{"id":"W4383904828","doi":"10.7759/cureus.41702","title":"The Pass/Fail Effect: A Longitudinal Study of United States Medical Licensing Examination (USMLE) Step 1 Performance Over a Decade","year":2023,"lang":"en","type":"article","venue":"Cureus","topic":"Innovations in Medical Education","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Medicine; United States Medical Licensing Examination; Descriptive statistics; Test (biology); Family medicine; Demography; Medical education; Medical school; Statistics","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":[0.004890453,0.0002174482,0.0003209714,0.001142068,0.0009625977,0.00106521,0.0006977817,0.0006468724,0.001616778],"category_scores_gemma":[0.01286993,0.0002847977,0.0005582229,0.001267637,0.0004730445,0.0009615569,0.001114415,0.001235673,0.0004511893],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009085573,"about_ca_system_score_gemma":0.002158981,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.07271954,"about_ca_topic_score_gemma":0.07411766,"domain_scores_codex":[0.9979943,0.0004747172,0.0001315773,0.000363161,0.0005608599,0.0004754076],"domain_scores_gemma":[0.9911302,0.001345732,0.003984038,0.0007544972,0.00144173,0.001343838],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000462049,0.00004994141,0.998127,0.000003805581,0.00002878962,0.00001934282,0.0002710761,0.00003309313,0.00004046331,0.00003635325,0.0001878437,0.00115607],"study_design_scores_gemma":[0.000001063,0.00006442005,0.9991325,0.000007384989,0.000008604426,0.00002290744,0.0003420837,0.00008461608,0.00003238091,0.00001371215,0.0002872529,0.000003160281],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9974262,0.0002878072,0.000208549,0.0002557012,0.0000194459,0.00001839309,0.0009779343,0.000008318722,0.0007974509],"genre_scores_gemma":[0.9983845,0.00009291661,0.00009656686,0.000061918,0.00001089345,0.00001800386,0.0007553458,0.000004521232,0.0005753019],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9951096,"threshold_uncertainty_score":0.1445925,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02612269425499893,"score_gpt":0.3535088670476444,"score_spread":0.3273861727926455,"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."}}