{"id":"W2028787956","doi":"10.1111/medu.12621","title":"A systematic review of validity evidence for checklists versus global rating scales in simulation‐based assessment","year":2015,"lang":"en","type":"review","venue":"Medical Education","topic":"Simulation-Based Education in Healthcare","field":"Medicine","cited_by":328,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; University of Calgary","funders":"","keywords":"Checklist; Context (archaeology); Confidence interval; Medicine; Reliability (semiconductor); MEDLINE; Systematic review; Validity; Psychometrics; Psychology; Clinical 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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1082056,0.002932586,0.0175244,0.01896425,0.001522098,0.006262984,0.00379661,0.003802037,0.003611643],"category_scores_gemma":[0.4008293,0.002547456,0.02225291,0.01684186,0.003080142,0.006272004,0.003503242,0.002865153,0.0003619648],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01032201,"about_ca_system_score_gemma":0.02304279,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00723664,"about_ca_topic_score_gemma":0.01818224,"domain_scores_codex":[0.8233465,0.093249,0.05418448,0.006804872,0.02142624,0.000988871],"domain_scores_gemma":[0.6091599,0.306471,0.05548964,0.006667633,0.02105034,0.001161366],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0006039249,0.00002780024,0.001976378,0.9252895,0.03757067,0.00006089982,0.0003534909,0.0001754063,0.0001100358,0.0004325581,0.0009733406,0.03242597],"study_design_scores_gemma":[0.001133813,0.0003350923,0.005282555,0.8581642,0.1278152,0.0002111572,0.0003637925,0.0003073456,0.0002008456,0.0008265292,0.005272947,0.00008666881],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.003058259,0.9915535,0.001389299,0.000784965,0.0004274164,0.001895002,0.0004355993,0.00002237228,0.0004335946],"genre_scores_gemma":[0.1139488,0.8591092,0.01216221,0.002386914,0.0004678926,0.01076603,0.0008942753,0.00005159629,0.000213136],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8917944,"threshold_uncertainty_score":0.5722528,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3337116927577552,"score_gpt":0.610282715440884,"score_spread":0.2765710226831288,"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."}}