| 1 |
context("rma.uni objects") |
| 2 |
set.seed(20190513) |
| 3 |
|
| 4 |
skip_if_not_installed("robumeta") |
| 5 |
skip_if_not_installed("metafor") |
| 6 |
|
| 7 |
library(robumeta, quietly=TRUE) |
| 8 |
suppressMessages(library(metafor, quietly=TRUE)) |
| 9 |
|
| 10 |
data(corrdat) |
| 11 |
corr_robu <- robu(effectsize ~ males + college + binge, data = corrdat, |
| 12 |
modelweights = "CORR", studynum = studyid, |
| 13 |
var.eff.size = var) |
| 14 |
corrdat$wt <- corr_robu$data.full$r.weights |
| 15 |
|
| 16 |
corr_meta <- rma(effectsize ~ males + college + binge, data = corrdat, |
| 17 |
weights = wt, vi = var, method = "FE") |
| 18 |
|
| 19 |
test_that("CR2 t-tests agree with robumeta for correlated effects", { |
| 20 |
robu_CR2 <- vcovCR(corr_meta, cluster = corrdat$studyid, target = 1 / corrdat$wt, type = "CR2") |
| 21 |
expect_true(check_CR(corr_meta, vcov = robu_CR2)) |
| 22 |
# expect_true(check_CR(corr_meta, vcov = "CR4", cluster = corrdat$studyid)) |
| 23 |
expect_equivalent(as.matrix(robu_CR2), corr_robu$VR.r) |
| 24 |
expect_equivalent(as.matrix(vcovCR(corr_meta, cluster = corrdat$studyid, |
| 25 |
inverse_var = TRUE, type = "CR2")), corr_robu$VR.r) |
| 26 |
|
| 27 |
CR2_ttests <- coef_test(corr_meta, vcov = robu_CR2, test = "Satterthwaite") |
| 28 |
expect_equal(corr_robu$dfs, CR2_ttests$df) |
| 29 |
expect_equal(corr_robu$reg_table$prob, CR2_ttests$p_Satt) |
| 30 |
}) |
| 31 |
|
| 32 |
data(hierdat) |
| 33 |
hier_meta <- rma(effectsize ~ binge + followup + sreport + age, data = hierdat, |
| 34 |
vi = var, method = "REML") |
| 35 |
hierdat$wt <- with(hier_meta, 1 / (vi + tau2)) |
| 36 |
hier_robu <- robu(effectsize ~ binge + followup + sreport + age, |
| 37 |
data = hierdat, studynum = studyid, |
| 38 |
var.eff.size = var, userweights = wt) |
| 39 |
|
| 40 |
test_that("CR2 t-tests agree with robumeta for user weighting", { |
| 41 |
|
| 42 |
robu_CR2_iv <- vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid) |
| 43 |
robu_CR2_not <- vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, |
| 44 |
target = hier_robu$data.full$avg.var.eff.size) |
| 45 |
expect_true(check_CR(hier_meta, vcov = robu_CR2_iv)) |
| 46 |
# expect_true(check_CR(hier_meta, vcov = "CR4", cluster = hierdat$studyid)) |
| 47 |
expect_true(check_CR(hier_meta, vcov = robu_CR2_not)) |
| 48 |
# expect_true(check_CR(hier_meta, vcov = "CR4", cluster = hierdat$studyid, |
| 49 |
# target = hier_robu$data.full$avg.var.eff.size)) |
| 50 |
|
| 51 |
expect_that(all.equal(hier_robu$VR.r, as.matrix(robu_CR2_iv)), is_a("character")) |
| 52 |
expect_equivalent(hier_robu$VR.r, as.matrix(robu_CR2_not)) |
| 53 |
|
| 54 |
# CR2_ttests <- coef_test(hier_meta, vcov = robu_CR2_not, test = "Satterthwaite") |
| 55 |
# expect_equal(hier_robu$dfs, CR2_ttests$df) |
| 56 |
# expect_equal(hier_robu$reg_table$prob, CR2_ttests$p_Satt) |
| 57 |
}) |
| 58 |
|
| 59 |
test_that("bread works", { |
| 60 |
expect_true(check_bread(corr_meta, cluster = corrdat$studyid, y = corrdat$effectsize)) |
| 61 |
X <- model_matrix(corr_meta) |
| 62 |
W <- corr_meta$weights |
| 63 |
V <- corr_meta$vi |
| 64 |
vcov_corr <- crossprod((sqrt(V) * W * X) %*% bread(corr_meta) / nobs(corr_meta)) |
| 65 |
attr(vcov_corr, "dimnames") <- attr(vcov(corr_meta), "dimnames") |
| 66 |
expect_equal(vcov(corr_meta), vcov_corr) |
| 67 |
|
| 68 |
expect_true(check_bread(hier_meta, cluster = hierdat$studyid, y = hierdat$effectsize)) |
| 69 |
expect_equal(vcov(hier_meta), bread(hier_meta) / nobs(hier_meta)) |
| 70 |
}) |
| 71 |
|
| 72 |
CR_types <- paste0("CR",0:4) |
| 73 |
|
| 74 |
test_that("order doesn't matter", { |
| 75 |
|
| 76 |
skip_on_cran() |
| 77 |
|
| 78 |
check_sort_order(hier_meta, hierdat, cluster = "studyid") |
| 79 |
|
| 80 |
}) |
| 81 |
|
| 82 |
test_that("clubSandwich works with dropped covariates", { |
| 83 |
dat_miss <- hierdat |
| 84 |
dat_miss$binge[sample.int(nrow(hierdat), size = round(nrow(hierdat) / 10))] <- NA |
| 85 |
dat_miss$followup[sample.int(nrow(hierdat), size = round(nrow(hierdat) / 20))] <- NA |
| 86 |
expect_warning(hier_drop <- rma(effectsize ~ binge + followup + sreport + age, |
| 87 |
data = dat_miss, vi = var, method = "REML")) |
| 88 |
|
| 89 |
|
| 90 |
subset_ind <- with(dat_miss, !is.na(binge) & !is.na(followup)) |
| 91 |
hier_complete <- rma(effectsize ~ binge + followup + sreport + age, |
| 92 |
subset = !is.na(binge) & !is.na(followup), |
| 93 |
data = dat_miss, vi = var, method = "REML") |
| 94 |
expect_error(vcovCR(hier_complete, type = "CR0", cluster = dat_miss$studyid)) |
| 95 |
|
| 96 |
CR_drop_A <- lapply(CR_types, function(x) vcovCR(hier_drop, type = x, cluster = dat_miss$studyid)) |
| 97 |
CR_drop_B <- lapply(CR_types, function(x) vcovCR(hier_drop, type = x, cluster = hierdat$studyid)) |
| 98 |
CR_complete <- lapply(CR_types, function(x) vcovCR(hier_complete, type = x, cluster = dat_miss$studyid[subset_ind])) |
| 99 |
expect_equal(CR_drop_A, CR_complete) |
| 100 |
expect_equal(CR_drop_B, CR_complete) |
| 101 |
|
| 102 |
test_drop_A <- lapply(CR_types, function(x) coef_test(hier_drop, vcov = x, cluster = dat_miss$studyid, test = "All", p_values = FALSE)) |
| 103 |
test_drop_B <- lapply(CR_types, function(x) coef_test(hier_drop, vcov = x, cluster = hierdat$studyid, test = "All", p_values = FALSE)) |
| 104 |
test_complete <- lapply(CR_types, function(x) coef_test(hier_complete, vcov = x, cluster = dat_miss$studyid[subset_ind], test = "All", p_values = FALSE)) |
| 105 |
compare_ttests(test_drop_A, test_complete) |
| 106 |
compare_ttests(test_drop_B, test_complete) |
| 107 |
}) |
| 108 |
|
| 109 |
test_that("clubSandwich works with missing variances", { |
| 110 |
|
| 111 |
dat_miss <- hierdat |
| 112 |
dat_miss$var[sample.int(nrow(hierdat), size = round(nrow(hierdat) / 10))] <- NA |
| 113 |
expect_warning(hier_drop <- rma(effectsize ~ binge + followup + sreport + age, |
| 114 |
data = dat_miss, vi = var, method = "REML")) |
| 115 |
|
| 116 |
|
| 117 |
subset_ind <- with(dat_miss, !is.na(var)) |
| 118 |
hier_complete <- rma(effectsize ~ binge + followup + sreport + age, |
| 119 |
subset = !is.na(var), |
| 120 |
data = dat_miss, vi = var, method = "REML") |
| 121 |
expect_error(vcovCR(hier_complete, type = "CR0", cluster = dat_miss$studyid)) |
| 122 |
|
| 123 |
CR_drop_A <- lapply(CR_types, function(x) vcovCR(hier_drop, type = x, cluster = dat_miss$studyid)) |
| 124 |
CR_drop_B <- lapply(CR_types, function(x) vcovCR(hier_drop, type = x, cluster = hierdat$studyid)) |
| 125 |
CR_complete <- lapply(CR_types, function(x) vcovCR(hier_complete, type = x, cluster = dat_miss$studyid[subset_ind])) |
| 126 |
expect_equal(CR_drop_A, CR_complete) |
| 127 |
expect_equal(CR_drop_B, CR_complete) |
| 128 |
|
| 129 |
}) |
| 130 |
|
| 131 |
|
| 132 |
test_that("vcovCR options work for CR2", { |
| 133 |
RE_var <- hier_meta$tau2 + hierdat$var |
| 134 |
CR2_iv <- vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid) |
| 135 |
expect_equal(vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, inverse_var = TRUE), CR2_iv) |
| 136 |
|
| 137 |
CR2_not <- vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, inverse_var = FALSE) |
| 138 |
attr(CR2_iv, "inverse_var") <- FALSE |
| 139 |
attr(CR2_iv, "target") <- attr(CR2_not, "target") |
| 140 |
expect_equal(CR2_not, CR2_iv) |
| 141 |
expect_equal(vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, target = RE_var), CR2_not) |
| 142 |
expect_equal(vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, target = RE_var, inverse_var = FALSE), CR2_not) |
| 143 |
expect_false(identical(vcovCR(hier_meta, type = "CR2", cluster = hierdat$studyid, target = hierdat$var), CR2_not)) |
| 144 |
}) |
| 145 |
|
| 146 |
|
| 147 |
test_that("vcovCR works with intercept-only model and user-specified weights.", { |
| 148 |
|
| 149 |
dat <- escalc(measure="RR", ai=tpos, bi=tneg, ci=cpos, di=cneg, data=dat.bcg) |
| 150 |
dat$wt <- sample(1:3, size = nrow(dat), replace = TRUE) |
| 151 |
|
| 152 |
res <- rma(yi, vi, weights = wt, data=dat) |
| 153 |
|
| 154 |
meta_rob <- robust(res, cluster=dat$trial) |
| 155 |
club_rob <- coef_test(res, vcov="CR1", cluster=dat$trial, test = "naive-t") |
| 156 |
expect_equal(meta_rob$se, club_rob$SE) |
| 157 |
expect_equal(meta_rob$zval, club_rob$tstat) |
| 158 |
expect_equal(meta_rob$dfs, club_rob$df_t) |
| 159 |
expect_equal(meta_rob$pval, club_rob$p_t) |
| 160 |
|
| 161 |
expect_true(check_CR(res, vcov = "CR2", cluster = dat$trial)) |
| 162 |
test_uni <- coef_test(res, vcov="CR2", cluster=dat$trial, test = "All") |
| 163 |
|
| 164 |
|
| 165 |
res <- rma.mv(yi, vi, W = wt, random = ~ 1 | trial, data=dat) |
| 166 |
meta_rob <- robust(res, cluster=dat$trial) |
| 167 |
club_rob <- coef_test(res, vcov="CR1", test = "naive-t") |
| 168 |
expect_equal(meta_rob$se, club_rob$SE) |
| 169 |
expect_equal(meta_rob$zval, club_rob$tstat) |
| 170 |
expect_equal(meta_rob$dfs, club_rob$df_t) |
| 171 |
expect_equal(meta_rob$pval, club_rob$p_t) |
| 172 |
|
| 173 |
expect_true(check_CR(res, vcov = "CR2")) |
| 174 |
test_mv <- coef_test(res, vcov="CR2", test = "All") |
| 175 |
|
| 176 |
expect_equal(test_uni, test_mv, tolerance = 10^-5) |
| 177 |
|
| 178 |
V_club <- vcovCR(res, type = "CR2") |
| 179 |
k <- res$k |
| 180 |
yi <- res$yi |
| 181 |
wi <- diag(res$W) |
| 182 |
W <- sum(wi) |
| 183 |
wi <- wi / W |
| 184 |
vi <- diag(res$M) |
| 185 |
V <- sum(vi) |
| 186 |
ei <- residuals_CS(res) |
| 187 |
M <- sum(wi^2 * vi) |
| 188 |
ai <- 1 / sqrt(1 - 2 * wi + M / vi) |
| 189 |
V_hand <- sum(wi^2 * ai^2 * ei^2) |
| 190 |
expect_equal(V_hand, as.numeric(V_club)) |
| 191 |
|
| 192 |
pi_theta_pj <- diag(vi) - tcrossprod(rep(1,k), wi * vi) - tcrossprod(wi * vi, rep(1, k)) + M |
| 193 |
df <- M^2 / sum(tcrossprod(ai^2 * wi^2) * (pi_theta_pj^2)) |
| 194 |
|
| 195 |
expect_equal(Inf, test_uni$df_z) |
| 196 |
expect_equal(k - 1, test_uni$df_t) |
| 197 |
expect_equal(df, test_uni$df_Satt, tolerance = 10^-5) |
| 198 |
|
| 199 |
}) |
| 200 |
|
| 201 |
test_that("clubSandwich agrees with metafor::robust() for CR0.", { |
| 202 |
|
| 203 |
test_CR0 <- conf_int(corr_meta, vcov = "CR0", cluster = corrdat$studyid, test = "naive-tp", p_values = TRUE) |
| 204 |
meta_CR0 <- robust(corr_meta, cluster = corrdat$studyid, adjust = FALSE) |
| 205 |
rob_CR0 <- conf_int(meta_CR0, vcov = "CR0", cluster = corrdat$studyid, test = "naive-tp", p_values = TRUE) |
| 206 |
expect_equal(test_CR0$SE, meta_CR0$se) |
| 207 |
expect_equal(test_CR0$df, rep(meta_CR0$df, length(test_CR0$df))) |
| 208 |
expect_equal(rob_CR0, test_CR0) |
| 209 |
|
| 210 |
club_F_CR0 <- Wald_test(corr_meta, constraints = constrain_zero(2:4), |
| 211 |
vcov = "CR0", cluster = corrdat$studyid, test = "Naive-Fp") |
| 212 |
rob_F_CR0 <- Wald_test(meta_CR0, constraints = constrain_zero(2:4), |
| 213 |
vcov = "CR0", cluster = corrdat$studyid, test = "Naive-Fp") |
| 214 |
expect_equal(club_F_CR0$Fstat, meta_CR0$QM) |
| 215 |
expect_equal(club_F_CR0$df_num, meta_CR0$QMdf[1]) |
| 216 |
expect_equal(club_F_CR0$df_denom, meta_CR0$QMdf[2]) |
| 217 |
expect_equal(club_F_CR0$p_val, meta_CR0$QMp) |
| 218 |
expect_equal(club_F_CR0, rob_F_CR0) |
| 219 |
|
| 220 |
}) |
| 221 |
|
| 222 |
test_that("clubSandwich agrees with metafor::robust() for CR1p.", { |
| 223 |
|
| 224 |
test_CR1 <- conf_int(corr_meta, vcov = "CR1p", cluster = corrdat$studyid, test = "naive-tp", p_values = TRUE) |
| 225 |
meta_CR1 <- robust(corr_meta, cluster = corrdat$studyid, adjust = TRUE) |
| 226 |
rob_CR1 <- conf_int(meta_CR1, vcov = "CR1p", cluster = corrdat$studyid, test = "naive-tp", p_values = TRUE) |
| 227 |
expect_equal(test_CR1$SE, meta_CR1$se) |
| 228 |
expect_equal(test_CR1$df, rep(meta_CR1$df, length(test_CR1$df))) |
| 229 |
expect_equal(rob_CR1, test_CR1) |
| 230 |
|
| 231 |
club_F_CR1 <- Wald_test(corr_meta, constraints = constrain_zero(2:4), |
| 232 |
vcov = "CR1p", cluster = corrdat$studyid, test = "Naive-Fp") |
| 233 |
rob_F_CR1 <- Wald_test(meta_CR1, constraints = constrain_zero(2:4), |
| 234 |
vcov = "CR1p", cluster = corrdat$studyid, test = "Naive-Fp") |
| 235 |
expect_equal(club_F_CR1$Fstat, meta_CR1$QM) |
| 236 |
expect_equal(club_F_CR1$df_num, meta_CR1$QMdf[1]) |
| 237 |
expect_equal(club_F_CR1$df_denom, meta_CR1$QMdf[2]) |
| 238 |
expect_equal(club_F_CR1$p_val, meta_CR1$QMp) |
| 239 |
expect_equal(club_F_CR1, rob_F_CR1) |
| 240 |
}) |
| 241 |
|
| 242 |
test_that("clubSandwich agrees with metafor::robust() for CR2.", { |
| 243 |
|
| 244 |
skip_if(packageVersion('metafor') < "3.1.31") |
| 245 |
test_CR2 <- conf_int(corr_meta, vcov = "CR2", cluster = corrdat$studyid, p_values = TRUE) |
| 246 |
meta_CR2 <- robust(corr_meta, cluster = corrdat$studyid, clubSandwich = TRUE) |
| 247 |
rob_CR2 <- conf_int(meta_CR2, vcov = "CR2", cluster = corrdat$studyid, p_values = TRUE) |
| 248 |
expect_equal(test_CR2$SE, meta_CR2$se) |
| 249 |
expect_equal(rob_CR2, test_CR2) |
| 250 |
|
| 251 |
club_F_CR2 <- Wald_test(corr_meta, constraints = constrain_zero(2:4), |
| 252 |
vcov = "CR2", cluster = corrdat$studyid, test = "All") |
| 253 |
rob_F_CR2 <- Wald_test(meta_CR2, constraints = constrain_zero(2:4), |
| 254 |
vcov = "CR2", cluster = corrdat$studyid, test = "All") |
| 255 |
expect_equal(subset(club_F_CR2, test == "HTZ")$Fstat, meta_CR2$QM) |
| 256 |
expect_equal(subset(club_F_CR2, test == "HTZ")$df_num, meta_CR2$QMdf[1]) |
| 257 |
expect_equal(subset(club_F_CR2, test == "HTZ")$df_denom, meta_CR2$QMdf[2]) |
| 258 |
expect_equal(subset(club_F_CR2, test == "HTZ")$p_val, meta_CR2$QMp) |
| 259 |
expect_equal(club_F_CR2, rob_F_CR2) |
| 260 |
}) |
| 261 |
|
| 262 |
test_that("clubSandwich methods work on robust.rma objects.", { |
| 263 |
|
| 264 |
hier_robust <- robust(hier_meta, cluster = hierdat$studyid, adjust = TRUE) |
| 265 |
|
| 266 |
expect_equal(residuals_CS(hier_meta), residuals_CS(hier_robust)) |
| 267 |
expect_equal(coef_CS(hier_meta), coef_CS(hier_robust)) |
| 268 |
expect_equal(model_matrix(hier_meta), model_matrix(hier_robust)) |
| 269 |
expect_equal(bread(hier_meta), bread(hier_robust)) |
| 270 |
expect_equal(v_scale(hier_meta), v_scale(hier_robust)) |
| 271 |
expect_equal(targetVariance(hier_meta, cluster = hierdat$studyid), |
| 272 |
targetVariance(hier_robust, cluster = hierdat$studyid)) |
| 273 |
expect_equal(weightMatrix(hier_meta, cluster = hierdat$studyid), |
| 274 |
weightMatrix(hier_robust, cluster = hierdat$studyid)) |
| 275 |
|
| 276 |
hier_club <- robust(hier_meta, cluster = hierdat$studyid, adjust = FALSE, clubSandwich = TRUE) |
| 277 |
|
| 278 |
expect_equal(residuals_CS(hier_meta), residuals_CS(hier_club)) |
| 279 |
expect_equal(coef_CS(hier_meta), coef_CS(hier_club)) |
| 280 |
expect_equal(model_matrix(hier_meta), model_matrix(hier_club)) |
| 281 |
expect_equal(bread(hier_meta), bread(hier_club)) |
| 282 |
expect_equal(v_scale(hier_meta), v_scale(hier_club)) |
| 283 |
expect_equal(targetVariance(hier_meta, cluster = hierdat$studyid), |
| 284 |
targetVariance(hier_club, cluster = hierdat$studyid)) |
| 285 |
expect_equal(weightMatrix(hier_meta, cluster = hierdat$studyid), |
| 286 |
weightMatrix(hier_club, cluster = hierdat$studyid)) |
| 287 |
|
| 288 |
}) |
| 289 |
|
| 290 |
|
| 291 |
test_that("clubSandwich works with weights of zero.", { |
| 292 |
|
| 293 |
data("SATcoaching") |
| 294 |
n_SAT <- nrow(SATcoaching) |
| 295 |
SATcoaching$wt <- rpois(n_SAT, lambda = 0.8) |
| 296 |
table(SATcoaching$wt) |
| 297 |
|
| 298 |
rma_full <- rma.uni(yi = d, vi = V, weights = wt, |
| 299 |
mods = ~ year + test, |
| 300 |
data = SATcoaching, method = "FE") |
| 301 |
|
| 302 |
SAT_sub <- subset(SATcoaching, wt > 0) |
| 303 |
rma_sub <- rma.uni(yi = d, vi = V, weights = wt, |
| 304 |
mods = ~ year + test, |
| 305 |
data = SAT_sub, method = "FE") |
| 306 |
# Note that this only works for method = "FE" because |
| 307 |
# tau.sq estimators differ between rma_full and rma_sub |
| 308 |
|
| 309 |
CR_full <- lapply(CR_types, function(x) vcovCR(rma_full, cluster = SATcoaching$study, type = x)) |
| 310 |
CR_sub <- lapply(CR_types, function(x) vcovCR(rma_sub, cluster = SAT_sub$study, type = x)) |
| 311 |
expect_equal(CR_full, CR_sub, check.attributes = FALSE) |
| 312 |
|
| 313 |
test_full <- lapply(CR_types, function(x) coef_test(rma_full, vcov = x, cluster = SATcoaching$study, test = c("z","naive-t","Satterthwaite"), p_values = TRUE)) |
| 314 |
test_sub <- lapply(CR_types, function(x) coef_test(rma_sub, vcov = x, cluster = SAT_sub$study, test = c("z","naive-t","Satterthwaite"), p_values = TRUE)) |
| 315 |
expect_equal(test_full, test_sub, check.attributes = FALSE) |
| 316 |
|
| 317 |
dat_miss <- SATcoaching |
| 318 |
miss_indicator <- sample.int(n_SAT, size = round(n_SAT / 10)) |
| 319 |
dat_miss$year[miss_indicator] <- NA |
| 320 |
with(dat_miss, table(wt, is.na(year))) |
| 321 |
|
| 322 |
expect_warning( |
| 323 |
rma_dropped <- rma.uni(yi = d, vi = V, weights = wt, |
| 324 |
mods = ~ year + test, |
| 325 |
data = dat_miss, method = "FE") |
| 326 |
) |
| 327 |
dat_complete <- subset(dat_miss, !is.na(year)) |
| 328 |
rma_complete <- rma.uni(yi = d, vi = V, weights = wt, |
| 329 |
mods = ~ year + test, |
| 330 |
data = dat_complete, method = "FE") |
| 331 |
|
| 332 |
CR_drop <- lapply(CR_types, function(x) vcovCR(rma_dropped, cluster = dat_miss$study, type = x)) |
| 333 |
CR_complete <- lapply(CR_types, function(x) vcovCR(rma_complete, cluster = dat_complete$study, type = x)) |
| 334 |
expect_equal(CR_drop, CR_complete) |
| 335 |
|
| 336 |
test_drop <- lapply(CR_types, function(x) coef_test(rma_dropped, vcov = x, cluster = dat_miss$study, test = "All", p_values = FALSE)) |
| 337 |
test_complete <- lapply(CR_types, function(x) coef_test(rma_complete, vcov = x, cluster = dat_complete$study, test = "All", p_values = FALSE)) |
| 338 |
expect_equal(test_drop, test_complete) |
| 339 |
|
| 340 |
}) |