{"id":"W1979809802","doi":"10.1007/s10459-008-9129-z","title":"The effect of reused questions on repeat examinees","year":2008,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Radiology practices and education","field":"Medicine","cited_by":36,"is_retracted":false,"has_abstract":false,"ca_institutions":"Medical Council of Canada","funders":"","keywords":"Test (biology); Psychology; Multiple choice; Block (permutation group theory); Social psychology; Medicine; Medical education; Significant difference; Mathematics","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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.04955291,0.0008794708,0.001708689,0.002050761,0.001450152,0.001852893,0.002522077,0.004088035,0.007732892],"category_scores_gemma":[0.4934974,0.0007822232,0.003276812,0.001284081,0.001546317,0.003287083,0.00313261,0.003516539,0.0007881829],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001791267,"about_ca_system_score_gemma":0.00295719,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01012876,"about_ca_topic_score_gemma":0.01152482,"domain_scores_codex":[0.9342001,0.04380625,0.004628464,0.005206712,0.01001796,0.00214052],"domain_scores_gemma":[0.2074524,0.7301222,0.02324972,0.02134527,0.01277505,0.005055296],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.07264953,0.009170974,0.7323942,0.001068403,0.002948366,0.001082085,0.01256387,0.001876217,0.004036857,0.0005746558,0.002668277,0.1589666],"study_design_scores_gemma":[0.0008256841,0.01622321,0.9684092,0.0005081471,0.001661259,0.0008500649,0.003540907,0.002015848,0.002784192,0.0005273463,0.002550419,0.0001037392],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9944748,0.0009594057,0.0008040114,0.0005712596,0.0001436376,0.0001917845,0.0002339831,0.00008135954,0.002539622],"genre_scores_gemma":[0.994303,0.0003415995,0.001568507,0.0004646808,0.0001036858,0.0002101437,0.0002593127,0.00008787755,0.002661117],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9504471,"threshold_uncertainty_score":0.2620639,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0248753314000006,"score_gpt":0.4252534904192208,"score_spread":0.4003781590192202,"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."}}