{"id":"W2000644252","doi":"10.1007/s10459-014-9547-z","title":"Examiners and content and site: Oh My! A national organization’s investigation of score variation in large-scale performance assessments","year":2014,"lang":"en","type":"article","venue":"Advances in Health Sciences Education","topic":"Innovations in Medical Education","field":"Medicine","cited_by":35,"is_retracted":false,"has_abstract":false,"ca_institutions":"Medical Council of Canada; Queen's University","funders":"","keywords":"Variance (accounting); Rasch model; Construct (python library); Scale (ratio); Variation (astronomy); Multilevel model; Medical education; Psychology; Construct validity; United States Medical Licensing Examination; Explained variation; Applied psychology; Medicine; Psychometrics; Clinical psychology; Statistics; Computer science; Medical school; Geography; Accounting; Business; 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":[0.01952935,0.0002217119,0.00029607,0.002425916,0.001523817,0.001024284,0.0007737657,0.000490476,0.002026052],"category_scores_gemma":[0.08121557,0.0003731483,0.0004183402,0.002526303,0.0008697124,0.001050349,0.001871503,0.0007797442,0.0006917201],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001145715,"about_ca_system_score_gemma":0.001519689,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01662171,"about_ca_topic_score_gemma":0.03454418,"domain_scores_codex":[0.980008,0.01073108,0.002342155,0.00214643,0.003868185,0.0009040515],"domain_scores_gemma":[0.8979034,0.03741281,0.01635176,0.01293664,0.03247813,0.00291723],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000180782,0.000114594,0.9779674,0.00002580204,0.00007539832,0.00004865398,0.003280842,0.00005825875,0.0005669883,0.0001661688,0.003039594,0.01447557],"study_design_scores_gemma":[0.000008363651,0.000135377,0.995479,0.00001520737,0.00002097838,0.0000673032,0.002086545,0.0001450592,0.0003478531,0.0000572867,0.001627233,0.000009659684],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9876787,0.000193634,0.002452032,0.0008309468,0.0001060083,0.0001766274,0.001608509,0.00006761273,0.006886101],"genre_scores_gemma":[0.9960736,0.00003688506,0.001291019,0.0002811619,0.00002618742,0.0001297491,0.0006524912,0.00003514291,0.001473702],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9804707,"threshold_uncertainty_score":0.1032823,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02602832990857098,"score_gpt":0.3716996520777023,"score_spread":0.3456713221691313,"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."}}