{"id":"W2131795990","doi":"10.1097/acm.0000000000000147","title":"The Reliability, Validity, and Feasibility of Multisource Feedback Physician Assessment","year":2014,"lang":"en","type":"review","venue":"Academic Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":185,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Generalizability theory; PsycINFO; CINAHL; MEDLINE; Reliability (semiconductor); Construct validity; Psychology; Interpersonal communication; Medicine; Validity; Medical education; Psychometrics; Clinical psychology; Applied psychology; Family medicine; Nursing; Social 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":["metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.006522506,0.0004667977,0.002554253,0.0001966109,0.0001943177,0.000006409552,0.0004921393,0.0007700894,0.00005179076],"category_scores_gemma":[0.008004352,0.0002561515,0.0001899787,0.0008119461,0.001723339,0.00004768674,0.0002242643,0.002901934,0.00001643516],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003948531,"about_ca_system_score_gemma":0.0005586333,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000337822,"about_ca_topic_score_gemma":6.162788e-7,"domain_scores_codex":[0.9952365,0.0005672829,0.002059955,0.0006632078,0.001081619,0.0003914746],"domain_scores_gemma":[0.9949144,0.001932762,0.00124916,0.00130021,0.0004291462,0.0001743885],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001152398,0.00004162357,0.0005945605,0.0225317,0.00008042448,0.000001124444,0.0003116377,8.672917e-8,0.000003149408,0.001294162,0.02249623,0.9526338],"study_design_scores_gemma":[0.0007727261,0.0003301257,0.0007338446,0.02320817,0.00167721,0.00007245778,0.0003711332,0.0001205115,9.919607e-7,0.0009902968,0.9715238,0.0001987947],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.001427687,0.9854055,0.0005586394,0.006373548,0.001180789,0.002583345,0.000007718408,0.00005825675,0.002404529],"genre_scores_gemma":[0.001885202,0.9937044,0.0004415257,0.0009199186,0.00174244,0.0001339715,0.0001773691,0.00005846137,0.0009366549],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.952435,"threshold_uncertainty_score":0.9999891,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1259323570804783,"score_gpt":0.4862627857752188,"score_spread":0.3603304286947404,"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."}}