{"id":"W2468688964","doi":"10.1177/2158244016653791","title":"Differential Performance on National Exams","year":2016,"lang":"en","type":"article","venue":"SAGE Open","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Differential item functioning; Test (biology); Psychology; Certificate; Item response theory; Subject (documents); Social psychology; Psychometrics; Clinical psychology; Mathematics; Computer science","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":[],"consensus_categories":[],"category_scores_codex":[0.02356843,0.0002602944,0.0005887336,0.002133059,0.0007738187,0.00152988,0.0007633578,0.0003963836,0.00234133],"category_scores_gemma":[0.1012652,0.0001287111,0.0004103379,0.001539933,0.001171745,0.001390482,0.00311785,0.000550567,0.0004453734],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001136009,"about_ca_system_score_gemma":0.001042132,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00248286,"about_ca_topic_score_gemma":0.004240077,"domain_scores_codex":[0.9763581,0.01063003,0.001968981,0.002404504,0.007345931,0.001292481],"domain_scores_gemma":[0.9235983,0.03582906,0.01527234,0.01028821,0.0129497,0.002062411],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0002429428,0.0001605383,0.9559525,0.00003525819,0.00005553736,0.00006503467,0.002603131,0.0003762118,0.0007267218,0.001553846,0.000212448,0.03801595],"study_design_scores_gemma":[0.00001003908,0.0003804171,0.9913399,0.0000335614,0.0000273527,0.0002197515,0.002226131,0.001028187,0.002235904,0.001230226,0.001251606,0.00001697971],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9914407,0.00009639392,0.001880775,0.0001121043,0.0000164945,0.00004304541,0.00008874595,0.00001459317,0.00630712],"genre_scores_gemma":[0.9983133,0.00002552102,0.0008859206,0.0000340738,0.000006222948,0.00002418813,0.00009527506,0.000005258967,0.0006102493],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02356843,"threshold_uncertainty_score":0.1246432,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6708101331326844,"score_gpt":0.5210443421341487,"score_spread":0.1497657909985357,"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."}}