, 2L, 1L, 2L, 1L, 2L
    ),
    y = c(
      0.0586978983148275, -0.159870038198774, 0.0125690871484012,
      -0.0152529928817782, 0.092433880558952, 0.0359796249184537,
      -0.00786545388312909, 0.0340005375703463, 0.165294695432772,
      0.0201040753050847, 0.0741924965491503, -0.0345053066539826,
      0.0108194665250311, -0.163941830205729, 0.310344189786906,
      -0.106627229564326
    ),
    A = c(
      "A1", "A1", "A1", "A1", "A1", "A1", "A1", "A1", "A2",
      "A2", "A2", "A2", "A2", "A2", "A2", "A2"
    ),
    B = c(
      "B1", "B1", "B1", "B1", "B2", "B2", "B2", "B2", "B1",
      "B1", "B1", "B1", "B2", "B2", "B2", "B2"
    ),
    C = c(
      "C1", "C1", "C2", "C2", "C1", "C1", "C2", "C2", "C1",
      "C1", "C2", "C2", "C1", "C1", "C2", "C2"
    )
  )
  suppressMessages(mod <- afex::aov_ez("subject", "y", data, within = c("A", "B", "C")))
  tab <- as.data.frame(anova(mod, es = "pes"))
  res <- eta_squared(mod)

  tab <- tab[order(rownames(tab)), ]
  res <- res[order(res$Parameter), ]

  expect_equal(res$Eta2_partial, tab$pes, tolerance = 0.001)
})


test_that("afex | mixed()", {
  skip_if_not_installed("afex")
  skip_if_not_installed("lmerTest")
  skip_if_not_installed("base", minimum_version = "3.6.1")

  data(md_15.1, package = "afex")
  # random intercept plus random slope
  suppressMessages(t15.4a <- afex::mixed(iq ~ timecat + (1 + time | id),
    data = md_15.1,
    method = "S"
  ))
  expect_equal(
    eta_squared(t15.4a),
    eta_squared(t15.4a$full_model)
  )

  # Intercept
  data("stroop", package = "afex")
  stroop <- subset(stroop, study == 1 & acc == 1 & trialnum < 20)
  suppressMessages(m1 <- afex::mixed(rt ~ condition + (condition | pno), data = stroop, method = "KR"))
  suppressMessages(m2 <- afex::mixed(rt ~ condition + (condition | pno), data = stroop, test_intercept = TRUE, method = "KR"))

  expect_warning(a1a <- eta_squared(m1, include_intercept = TRUE), regexp = "Intercept")
  expect_warning(a1b <- eta_squared(m1, include_intercept = FALSE), regexp = NA)
  expect_equal(a1a, a1b)
  expect_equal(nrow(a1a), 1L)

  expect_warning(a2a <- eta_squared(m2, include_intercept = TRUE), regexp = NA)
  expect_warning(a2b <- eta_squared(m2, include_intercept = FALSE), regexp = NA)
  expect_equal(nrow(a2a), 2L)
  expect_equal(nrow(a2b), 1L)
  expect_equal(a1a, a2a[2, ], ignore_attr = TRUE)
})


## car ---------------------------------------------------------------------
test_that("car MVM", {
  skip_if_not_installed("afex")
  skip_if_not_installed("car")

  # Simple ---
  ds <- data.frame(
    I = c(116, 96, 120, 110, 116, 126, 86, 80),
    II = c(76, 93, 112, 113, 75, 120, 90, 105),
    III = c(85, 63, 89, 60, 115, 101, 129, 67),
    IV = c(50, 87, 100, 60, 79, 70, 65, 65),
    id = 1:8
  )

  ds_long <-
    datawizard::reshape_longer(ds,
      select = 1:4,
      names_to = "ind_var",
      values_to = "score"
    )


  fit <- lm(cbind(I, II, III, IV) ~ 1, data = ds)
  in_rep <- data.frame(ind_var = gl(4, 1))
  suppressMessages(A_car <- car::Anova(fit, idata = in_rep, idesign = ~ind_var))

  eta_car <- effectsize::eta_squared(A_car, ci = NULL)[[2]]

  eta_afex <- afex::aov_ez("id", "score", ds_long,
    within = "ind_var",
    anova_table = list(es = "pes")
  )$anova_table$pes

  expect_equal(eta_car, eta_afex)

  # Complex ---
  data(obk.long, package = "afex")

  suppressMessages(mod <- afex::aov_ez("id", "value", obk.long,
    between = c("treatment", "gender"),
    within = c("phase", "hour"),
    observed = "gender"
  ))
  expect_equal(
    sort(eta_squared(mod$Anova, generalized = "gender")[[2]]),
    sort(mod$anova_table$ges)
  )
})


test_that("Anova.mlm Manova", {
  skip_if_not_installed("car")

  data("mtcars")
  mtcars <- mtcars[c(1L, 2L, 3L, 4L, 5L, 6L, 7L, 8L, 9L, 18L, 29L, 31L), ]
  mtcars$am_f <- factor(mtcars$am)
  mtcars$cyl_f <- factor(mtcars$cyl)

  mod <- lm(cbind(mpg, qsec, disp) ~ am_f * cyl_f, data = mtcars)

  Manova <- car::Manova(mod, type = 2)

  expect_true(is.null(summary(Manova, univariate = TRUE)[["univariate.tests"]]))
  expect_error(eta_squared(Manova), regexp = NA)
  expect_equal(
    eta_squared(manova(mod))[[2]][2:3],
    eta_squared(Manova)[[2]][2:3]
  )

  Anova <- car::Anova(mod, idesign = ~g, idata = data.frame(g = factor(1:3)))
  mtcars$id <- factor(seq(nrow(mtcars)))
  mtcars_long <- datawizard::reshape_longer(mtcars,
    select = c("mpg", "qsec", "disp"), names_to = "g"
  )
  a1 <- aov(value ~ am_f * cyl_f * g + Error(id / g), data = mtcars_long)

  A1 <- eta_squared(Anova)
  A2 <- eta_squared(a1)
  expect_equal(A1$Parameter, A2$Parameter)
  expect_equal(A1[c(2:4, 6:7), ], A2[c(2:4, 6:7), -1], ignore_attr = TRUE)
})

## merMod --------------------

test_that("merMod and lmerModLmerTest", {
  skip_if_not_installed("lmerTest")
  skip_if_not_installed("lme4")

  data("sleepstudy", package = "lme4")

  m <- lme4::lmer(Reaction ~ Days + (Days | Subject), sleepstudy)
  mtest <- lmerTest::lmer(Reaction ~ Days + (Days | Subject), sleepstudy)

  expect_equal(
    eta_squared(m),
    eta_squared(mtest)
  )
})


## tidymodels -------------------
test_that("ets_squared | tidymodels", {
  skip_on_cran()
  skip_if_not_installed("parsnip")
  skip_if_not_installed("base", minimum_version = "3.6.1")

  set.seed(123)
  mod_lm <- parsnip::linear_reg(engine = "lm", mode = "regression")
  mod_lm <- parsnip::fit(mod_lm, mpg ~ am + vs, data = mtcars)

  set.seed(123)
  tidy_lm <- eta_squared(mod_lm)
  lm_lm <- eta_squared(lm(mpg ~ am + vs, data = mtcars))

  expect_equal(tidy_lm, lm_lm, tolerance = 0.001)
})


## GAMs -------------------
test_that("ets_squared | gam", {
  skip_on_cran()
  skip_if_not_installed("mgcv")

  set.seed(2) ## simulate some data...
  dat <- mgcv::gamSim(1, n = 400, dist = "normal", scale = 2, verbose = FALSE)
  b <- mgcv::gam(y ~ x0 + s(x1) + s(x2) + t2(x1, x2) + s(x3), data = dat)

  expect_error(out <- eta_squared(b), regexp = NA)
  expect_warning(eta_squared(b), regexp = NA)
  expect_output(print(out), "Type III")
})

## rms -------------------
test_that("ets_squared | rms", {
  skip_on_cran()
  skip_if_not_installed("rms")
  data("mtcars")

  b <- rms::ols(mpg ~ cyl + am, data = mtcars)
  expect_error(out <- eta_squared(b), regexp = NA)
  expect_output(print(out), "Type II")

  skip_if_not_installed("car")
  skip_if_not_installed("base", minimum_version = "3.6.1")
  b_lm <- car::Anova(lm(mpg ~ cyl + am, data = mtcars), type = 2)
  out_lm <- eta_squared(b_lm)
  expect_equal(out[1:2, ], out_lm, ignore_attr = TRUE)
})