{"id":"W4402461068","doi":"10.7759/cureus.69151","title":"Item Analysis of Multiple-Choice Question (MCQ)-Based Exam Efficiency Among Postgraduate Pediatric Medical Students: An Observational, Cross-Sectional Study From Saudi Arabia","year":2024,"lang":"en","type":"article","venue":"Cureus","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Child, Adolescent and Family Mental Health","funders":"","keywords":"Medicine; Observational study; Cross-sectional study; Multiple choice; Family medicine; Medical education; Pathology; Internal medicine; Significant difference","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0158618,0.0002690843,0.0006377148,0.001621165,0.0002662229,0.0008831903,0.002078129,0.000217455,0.0009165799],"category_scores_gemma":[0.2177727,0.000204773,0.0003391707,0.0109611,0.0001759648,0.0006461728,0.0002682941,0.0003880883,0.00002922951],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001312216,"about_ca_system_score_gemma":0.0002722319,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003356615,"about_ca_topic_score_gemma":0.001293801,"domain_scores_codex":[0.9889446,0.001475329,0.001749387,0.001272543,0.006212908,0.0003452577],"domain_scores_gemma":[0.9116782,0.08566549,0.0005645357,0.0008907863,0.0009234003,0.000277585],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00002833242,0.0008336735,0.9923956,0.000007686498,0.000278626,0.00002645133,0.0002093569,0.004325379,0.00003238947,0.00004760219,0.000146157,0.001668709],"study_design_scores_gemma":[0.0004804088,0.0002084456,0.8688908,0.00001490527,0.0003287315,9.051824e-7,0.0001918067,0.1289691,0.000006164931,0.0006808615,0.00004330233,0.0001845739],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9779914,0.0009966397,0.01836577,0.00008714759,0.001832434,0.0003365345,0.000143542,0.0001789535,0.00006754838],"genre_scores_gemma":[0.9966038,0.00002082235,0.002590155,0.00007544275,0.0004586297,0.00003264013,0.0001088163,0.0000181424,0.00009148467],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2019109,"threshold_uncertainty_score":0.9999967,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.474351599505167,"score_gpt":0.5321652294323643,"score_spread":0.0578136299271973,"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."}}