{"id":"W2461669997","doi":"10.1097/acm.0000000000000919","title":"Does Making the Numerical Values of Verbal Anchors on a Rating Scale Available to Examiners Inflate Scores on a Long Case Examination?","year":2015,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Rating scale; Scale (ratio); Cohort; Psychology; Medicine; Clinical psychology; Medical education; Developmental psychology; Internal medicine","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0878129,0.0004859311,0.0006292447,0.0008488573,0.000852243,0.002472144,0.001417283,0.001293404,0.002308795],"category_scores_gemma":[0.4582659,0.0006676959,0.0009106931,0.001187946,0.003148361,0.002577402,0.002033536,0.001320632,0.001188366],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001124582,"about_ca_system_score_gemma":0.001444821,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002406345,"about_ca_topic_score_gemma":0.004005459,"domain_scores_codex":[0.9078192,0.06032116,0.007616311,0.004381761,0.01871773,0.001143801],"domain_scores_gemma":[0.4241712,0.3693203,0.1208349,0.05203835,0.02976727,0.003868035],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004200351,0.0008495745,0.6827487,0.0008928211,0.0006107735,0.0002186136,0.007385252,0.0008127388,0.00702063,0.0009706531,0.004967749,0.2893221],"study_design_scores_gemma":[0.0003214718,0.005307799,0.9598635,0.001402662,0.0005283973,0.0009710079,0.003832751,0.002464281,0.01037029,0.001279381,0.0134505,0.0002079754],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9604561,0.003763984,0.01646005,0.005860451,0.001389729,0.0003615175,0.0002836479,0.000245459,0.01117909],"genre_scores_gemma":[0.9855648,0.0006726449,0.01122545,0.001102681,0.0003477439,0.000197254,0.0001449197,0.00006125979,0.0006831564],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0878129,"threshold_uncertainty_score":0.4644046,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05638001853164948,"score_gpt":0.3807799310911065,"score_spread":0.324399912559457,"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."}}