{"id":"W2060526915","doi":"10.1177/0163278702250082","title":"A Validity Study Of Expert Judgment Procedures For Setting Cutoff Scores On High-Stakes Credentialing Examinations Using Cluster Analysis","year":2003,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; Royal College of Physicians and Surgeons of Canada; Scanimetrics (Canada)","funders":"","keywords":"Cutoff; Licensure; Credentialing; Cluster (spacecraft); Categorization; Clinical judgment; Test (biology); Psychology; Medical physics; Statistics; Medicine; Computer science; Artificial intelligence; Medical education; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.007039423,0.0001514601,0.0004476477,0.0002477428,0.0009197303,0.00001972801,0.0000879876,0.00007744652,0.0001237825],"category_scores_gemma":[0.05547954,0.00009825001,0.0001505311,0.0006421781,0.00004608755,0.0000467998,0.00002920849,0.0002054401,0.000002339639],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002169346,"about_ca_system_score_gemma":0.001238441,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000330589,"about_ca_topic_score_gemma":0.0002751138,"domain_scores_codex":[0.9957862,0.001798196,0.0008665644,0.0003519753,0.000903458,0.0002935344],"domain_scores_gemma":[0.9874585,0.01058614,0.0006238053,0.0004856473,0.0006910925,0.0001547855],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003157023,0.04915012,0.4374299,0.001952075,0.005890951,0.000006628532,0.2515596,0.1555144,0.002042566,0.007061516,0.04126181,0.04497346],"study_design_scores_gemma":[0.02145904,0.00746136,0.6430835,0.01159138,0.01069771,0.000008987932,0.09268834,0.2071869,0.002023793,0.002765286,0.0003371076,0.0006966353],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9783287,0.0001981181,0.008767989,0.006943514,0.0004839996,0.005147293,0.00001751359,0.00003547583,0.00007737245],"genre_scores_gemma":[0.9937349,0.00002517626,0.003478941,0.001614707,0.0001319688,0.0007561183,0.00007217855,0.00001966764,0.0001663652],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2056535,"threshold_uncertainty_score":0.9524766,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2742225804547531,"score_gpt":0.5179446169931036,"score_spread":0.2437220365383505,"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."}}