1[2 {"id":"C1","text":"CR0 with a normal critical value undercovers in every cell and fails (coverage < 0.90) at G = 5 and G = 10 for all ICCs.","results_keys":["coverage","failure_cells"],"status":"supported","caveat":"Gaussian DGP, one cluster-size pattern."},3 {"id":"C2","text":"11 of 27 (procedure, cell) pairs are failure regimes; all are at G = 5 or G = 10, none at G = 30.","results_keys":["failure_cells","n_failure_cells"],"status":"supported","caveat":"CR1 G=5 classifications are within ~3 MC SE of the 0.90 threshold."},4 {"id":"C3","text":"CR0 and CR1 variance estimators are biased downward; at G = 5 with ICC > 0, CR0 recovers about 50% and CR1 about 63% of the exact Var(ybar).","results_keys":["rel_bias_var"],"status":"supported","caveat":"MC SE of relative bias about 0.013-0.016; confirmed by a closed-form expectation."},5 {"id":"C4","text":"CR1 has the highest average coverage over the G = 5 cells (0.894), ahead of WCB (0.856) and CR0 (0.747).","results_keys":["best_procedure_small_G","coverage"],"status":"supported","caveat":"WCB at G=5 uses Rademacher weights with only 32 distinct draws; other weight schemes may perform differently."},6 {"id":"C5","text":"Only 5 pairs are consistent with nominal 95% coverage: CR1 at G=30 (ICC 0, 0.2) and WCB at G=30 (all ICCs).","results_keys":["nominal_cells","mc_se_coverage"],"status":"supported","caveat":"The criterion is a 2-MC-SE band; failing to reject nominal coverage is not proof of exact coverage."},7 {"id":"C6","text":"At G = 10 with ICC > 0, WCB is closer to nominal than CR1 (about 0.935 vs 0.917).","results_keys":["coverage","mc_se_coverage"],"status":"supported","caveat":"The difference is about 2-3 marginal MC SEs; the paired design makes it more precise than the marginal SEs suggest, but no formal paired test is reported."}8]9