{"id":"W2962737799","doi":"10.1097/acm.0000000000002888","title":"Examinee Cohort Size and Item Analysis Guidelines for Health Professions Education Programs: A Monte Carlo Simulation Study","year":2019,"lang":"en","type":"article","venue":"Academic Medicine","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":10,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia; McGill University; Office of the Chief Medical Examiner; Université du Québec à Montréal; Université de Sherbrooke","funders":"","keywords":"Cohort; Reliability (semiconductor); Medicine; Cohort study; Monte Carlo method; Control (management); Medical education; Statistics; Family medicine; Psychology; Computer science; Mathematics; Internal medicine","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1037892,0.000804905,0.00104664,0.0008347586,0.001364057,0.002298923,0.002128246,0.002007204,0.002830459],"category_scores_gemma":[0.3069947,0.0009382819,0.002541008,0.001008441,0.001589302,0.001933552,0.001605999,0.003421779,0.0002939266],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005270703,"about_ca_system_score_gemma":0.005901564,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02524479,"about_ca_topic_score_gemma":0.01830036,"domain_scores_codex":[0.9190996,0.06959369,0.001686384,0.003798802,0.004420715,0.001400808],"domain_scores_gemma":[0.4313642,0.5250194,0.01931553,0.01487256,0.007614249,0.001814115],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0130677,0.007745873,0.6156855,0.0005498581,0.003136691,0.0006068553,0.004048118,0.2479718,0.002282999,0.01663178,0.004007679,0.0842651],"study_design_scores_gemma":[0.00327835,0.01491945,0.2049787,0.0004166043,0.002177012,0.0003196563,0.001634342,0.7551696,0.004163728,0.008933213,0.003727934,0.000281301],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9546943,0.0002562827,0.0406011,0.00051504,0.0000357695,0.0017776,0.000343614,0.0001005499,0.001675678],"genre_scores_gemma":[0.9700693,0.00007706808,0.02725217,0.0002086263,0.000009664894,0.001628974,0.0001956335,0.00002410809,0.0005345289],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1037892,"threshold_uncertainty_score":0.5488966,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6505779948723512,"score_gpt":0.6261768352554835,"score_spread":0.02440115961686773,"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."}}