{"id":"W2120950549","doi":"10.1097/00001888-200410001-00002","title":"An Examination of the Appropriateness of Using a Common Peer Assessment Instrument to Assess Physician Skills across Specialties","year":2004,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":41,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Cronbach's alpha; Generalizability theory; Specialty; Medicine; Reliability (semiconductor); Family medicine; Psychometrics; Psychology; Clinical 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07829975,0.000501724,0.0005692933,0.002854924,0.0008639338,0.001492073,0.001040012,0.001331766,0.0008985696],"category_scores_gemma":[0.33057,0.000401001,0.001316718,0.002261007,0.002041072,0.002155015,0.001972506,0.001202195,0.0002554951],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00143978,"about_ca_system_score_gemma":0.001700404,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001694234,"about_ca_topic_score_gemma":0.002022711,"domain_scores_codex":[0.9087552,0.0522232,0.01348321,0.003840015,0.02047231,0.001226157],"domain_scores_gemma":[0.6146184,0.2842169,0.0333677,0.01324218,0.05186789,0.002686915],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006172028,0.0003630752,0.940147,0.0003730728,0.0003495592,0.0001958233,0.003980868,0.000540729,0.002092488,0.0004243982,0.0006442346,0.0502716],"study_design_scores_gemma":[0.000106593,0.002815295,0.977796,0.0003330743,0.0002175863,0.0008684053,0.004105538,0.008885193,0.002329624,0.0005644713,0.001922791,0.00005538954],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9889898,0.0003460526,0.006694562,0.0006035903,0.0001083881,0.0004703961,0.0001068079,0.00004244486,0.00263797],"genre_scores_gemma":[0.9915215,0.0001037152,0.007670803,0.0001277412,0.00004020292,0.0002885611,0.000112474,0.00001350256,0.0001215392],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9217002,"threshold_uncertainty_score":0.4140936,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.06326721913361225,"score_gpt":0.4338719509033059,"score_spread":0.3706047317696937,"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."}}