{"id":"W1994303059","doi":"10.1080/00220970309602065","title":"Pairwise Multiple Comparisons: New Yardstick, New Results","year":2003,"lang":"en","type":"article","venue":"The Journal of Experimental Education","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":19,"is_retracted":false,"has_abstract":true,"ca_institutions":"York University","funders":"","keywords":"Bonferroni correction; Multiple comparisons problem; False discovery rate; Statistics; Pairwise comparison; Word error rate; Type I and type II errors; Computer science; Mathematics; Artificial intelligence; Biology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003277078,0.0001504584,0.0004012947,0.00005865136,0.00009118356,0.00003694886,0.0003074801,0.00007325236,0.0003723752],"category_scores_gemma":[0.05842985,0.00009917243,0.0001347407,0.000152883,0.00008150721,0.0001109841,0.00002749082,0.000302983,0.00003825319],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001423933,"about_ca_system_score_gemma":0.0009129466,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000084901,"about_ca_topic_score_gemma":0.000003447668,"domain_scores_codex":[0.9969847,0.001063368,0.001246615,0.0001176323,0.0004051283,0.0001825999],"domain_scores_gemma":[0.9822792,0.01601603,0.0009168619,0.0003541022,0.0001321414,0.0003016949],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001480835,0.001931797,0.0002514158,0.0000165598,0.0001261564,0.000002035597,0.004872892,0.00001781444,0.01324609,0.05842538,0.911348,0.008281017],"study_design_scores_gemma":[0.005151952,0.0008200135,0.0005331436,0.0002754017,0.000264841,0.0001526441,0.008985328,0.00006402931,0.1495995,0.807889,0.02596706,0.0002971235],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3668937,0.01160846,0.539537,0.01361531,0.02398842,0.002692967,0.00003619051,0.0001415101,0.04148644],"genre_scores_gemma":[0.3654971,0.00003907985,0.6316463,0.0002493696,0.0008116824,0.000002370932,5.123666e-7,0.00002028763,0.001733354],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8853809,"threshold_uncertainty_score":0.9495014,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5586051573696436,"score_gpt":0.5803779345219491,"score_spread":0.02177277715230552,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}