{"id":"W2606226254","doi":"10.1097/acm.0000000000001669","title":"The Hidden Value of Narrative Comments for Assessment: A Quantitative Reliability Analysis of Qualitative Data","year":2017,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":122,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; The Wilson Centre","funders":"","keywords":"Generalizability theory; Reliability (semiconductor); Psychology; Narrative; Set (abstract data type); Argument (complex analysis); Medical education; Medicine; Family medicine; Computer science; Internal medicine; Developmental psychology","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2341258,0.0008202736,0.001002046,0.005685174,0.002422353,0.00305884,0.001498215,0.0006712523,0.0009765574],"category_scores_gemma":[0.5376619,0.0007178625,0.001343671,0.003700969,0.006218831,0.003021554,0.004092351,0.001160262,0.0002873276],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004714085,"about_ca_system_score_gemma":0.003974938,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001735999,"about_ca_topic_score_gemma":0.001859037,"domain_scores_codex":[0.6684991,0.2335012,0.02825682,0.0107762,0.05664648,0.002320229],"domain_scores_gemma":[0.2464289,0.5858145,0.0457313,0.02725602,0.09336016,0.001409095],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"observational","study_design_scores_codex":[0.00181987,0.0004273099,0.3518069,0.003742959,0.00101267,0.0003719365,0.4768886,0.001577007,0.008509248,0.006487722,0.002668266,0.1446875],"study_design_scores_gemma":[0.0003902963,0.005134203,0.5772001,0.007900923,0.001426167,0.001560736,0.2954272,0.03874947,0.02674329,0.01835306,0.02632542,0.0007891257],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9236074,0.000503707,0.06390198,0.0006709022,0.0001985511,0.004321462,0.0004741278,0.0001434592,0.006178448],"genre_scores_gemma":[0.9764811,0.00008850639,0.01946563,0.0001733469,0.00004321268,0.003232487,0.0001813264,0.00006214371,0.0002722448],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7658742,"threshold_uncertainty_score":0.9444592,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3000704535187511,"score_gpt":0.6009780716740539,"score_spread":0.3009076181553028,"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."}}