{"id":"W4294351388","doi":"10.36834/cmej.74860","title":"Validity evidence for the Quality of Assessment for Learning score: a quality metric for supervisor comments in Competency Based Medical Education","year":2022,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University; Impact; University of Ottawa; University of Calgary; University of Saskatchewan","funders":"McMaster University; University of Saskatchewan","keywords":"Generalizability theory; Competence (human resources); Supervisor; Cronbach's alpha; Narrative; Psychology; Medical education; Variance (accounting); Medicine; Clinical psychology; Social psychology; Psychometrics; Management; Accounting; Developmental psychology","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"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":[],"category_scores_codex":[0.2474886,0.0006049428,0.001558009,0.007713587,0.001775192,0.003525696,0.002196092,0.002019913,0.002374024],"category_scores_gemma":[0.5477147,0.0005632033,0.003391464,0.006205889,0.005357759,0.004012183,0.004171124,0.001853348,0.0004701449],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005380675,"about_ca_system_score_gemma":0.00634408,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003469965,"about_ca_topic_score_gemma":0.004543491,"domain_scores_codex":[0.6667301,0.1842732,0.04319267,0.01049376,0.0932456,0.002064657],"domain_scores_gemma":[0.2439785,0.5538899,0.09345819,0.02157699,0.0837672,0.003329295],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001383358,0.0004064682,0.8820864,0.003405401,0.002224248,0.00005255105,0.003870868,0.00111958,0.0004070262,0.004046188,0.003136133,0.09786174],"study_design_scores_gemma":[0.0008045179,0.003658778,0.9280753,0.007886439,0.001728559,0.0005009826,0.005509684,0.0180236,0.002902142,0.01246405,0.01815539,0.0002907012],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7659529,0.01874829,0.1260827,0.01075953,0.001884856,0.009522355,0.005819356,0.000286329,0.06094364],"genre_scores_gemma":[0.9619758,0.0009488991,0.03091256,0.000861994,0.0002226766,0.003225129,0.001238383,0.00007475102,0.000539735],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2474886,"threshold_uncertainty_score":0.9279805,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1675569394145632,"score_gpt":0.4977005813903227,"score_spread":0.3301436419757596,"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."}}