{"id":"W2793261989","doi":"10.1080/10401334.2017.1414609","title":"Quality Evaluation Scores are no more Reliable than Gestalt in Evaluating the Quality of Emergency Medicine Blogs: A METRIQ Study","year":2018,"lang":"en","type":"article","venue":"Teaching and Learning in Medicine","topic":"Social Media in Health Education","field":"Social Sciences","cited_by":29,"is_retracted":false,"has_abstract":true,"ca_institutions":"Queen's University; McMaster University; Western University; University of Alberta; University of Saskatchewan","funders":"Royal College of Physicians and Surgeons of Canada; Canadian Association of Emergency Physicians","keywords":"Gestalt psychology; Intraclass correlation; Reliability (semiconductor); Quality (philosophy); Metric (unit); Medicine; Emergency department; Psychology; Medical education; Clinical psychology; Nursing; Psychometrics","routes":{"ca_aff":true,"ca_fund":true,"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":[0.1270833,0.000148712,0.0005289284,0.0002949413,0.001078921,0.00000760925,0.0002540767,0.0001140334,0.0003166838],"category_scores_gemma":[0.2812271,0.0001057888,0.00002437702,0.0009860428,0.0007075016,0.0001267273,0.00004509318,0.001293548,0.000003208166],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002704779,"about_ca_system_score_gemma":0.0002884835,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.3100207,"about_ca_topic_score_gemma":0.03293718,"domain_scores_codex":[0.980629,0.01545544,0.001273374,0.0004001872,0.001856719,0.0003852631],"domain_scores_gemma":[0.9900616,0.00796853,0.0009720996,0.0002865054,0.000601902,0.0001093693],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000423051,0.00008663116,0.6290057,0.0000562119,0.000005975286,1.270844e-7,0.3568515,0.00002976907,0.00009330107,0.000153437,0.00009703271,0.01357804],"study_design_scores_gemma":[0.0008330667,0.0006495032,0.7138724,0.0004272125,0.00002765877,7.725145e-8,0.282458,0.0008853328,5.93622e-7,0.0004489094,0.0003204084,0.00007672885],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9826159,0.001099009,0.00001458043,0.009297495,0.002464408,0.001598903,3.239773e-7,0.00004686335,0.002862491],"genre_scores_gemma":[0.9972283,0.0001455945,0.0001602517,0.00009194045,0.001954808,0.0002156053,0.000005977903,0.00001286359,0.0001846325],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2770835,"threshold_uncertainty_score":0.9847092,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3946124905160667,"score_gpt":0.5938683588428553,"score_spread":0.1992558683267886,"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."}}