{"id":"W2998758675","doi":"10.1016/j.amjsurg.2020.01.019","title":"The impact of rater training on the psychometric properties of standardized surgical skill assessment tools","year":2020,"lang":"en","type":"article","venue":"The American Journal of Surgery","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Manitoba","funders":"","keywords":"Checklist; Construct validity; Knot tying; Reliability (semiconductor); Inter-rater reliability; Internal consistency; Psychology; Rating scale; Medicine; Psychometrics; Medical physics; Clinical psychology; Surgery; Developmental psychology","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":[],"consensus_categories":[],"category_scores_codex":[0.002771862,0.0001308473,0.0008069816,0.0001025775,0.00009935982,0.00003186371,0.0001694478,0.00002243475,0.0001532706],"category_scores_gemma":[0.001202702,0.00004358447,0.0006706577,0.0009542989,0.0004964581,0.00005900142,0.00001991965,0.0004647514,0.000001340081],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004362738,"about_ca_system_score_gemma":0.0003521155,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000008632152,"about_ca_topic_score_gemma":4.802493e-8,"domain_scores_codex":[0.9977531,0.0004926534,0.0007743617,0.00008485634,0.0006879013,0.000207128],"domain_scores_gemma":[0.9934488,0.004960197,0.0009841339,0.0001878844,0.0002775477,0.0001414586],"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":[0.03519187,0.0005535953,0.11869,0.0000567768,0.003479471,0.000185013,0.008865128,0.003653185,0.006165775,0.0004810097,0.001639113,0.8210391],"study_design_scores_gemma":[0.006429205,0.00400305,0.9539538,0.001025755,0.0003658566,0.0004013519,0.01997376,0.001817939,0.003879081,0.0001217876,0.007689468,0.000338928],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.989807,0.0002950484,0.00005105462,0.008660894,0.00008132645,0.000120513,0.000005351942,0.000006584167,0.0009722339],"genre_scores_gemma":[0.9989643,0.0003426869,0.0000269652,0.0004634686,0.0001741004,0.000001993899,5.305839e-7,0.00001457109,0.00001144844],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8352638,"threshold_uncertainty_score":0.2019139,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1798463362986748,"score_gpt":0.3749284874325116,"score_spread":0.1950821511338368,"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."}}