{"id":"W2984746914","doi":"10.1371/journal.pone.0224675","title":"Institutional differences in USMLE Step 1 and 2 CK performance: Cross-sectional study of 89 US allopathic medical schools","year":2019,"lang":"en","type":"article","venue":"PLoS ONE","topic":"Innovations in Medical Education","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"York University","keywords":"United States Medical Licensing Examination; Licensure; Medicine; Family medicine; Observational study; Standardized test; Medical education; Medical school; Psychology; Internal medicine; Mathematics education","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002339536,0.0002782306,0.0003149654,0.001307724,0.0005646936,0.0007824297,0.0005848194,0.000492343,0.001556435],"category_scores_gemma":[0.005010432,0.0005483222,0.0004233713,0.001791621,0.0005383429,0.0008982816,0.001262471,0.0006473023,0.0004572225],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007106813,"about_ca_system_score_gemma":0.0007374258,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01010024,"about_ca_topic_score_gemma":0.01565806,"domain_scores_codex":[0.9984854,0.0003978692,0.00022594,0.0003177743,0.0003413455,0.0002317656],"domain_scores_gemma":[0.9944553,0.0009900086,0.002800504,0.0004054868,0.0006369837,0.0007118435],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001026004,0.00003019863,0.9996396,0.000001548746,0.000009849804,0.000005318397,0.00005937348,0.000007916895,0.00002837037,0.000003369841,0.00002318726,0.0001808648],"study_design_scores_gemma":[0.00000157741,0.00004913336,0.9995449,0.000002312128,0.000007612381,0.00002616525,0.0002358559,0.00005108594,0.00003325167,0.000002682131,0.00004390673,0.000001542176],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9995473,0.00003021013,0.00003967678,0.00001154298,0.00000135324,0.000007716005,0.0002439659,0.000002147655,0.0001161989],"genre_scores_gemma":[0.9994833,0.00002698997,0.00006335174,0.00001736192,0.000002387304,0.000009754121,0.0003315329,0.000001539969,0.00006374476],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01010024,"threshold_uncertainty_score":0.02008289,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.05028206486359744,"score_gpt":0.3132429748019663,"score_spread":0.2629609099383689,"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."}}