tests/testthat/test-reliability_adapters.R

In EFAtools: Fast and Flexible Implementations of Exploratory Factor Analysis Tools

## Tests for the .rel_adapt_* front-end adapters. Each normalizes one input source
## to the spec consumed by .reliability_core(); the spec must drive the core to the
## same numbers the current OMEGA path produces for that input.

efa_mod <- EFA(test_models$baseline$cormat, N = 500, n_factors = 3,
               type = "EFAtools", method = "PAF", rotation = "promax")
sl_mod <- efa_schmid_leiman(efa_mod, estimate_control = estimate_control(type = "EFAtools"), estimator = "PAF")
schmid_mod <- psych::schmid(test_models$baseline$cormat, nfactors = 3,
                            n.obs = 500, fm = "pa", rotate = "Promax")
fc <- sl_mod$sl[, c("F1", "F2", "F3")] >= .2

test_that(".rel_adapt_SL reproduces the SL OMEGA path", {
  spec <- .rel_adapt_SL(sl_mod, factor_corres = fc, type = "EFAtools")

  expect_identical(
    .reliability_core(spec, "correlation"),
    .OMEGA_FLEX(sl_mod, type = "EFAtools", factor_corres = fc,
                variance = "correlation"))
  expect_identical(
    .reliability_core(spec, "sums_load"),
    .OMEGA_FLEX(sl_mod, type = "EFAtools", factor_corres = fc,
                variance = "sums_load"))
  expect_identical(
    .reliability_core(spec, "correlation", add_ind = FALSE),
    .OMEGA_FLEX(sl_mod, type = "EFAtools", factor_corres = fc,
                variance = "correlation", add_ind = FALSE))

  # The stored correlation matrix is picked up from the SL object.
  expect_identical(spec$cormat, sl_mod$orig_R)
})

test_that(".rel_adapt_schmid reproduces the psych::schmid OMEGA path", {
  # Cormat reconstructed from the schmid pattern and Phi.
  expect_equal(
    .reliability_core(.rel_adapt_schmid(schmid_mod, type = "psych"), "correlation"),
    .OMEGA_FLEX(schmid_mod, type = "psych", variance = "correlation"))

  # User-supplied cormat.
  expect_equal(
    .reliability_core(
      .rel_adapt_schmid(schmid_mod, type = "psych",
                        cormat = test_models$baseline$cormat), "correlation"),
    .OMEGA_FLEX(schmid_mod, type = "psych", cormat = test_models$baseline$cormat,
                variance = "correlation"))
})

test_that(".rel_adapt_manual reproduces the model = NULL OMEGA path", {
  spec <- .rel_adapt_manual(g_load = sl_mod$sl[, "g"],
                            s_load = sl_mod$sl[, c("F1", "F2", "F3")],
                            u2 = sl_mod$sl[, "u2"],
                            var_names = rownames(sl_mod$sl),
                            factor_corres = fc, type = "EFAtools",
                            cormat = test_models$baseline$cormat)

  expect_identical(
    .reliability_core(spec, "correlation"),
    .OMEGA_FLEX(model = NULL, type = "EFAtools", var_names = rownames(sl_mod$sl),
                g_load = sl_mod$sl[, "g"], s_load = sl_mod$sl[, c("F1", "F2", "F3")],
                u2 = sl_mod$sl[, "u2"], cormat = test_models$baseline$cormat,
                factor_corres = fc, variance = "correlation"))
  expect_identical(
    .reliability_core(spec, "sums_load"),
    .OMEGA_FLEX(model = NULL, type = "EFAtools", var_names = rownames(sl_mod$sl),
                g_load = sl_mod$sl[, "g"], s_load = sl_mod$sl[, c("F1", "F2", "F3")],
                u2 = sl_mod$sl[, "u2"], cormat = test_models$baseline$cormat,
                factor_corres = fc, variance = "sums_load"))
})

test_that(".rel_adapt_manual checks cormat against the loadings, not the output labels", {
  # `var_names` names the result's rows; it is not a claim about what the correlation
  # matrix is labelled with. Relabelling the output must therefore not reject a correct
  # matrix -- the axis to match is the loadings' own row names.
  s_load <- unclass(sl_mod$sl)[, c("F1", "F2", "F3")]
  manual <- function(...) {
    .rel_adapt_manual(g_load = sl_mod$sl[, "g"], s_load = s_load,
                      u2 = sl_mod$sl[, "u2"], factor_corres = fc, type = "EFAtools",
                      ...)
  }
  cm <- test_models$baseline$cormat

  expect_identical(manual(var_names = paste0("Item", 1:18), cormat = cm)$cormat, cm)
  # The loadings' names still drive a permutation back into the solution's order.
  expect_identical(manual(var_names = paste0("Item", 1:18),
                          cormat = cm[18:1, 18:1])$cormat, cm)
  # Unlabelled components are matched by position: only the dimension is checked.
  bare <- s_load
  rownames(bare) <- NULL
  expect_identical(
    .rel_adapt_manual(g_load = sl_mod$sl[, "g"], s_load = bare, u2 = sl_mod$sl[, "u2"],
                      var_names = paste0("Item", 1:18), factor_corres = fc,
                      type = "EFAtools", cormat = cm[18:1, 18:1])$cormat,
    cm[18:1, 18:1])
  expect_error(manual(var_names = rownames(cm), cormat = cm[1:6, 1:6]),
               class = "efa_reliability_cormat_dim")
})

test_that(".rel_adapt_efa gives model-appropriate correlated-factors coefficients", {
  spec <- .rel_adapt_efa(efa_mod)

  # No general factor: g_load is zero, so omega hierarchical and ECV vanish.
  expect_true(all(spec$g_load == 0))
  expect_no_warning(om <- .reliability_core(spec, "correlation"))
  expect_true(all(om[, "hier"] == 0))
  expect_equal(unname(om["g", "ECV"]), 0)

  # Whole-scale omega total is the model-implied common variance of all variables over
  # their observed total variance.
  common <- spec$s_load %*% spec$Phi %*% t(spec$s_load)
  expect_equal(unname(om["g", "tot"]), sum(common) / sum(spec$cormat))

  # Each group row's omega subscale is that factor's congeneric omega from its own
  # indicators, while its omega total is the composite's full model-implied common
  # variance. The two differ whenever the composite draws on another factor, and on
  # this fit the total is the larger of the two (it need not be: the total is a
  # quadratic form in Phi and the cross terms can carry either sign).
  for (j in seq_len(ncol(spec$s_load))) {
    mem <- which(spec$map[, j] == 1)
    Vgr <- sum(spec$cormat[mem, mem])
    cong <- sum(spec$s_load[mem, j])^2 / Vgr
    expect_equal(unname(om[spec$fac_names[j], "sub"]), cong)
    expect_equal(unname(om[spec$fac_names[j], "hier"]), 0)
    expect_equal(unname(om[spec$fac_names[j], "tot"]), sum(common[mem, mem]) / Vgr)
    expect_gt(om[spec$fac_names[j], "tot"], cong)
  }

  # Works for a two-factor oblique EFA too (no Schmid-Leiman, so no identification
  # floor at three factors).
  efa2 <- EFA(test_models$baseline$cormat, N = 500, n_factors = 2,
              type = "EFAtools", method = "PAF", rotation = "promax")
  expect_no_error(.reliability_core(.rel_adapt_efa(efa2), "correlation"))
})

test_that(".rel_adapt_efa carries Phi in the column order of the loadings", {
  # The loading columns are ordered by the factor number in their labels, and the
  # group-row omega totals are a quadratic form in Phi, so Phi must be permuted the
  # same way or every correlated-factors total is scored against the wrong factors.
  efa_perm <- efa_mod
  colnames(efa_perm$rot_loadings) <- c("F2", "F3", "F1")
  spec <- .rel_adapt_efa(efa_perm)

  expect_equal(unname(spec$s_load), unname(unclass(efa_mod$rot_loadings)[, c(3, 1, 2)]))
  expect_equal(unname(spec$Phi), unname(efa_mod$Phi[c(3, 1, 2), c(3, 1, 2)]))
})

test_that(".rel_adapt_efa handles rotations that leave the loading columns unlabelled", {
  skip_if_not_installed("GPArotation")

  # efa_fit() labels the factor columns for every rotation, but the adapter also sees
  # loading matrices from other sources (a bare pattern matrix, an externally built fit),
  # where the labels can be missing. Unlabelled columns must not cost the adapter its
  # group factors, so construct that state explicitly.
  efa_obl <- efa_fit(test_models$baseline$cormat, N = 500, n_factors = 3,
                     estimator = "PAF", rotation = "oblimin")
  colnames(efa_obl$rot_loadings) <- NULL
  dimnames(efa_obl$Phi) <- NULL

  spec <- .rel_adapt_efa(efa_obl)

  # All three factors survive, and every group factor still gets a row label.
  expect_identical(dim(spec$s_load), c(18L, 3L))
  expect_identical(spec$fac_names, c("F1", "F2", "F3"))
  expect_identical(spec$s_load, unclass(efa_obl$rot_loadings))

  # Communalities come from the real pattern, not from an empty loading matrix.
  expect_true(all(spec$u2 > 0 & spec$u2 < 1))

  om <- .reliability_core(spec, "correlation")
  expect_identical(rownames(om), c("g", "F1", "F2", "F3"))
  # Each group factor scores a real congeneric omega. Tested on the group rows
  # only: the g row carries a "sub" value whether or not the group factors
  # survived, so it cannot discriminate a lost factor.
  expect_true(all(om[c("F1", "F2", "F3"), "sub"] > 0 &
                    om[c("F1", "F2", "F3"), "sub"] < 1))

  # Whole-scale omega total is 1' L Phi L' 1 / 1'R1, and L Phi L' is invariant to the
  # choice of oblique rotation, so the promax fit of the same extraction must agree on
  # it.
  om_pro <- .reliability_core(.rel_adapt_efa(efa_mod), "correlation")
  expect_equal(unname(om["g", "tot"]), unname(om_pro["g", "tot"]),
               tolerance = 1e-8)
})

test_that(".rel_adapt_bifactor reproduces the manual OMEGA path", {
  # A raw orthogonal-bifactor loading matrix: g in the first column.
  g_load <- rep(0.5, 6)
  s_load <- matrix(0, 6, 2)
  s_load[1:3, 1] <- 0.4
  s_load[4:6, 2] <- 0.4
  L <- cbind(g = g_load, s_load)
  colnames(L) <- c("g", "F1", "F2")
  rownames(L) <- paste0("V", 1:6)
  u2 <- 1 - rowSums(L^2)
  Rm <- L %*% t(L) + diag(u2)
  map <- abs(s_load) > 0

  spec <- .rel_adapt_bifactor(L, fac_names = c("F1", "F2"))
  expect_identical(spec$u2, u2)
  expect_identical(spec$cormat, Rm)

  ref_corr <- .OMEGA_FLEX(model = NULL, type = "EFAtools", var_names = rownames(L),
                          g_load = g_load, s_load = s_load, u2 = u2, cormat = Rm,
                          factor_corres = map, fac_names = c("F1", "F2"),
                          variance = "correlation")
  ref_sums <- .OMEGA_FLEX(model = NULL, type = "EFAtools", var_names = rownames(L),
                          g_load = g_load, s_load = s_load, u2 = u2,
                          factor_corres = map, fac_names = c("F1", "F2"),
                          variance = "sums_load")

  expect_identical(.reliability_core(spec, "correlation"), ref_corr)
  expect_identical(.reliability_core(spec, "sums_load"), ref_sums)
})

test_that("adapters reject a non-correlation cormat via .is_cormat", {
  bad <- matrix(rnorm(36), 6, 6)
  expect_error(.rel_adapt_SL(sl_mod, factor_corres = fc, cormat = bad),
               class = "efa_reliability_not_cormat")
  expect_error(.rel_adapt_bifactor(cbind(rep(0.5, 6), matrix(0.4, 6, 1)),
                                   cormat = bad),
               class = "efa_reliability_not_cormat")
  # the guard runs before the coercion, so a data frame that is not a correlation matrix is
  # still rejected rather than silently coerced
  expect_error(.rel_adapt_SL(sl_mod, factor_corres = fc, cormat = as.data.frame(bad)),
               class = "efa_reliability_not_cormat")
})

test_that(".rel_check_cormat matches the supplied matrix to the solution's variables", {
  cm <- test_models$baseline$cormat
  nms <- rownames(cm)

  # Same variables, other order: an unambiguous permutation, so it is reordered.
  expect_identical(.rel_check_cormat(cm[18:1, 18:1], nms, 18L), cm)
  # Already in the solution's order, or unlabelled: handed back untouched.
  expect_identical(.rel_check_cormat(cm, nms, 18L), cm)
  expect_identical(.rel_check_cormat(unname(cm), nms, 18L), unname(cm))

  # A different size, or a different set of variables, cannot be matched.
  expect_error(.rel_check_cormat(cm[1:6, 1:6], nms, 18L),
               class = "efa_reliability_cormat_dim")
  other <- cm
  dimnames(other) <- list(paste0("X", 1:18), paste0("X", 1:18))
  expect_error(.rel_check_cormat(other, nms, 18L),
               class = "efa_reliability_cormat_names")
  # Row and column names that disagree do not identify the variables either.
  mixed <- cm
  colnames(mixed) <- paste0("X", 1:18)
  expect_error(.rel_check_cormat(mixed, nms, 18L),
               class = "efa_reliability_cormat_names")

  # Without a solution to check against, only the correlation-matrix check applies.
  expect_identical(.rel_check_cormat(cm[18:1, 18:1]), cm[18:1, 18:1])
})

test_that("a cormat supplied as a data frame is accepted and handed on as a matrix", {
  # read.csv(file, row.names = 1) returns a data frame, and this guard is the only check a
  # user-supplied cormat passes on the reliability route -- it does not go through
  # .prepare_cor_input(). Without the coercion the omega columns still come out right (they
  # only sum and subset the matrix) and only the standardized-alpha branch fails, on diag();
  # efa_reliability() always requests that branch, so the whole call breaks.
  cm <- test_models$baseline$cormat
  cdf <- as.data.frame(cm)

  expect_true(is.matrix(.rel_check_cormat(cdf)))
  expect_identical(.rel_check_cormat(cdf), cm)

  # the adapters, and the exported entry point, give the matrix route's answer
  expect_equal(.rel_adapt_SL(sl_mod, factor_corres = fc, cormat = cdf),
               .rel_adapt_SL(sl_mod, factor_corres = fc, cormat = cm))
  expect_equal(efa_reliability(efa_mod, cormat = cdf),
               efa_reliability(efa_mod, cormat = cm))
})

test_that(".rel_adapt_efa requires an oblique solution", {
  efa_orth <- EFA(test_models$baseline$cormat, N = 500, n_factors = 3,
                  type = "EFAtools", method = "PAF", rotation = "varimax")
  expect_error(.rel_adapt_efa(efa_orth), class = "efa_reliability_not_oblique")
})

test_that(".rel_adapt_efa reads a one-factor solution from its unrotated loadings", {
  cm <- test_models$baseline$cormat
  efa_1 <- EFA(cm, N = 500, n_factors = 1, type = "EFAtools", method = "PAF")
  spec <- .rel_adapt_efa(efa_1)

  # One factor is nothing to rotate, so the unrotated loadings are the solution and the
  # oblique requirement does not apply. The spec is the ordinary one-factor one; the
  # front-end rewrites it into the single-factor spec the core scores.
  expect_identical(dim(spec$s_load), c(18L, 1L))
  expect_equal(unname(spec$s_load), unname(unclass(efa_1$unrot_loadings)))
  expect_equal(spec$u2, 1 - as.numeric(unclass(efa_1$unrot_loadings))^2,
               ignore_attr = TRUE)

  # Asking for a rotation leaves rot_loadings a copy of the unrotated ones and no Phi, so
  # the same solution arrives at the same spec.
  efa_1_rot <- suppressWarnings(suppressMessages(
    EFA(cm, N = 500, n_factors = 1, type = "EFAtools", method = "PAF",
        rotation = "promax")))
  expect_equal(.rel_adapt_efa(efa_1_rot), spec)

  # variance = "correlation" scores it against the object's own correlation matrix ...
  expect_equal(spec$cormat, efa_1$orig_R)
  # ... a supplied one takes precedence, and with neither the model-implied matrix stands
  # in, exactly as for an oblique solution.
  expect_equal(.rel_adapt_efa(efa_1, cormat = test_models$baseline$cormat)$cormat, cm)
  efa_no_R <- efa_1
  efa_no_R$orig_R <- NA
  expect_equal(.rel_adapt_efa(efa_no_R)$cormat,
               tcrossprod(as.numeric(unclass(efa_1$unrot_loadings))) + diag(spec$u2),
               ignore_attr = TRUE)
})

## lavaan adapters -----------------------------------------------------------

test_that(".rel_adapt_lavaan reproduces the bifactor OMEGA path", {
  skip_if_not_installed("lavaan")
  mod <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
          F2 =~ V7 + V8 + V9 + V10 + V11 + V12
          F3 =~ V13 + V14 + V15 + V16 + V17 + V18
          g =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + V10 + V11 + V12 +
               V13 + V14 + V15 + V16 + V17 + V18'
  fit <- suppressWarnings(lavaan::cfa(mod, sample.cov = test_models$baseline$cormat,
                                      sample.nobs = 500, estimator = "ml",
                                      orthogonal = TRUE))

  adapt <- .rel_adapt_lavaan(fit, g_name = "g")
  expect_length(adapt$groups, 1L)
  expect_false(adapt$groups[[1]]$single)
  expect_identical(adapt$variance, "sums_load")

  expect_equal(.reliability_core(adapt$groups[[1]], "sums_load"),
               .OMEGA_LAVAAN(fit, g_name = "g"))
})

test_that(".rel_adapt_lavaan Schmid-Leiman transforms a second-order model", {
  skip_if_not_installed("lavaan")
  mod <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
          F2 =~ V7 + V8 + V9 + V10 + V11 + V12
          F3 =~ V13 + V14 + V15 + V16 + V17 + V18
          g =~ F1 + F2 + F3'
  fit <- suppressWarnings(lavaan::cfa(mod, sample.cov = test_models$baseline$cormat,
                                      sample.nobs = 500, estimator = "ml"))

  spec <- .rel_adapt_lavaan(fit, g_name = "g")$groups[[1]]
  expect_equal(.reliability_core(spec, "sums_load"),
               suppressMessages(.OMEGA_LAVAAN(fit, g_name = "g")))
})

test_that(".rel_adapt_lavaan handles a single factor as a special path", {
  skip_if_not_installed("lavaan")
  mod <- 'g =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + V10 + V11 + V12 +
               V13 + V14 + V15 + V16 + V17 + V18'
  fit <- suppressWarnings(lavaan::cfa(mod, sample.cov = test_models$baseline$cormat,
                                      sample.nobs = 500, estimator = "ml",
                                      orthogonal = TRUE))

  sf <- .rel_adapt_lavaan(fit)$groups[[1]]
  expect_true(sf$single)

  ref <- suppressMessages(.OMEGA_LAVAAN(fit))  # c(Omega, H)
  omega <- sum(sf$g_load)^2 / (sum(sf$g_load)^2 + sum(sf$u2))
  expect_equal(unname(omega), unname(ref["Omega"]))
  expect_equal(unname(.h_index(sf$g_load)), unname(ref["H"]))
})

test_that(".rel_adapt_lavaan returns one spec per group", {
  skip_if_not_installed("lavaan")
  mod <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
          F2 =~ V7 + V8 + V9 + V10 + V11 + V12
          F3 =~ V13 + V14 + V15 + V16 + V17 + V18
          g =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + V10 + V11 + V12 +
               V13 + V14 + V15 + V16 + V17 + V18'
  fit <- suppressWarnings(lavaan::cfa(mod, sample.cov =
                                        list(test_models$baseline$cormat,
                                             test_models$baseline$cormat),
                                      sample.nobs = c(500, 500),
                                      estimator = "ml", orthogonal = TRUE))

  adapt <- .rel_adapt_lavaan(fit, g_name = "g", group_names = c("Some", "Others"))
  expect_identical(adapt$group_names, c("Some", "Others"))
  expect_length(adapt$groups, 2L)

  ref <- .OMEGA_LAVAAN(fit, g_name = "g", group_names = c("Some", "Others"))
  expect_equal(.reliability_core(adapt$groups[[1]], "sums_load"), ref$Some)
  expect_equal(.reliability_core(adapt$groups[[2]], "sums_load"), ref$Others)
})

test_that(".rel_adapt_lavaan normalizes a correlated-factors fit to an oblique spec", {
  skip_if_not_installed("lavaan")
  mod <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
          F2 =~ V7 + V8 + V9 + V10 + V11 + V12
          F3 =~ V13 + V14 + V15 + V16 + V17 + V18'
  fit <- suppressWarnings(lavaan::cfa(mod, sample.cov = test_models$baseline$cormat,
                                      sample.nobs = 500, estimator = "ml"))
  std <- suppressWarnings(lavaan::lavInspect(fit, "std",
                                             drop.list.single.group = FALSE))[[1]]

  adapt <- .rel_adapt_lavaan(fit)
  expect_true(adapt$correlated)
  expect_false(adapt$higher_order)
  # A simple structure would otherwise trip the "did you enter a bifactor model?" note.
  expect_false(adapt$few_loadings)

  spec <- adapt$groups[[1]]
  # The correlated-factors spec: no general factor, the oblique pattern as the group
  # loadings, and the standardized psi as their factor correlations.
  expect_true(all(spec$g_load == 0))
  expect_equal(spec$s_load, std$lambda)
  expect_equal(spec$Phi, std$psi)
  expect_equal(spec$u2, diag(std$theta))
  expect_identical(spec$fac_names, colnames(std$lambda))
  # The map is the model's own fixed-zero structure, as on the bifactor path: a
  # confirmatory model's unmodelled loadings are exact zeros, so the property holds
  # without repeating the adapter's tolerance here.
  expect_equal(spec$map, std$lambda != 0)
  # The factor correlations are substantial here, so a spec that dropped them would
  # score a materially different model.
  expect_gt(max(abs(std$psi[upper.tri(std$psi)])), .3)
})

test_that(".rel_adapt_lavaan reads the model structure, not the factor names", {
  skip_if_not_installed("lavaan")
  # An ordinary factor named "g" does not make a fit a bifactor one: no variable loads
  # on two or more factors, so it is a correlated-factors solution either way.
  mod_g <- 'g  =~ V1 + V2 + V3 + V4 + V5 + V6
            F2 =~ V7 + V8 + V9 + V10 + V11 + V12
            F3 =~ V13 + V14 + V15 + V16 + V17 + V18'
  fit_g <- suppressWarnings(lavaan::cfa(mod_g, sample.cov = test_models$baseline$cormat,
                                        sample.nobs = 500, estimator = "ml"))
  expect_true(.rel_adapt_lavaan(fit_g, g_name = "g")$correlated)

  # A second-order model has simple structure too, so it must not be told apart by that
  # alone: it routes its factor covariances through beta, which is what marks it. Were it
  # read as a correlated-factors fit, its psi -- the residual covariances of the
  # first-order factors, not their correlations -- would be scored as factor correlations.
  mod_ho <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
             F2 =~ V7 + V8 + V9 + V10 + V11 + V12
             F3 =~ V13 + V14 + V15 + V16 + V17 + V18
             g =~ F1 + F2 + F3'
  fit_ho <- suppressWarnings(lavaan::cfa(mod_ho, sample.cov = test_models$baseline$cormat,
                                         sample.nobs = 500, estimator = "ml"))
  expect_false(.rel_adapt_lavaan(fit_ho, g_name = "g")$correlated)
  expect_true(.rel_adapt_lavaan(fit_ho, g_name = "g")$higher_order)
  # And it is still named as a misspelled general factor rather than quietly rescored.
  expect_error(.rel_adapt_lavaan(fit_ho, g_name = "G"), class = "efa_reliability_g_name")

  # A bifactor fit has variables on two factors, so it keeps needing its general factor
  # named, whatever that name is.
  mod_bi <- 'F1 =~ V1 + V2 + V3 + V4 + V5 + V6
             F2 =~ V7 + V8 + V9 + V10 + V11 + V12
             F3 =~ V13 + V14 + V15 + V16 + V17 + V18
             gen =~ V1 + V2 + V3 + V4 + V5 + V6 + V7 + V8 + V9 + V10 + V11 + V12 +
                    V13 + V14 + V15 + V16 + V17 + V18'
  fit_bi <- suppressWarnings(lavaan::cfa(mod_bi, sample.cov = test_models$baseline$cormat,
                                         sample.nobs = 500, estimator = "ml",
                                         orthogonal = TRUE))
  expect_false(.rel_adapt_lavaan(fit_bi, g_name = "gen")$correlated)
  expect_error(.rel_adapt_lavaan(fit_bi, g_name = "g"), class = "efa_reliability_g_name")
})

rm(efa_mod, sl_mod, schmid_mod, fc)