{"id":"W2138928670","doi":"10.1177/0163278711425041","title":"A Comparison of Empirical Ranking Methods of Frequency and Severity Ratings of Clinical Presentations","year":2011,"lang":"en","type":"article","venue":"Evaluation & the Health Professions","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"Health Sciences Centre; University of Calgary","funders":"","keywords":"Ranking (information retrieval); Statistics; Psychology; Medicine; MEDLINE; Econometrics; Clinical psychology; 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":["metaresearch"],"category_scores_codex":[0.08238816,0.00008942387,0.0006088652,0.0001094065,0.0003391082,0.000007384873,0.0004087214,0.0000880565,0.00059499],"category_scores_gemma":[0.02009672,0.000050598,0.0001259513,0.0005831905,0.0003834685,0.0001432832,0.0001473038,0.0002951691,0.00000336043],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000206596,"about_ca_system_score_gemma":0.0008566431,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002120335,"about_ca_topic_score_gemma":0.00007918526,"domain_scores_codex":[0.9766811,0.01641747,0.003877875,0.0003617888,0.002463466,0.0001983604],"domain_scores_gemma":[0.988844,0.006629894,0.002318781,0.0006193613,0.001490537,0.00009740658],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00006981074,0.0005878839,0.8810493,0.00007019439,0.00002403424,1.474558e-8,0.03646154,0.00003064075,0.0003465058,0.00121115,0.002422912,0.07772598],"study_design_scores_gemma":[0.0005698503,0.00030614,0.8961906,0.0001321593,0.00004996523,2.315391e-7,0.01120301,0.01210219,0.0008232993,0.07843019,0.000141034,0.00005127185],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.942885,0.0005181855,0.04500188,0.007409013,0.0006276708,0.001835152,0.00001205843,0.000008973667,0.001702073],"genre_scores_gemma":[0.9348609,0.00002567408,0.06479463,0.0001925345,0.0000228074,0.00005347527,0.000002431301,0.000003529362,0.00004397219],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.07767472,"threshold_uncertainty_score":0.9881575,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.893414992052289,"score_gpt":0.7377671514789872,"score_spread":0.1556478405733018,"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."}}