{"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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01794747,0.0001849773,0.0008356856,0.0008104284,0.0001722545,0.00002732001,0.0004342799,0.0001388008,0.00007049614],"category_scores_gemma":[0.2805269,0.0001109518,0.00006770627,0.003998192,0.00006321575,0.0001770764,0.0001022631,0.0002920438,0.000004346129],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005489345,"about_ca_system_score_gemma":0.0002050232,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000365306,"about_ca_topic_score_gemma":0.00003296013,"domain_scores_codex":[0.9956262,0.0005960075,0.001590946,0.000745785,0.001126579,0.0003144865],"domain_scores_gemma":[0.9546409,0.04254633,0.0008993667,0.0005990874,0.001100372,0.0002139472],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002346434,0.00003029839,0.9037082,0.00001657904,0.0000905216,1.119407e-7,0.002678897,0.0006640982,0.00001796463,0.00002323386,0.004267483,0.0884792],"study_design_scores_gemma":[0.0009348918,0.0006980239,0.8613013,0.0001337892,0.0002635316,0.000001978806,0.03533772,0.09458928,6.125411e-7,0.001823798,0.004773611,0.0001414735],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9738386,0.002162973,0.009830921,0.01010768,0.0008533625,0.003004242,0.000003223434,0.00006577106,0.0001332182],"genre_scores_gemma":[0.988255,0.0001284863,0.004706444,0.001547685,0.0004141271,0.0001791715,0.000007870963,0.00001211139,0.004749074],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2625795,"threshold_uncertainty_score":0.7255335,"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."}}