{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1757258,0.0006286231,0.0006838363,0.004875101,0.001761761,0.002454167,0.001999151,0.001260246,0.0009548053],"category_scores_gemma":[0.5340194,0.0005000977,0.001210702,0.00225809,0.004028382,0.002579726,0.002456891,0.00110724,0.0003692811],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002542394,"about_ca_system_score_gemma":0.003744737,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004626573,"about_ca_topic_score_gemma":0.005023644,"domain_scores_codex":[0.7751122,0.1676691,0.01038838,0.008682983,0.03636031,0.001787034],"domain_scores_gemma":[0.3563556,0.5288065,0.02563065,0.0329931,0.05430814,0.001906023],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002691828,0.0006596837,0.8036743,0.0005152093,0.001519985,0.0001242891,0.01908545,0.007299677,0.002115273,0.01546714,0.002162136,0.1446851],"study_design_scores_gemma":[0.001012134,0.003612671,0.7071221,0.001020191,0.0006800433,0.00106436,0.01616299,0.2214158,0.01379235,0.02662136,0.006979767,0.0005162624],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8373851,0.000391745,0.1529873,0.0004520475,0.0001690767,0.00122685,0.0001749376,0.0001739809,0.00703889],"genre_scores_gemma":[0.9595755,0.00005437357,0.03929709,0.00005950716,0.00003221868,0.0005573067,0.0001745653,0.00003920836,0.0002102925],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1757258,"threshold_uncertainty_score":0.9293379,"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."}}