{"id":"W4396508255","doi":"10.36834/cmej.77596","title":"The impact of systematically repairing multiple choice questions with low discrimination on assessment reliability: an interrupted time series analysis","year":2024,"lang":"en","type":"article","venue":"Canadian Medical Education Journal","topic":"Medical Education and Admissions","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary","funders":"","keywords":"Summative assessment; Reliability (semiconductor); Quality management; Consistency (knowledge bases); Regression analysis; Medicine; Multiple choice; Statistics; Psychology; Computer science; Significant difference; Mathematics; Internal medicine; Formative assessment; Operations management; Engineering; Artificial intelligence","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.001703633,0.0001432399,0.0002946049,0.0005418282,0.0003264938,0.0001545358,0.000174755,0.0001150234,0.01499079],"category_scores_gemma":[0.06387931,0.00007573343,0.000202089,0.000854803,0.0002088253,0.0001917062,0.00000781473,0.0007382709,0.00001766147],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009933886,"about_ca_system_score_gemma":0.07841537,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004929037,"about_ca_topic_score_gemma":0.006058647,"domain_scores_codex":[0.9977598,0.0003380763,0.0006014041,0.0002221738,0.0008206302,0.0002579582],"domain_scores_gemma":[0.9871972,0.0006467936,0.0001377864,0.0004076005,0.0006319113,0.0109787],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0006858835,0.009920262,0.1189539,0.004529322,0.01170458,0.0005315949,0.02389634,0.002789517,0.0008659104,0.02150332,0.4236955,0.3809239],"study_design_scores_gemma":[0.001349514,0.003280906,0.6237193,0.01758604,0.002952267,0.002165823,0.01135781,0.2931404,0.00003539097,0.0006441961,0.04314423,0.0006240871],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7328266,0.0004639399,0.003008381,0.255388,0.001721338,0.001023421,0.00003590187,0.0001304361,0.005401983],"genre_scores_gemma":[0.9945964,0.00004085342,0.00098037,0.001665398,0.0004341636,0.0000395897,0.00008829396,0.00001857148,0.00213632],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5047654,"threshold_uncertainty_score":0.9859096,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01210954641671757,"score_gpt":0.3767930397346411,"score_spread":0.3646834933179235,"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."}}