{"id":"W1483774131","doi":"10.36834/cmej.36560","title":"Comparison of Student Performance on Internally Prepared Clerkship Examinations and NBME Subject Examinations","year":2011,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Innovations in Medical Education","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"McNemar's test; Medicine; Significant difference; Medical education; Internal medicine; Statistics; Mathematics","routes":{"ca_aff":true,"ca_fund":false,"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":[0.004177526,0.0003270395,0.0004199723,0.001247156,0.000344069,0.0008855985,0.0005092819,0.0004710457,0.002061488],"category_scores_gemma":[0.02550326,0.0001868565,0.0003491471,0.0007694348,0.0005902565,0.0004424768,0.001109044,0.0005942516,0.0006501994],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001249583,"about_ca_system_score_gemma":0.001062891,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004452694,"about_ca_topic_score_gemma":0.004823087,"domain_scores_codex":[0.9955217,0.001068682,0.0004252967,0.0004931,0.001802246,0.0006890116],"domain_scores_gemma":[0.9790876,0.007693422,0.006188191,0.001103018,0.003322478,0.002605314],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000321261,0.0001596041,0.9907837,0.00002456436,0.00004060688,0.00004422518,0.0006522065,0.00008968653,0.0006945332,0.00002120084,0.0001261684,0.007042263],"study_design_scores_gemma":[0.000005228368,0.0004508564,0.9986838,0.000004805132,0.00000679185,0.00005064032,0.0001923541,0.00007488152,0.0004200128,0.000006398804,0.0001007958,0.000003374777],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9994468,0.00003153659,0.00005745451,0.00001470387,0.000002797446,0.000007918932,0.00003166392,0.000003843988,0.0004033169],"genre_scores_gemma":[0.9996179,0.00001461799,0.00006410855,0.00001334982,0.00000297978,0.000006702171,0.000101166,0.000002867045,0.0001761862],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9958225,"threshold_uncertainty_score":0.02209312,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05273264572949835,"score_gpt":0.3821762039846095,"score_spread":0.3294435582551112,"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."}}