{"id":"W3208344337","doi":"10.1097/acm.0000000000004282","title":"A Question of Scale? Comparison of Generalizability in Ottawa and Chen Scales When Used to Formulate Ad Hoc Entrustment Decisions for the Core EPAs","year":2021,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Generalizability theory; Scale (ratio); Formative assessment; Reliability (semiconductor); Medical education; Psychology; Medicine; Mathematics education; Geography","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.123495,0.0007783615,0.001080991,0.004576157,0.001428431,0.002610655,0.002049957,0.001794081,0.002397012],"category_scores_gemma":[0.3249109,0.0008512788,0.003864771,0.004126066,0.003689762,0.003677079,0.003480922,0.001590737,0.0007054324],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002626343,"about_ca_system_score_gemma":0.001517131,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00442346,"about_ca_topic_score_gemma":0.007195344,"domain_scores_codex":[0.8943374,0.06102763,0.01520663,0.005491324,0.0225732,0.001363814],"domain_scores_gemma":[0.6193603,0.296114,0.02725073,0.02392338,0.03106968,0.002282024],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003011123,0.0004359254,0.8637378,0.0005737716,0.001743954,0.0001426496,0.01835353,0.001685229,0.000704576,0.002855644,0.004241175,0.1025147],"study_design_scores_gemma":[0.0007121883,0.003262608,0.9486477,0.0008451922,0.0004646328,0.0004037841,0.01306119,0.01437071,0.001466426,0.004338297,0.01215972,0.0002676037],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9593097,0.00183949,0.01691999,0.001421889,0.0007835972,0.003934773,0.00117222,0.000191564,0.01442683],"genre_scores_gemma":[0.9833946,0.0003332509,0.01085352,0.0004630818,0.0001728032,0.003361851,0.0006860131,0.00009309703,0.0006417229],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.876505,"threshold_uncertainty_score":0.6531119,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09501778999012343,"score_gpt":0.4370391930029333,"score_spread":0.3420214030128099,"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."}}