{"id":"W4396530533","doi":"10.36834/cmej.77580","title":"‘What would my peers say?’ Comparing the opinion-based method with the prediction-based method in Continuing Medical Education course evaluation","year":2024,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Likert scale; Scale (ratio); Computer science; Continuing medical education; Point (geometry); Statistics; Medical education; Psychology; Machine learning; Medicine; Mathematics; Continuing education","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1131689,0.001085675,0.001549943,0.003442199,0.0008621287,0.001758734,0.001594484,0.001524196,0.005246979],"category_scores_gemma":[0.2581727,0.0005191715,0.001785471,0.002515828,0.001213987,0.002835192,0.002449134,0.001338672,0.0009250054],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001281843,"about_ca_system_score_gemma":0.00115735,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001780679,"about_ca_topic_score_gemma":0.003215807,"domain_scores_codex":[0.786839,0.1783317,0.009737137,0.007655791,0.0165476,0.0008887884],"domain_scores_gemma":[0.5128365,0.4265361,0.01482522,0.01196372,0.03170911,0.002129391],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.02187304,0.00340962,0.2528736,0.002641684,0.002576311,0.000216531,0.00790099,0.005229619,0.004560227,0.003216396,0.005217534,0.6902845],"study_design_scores_gemma":[0.008646357,0.03349167,0.5549492,0.001237083,0.002991721,0.001559672,0.006234244,0.3490465,0.01628483,0.01012765,0.01426336,0.001167779],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7366656,0.0008805249,0.2405189,0.0006573346,0.0008259579,0.009021224,0.0006276923,0.0006388768,0.01016399],"genre_scores_gemma":[0.7529922,0.0003021952,0.2344715,0.000445958,0.0002394234,0.00971586,0.0004487618,0.0001582035,0.001226035],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1131689,"threshold_uncertainty_score":0.5985017,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02446761684646262,"score_gpt":0.4104752178828711,"score_spread":0.3860076010364085,"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."}}