{"id":"W4376626995","doi":"10.58379/rshg8366","title":"DIF investigations across groups of gender and academic background in a large-scale high-stakes language test ","year":2015,"lang":"en","type":"article","venue":"Studies in Language Assessment","topic":"Student Assessment and Feedback","field":"Social Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"Queen's University","keywords":"Test (biology); Differential item functioning; Psychology; Quality (philosophy); Reliability (semiconductor); Scale (ratio); Gender bias; Social psychology; Applied psychology; Mathematics education; Medical education; Item response theory; Developmental psychology; Psychometrics; Medicine","routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001939991,0.0001865303,0.0004069052,0.0001291336,0.0001860482,0.00004365888,0.0002733563,0.0001244726,0.00003273934],"category_scores_gemma":[0.0004115948,0.0001718062,0.00003904311,0.0006056115,0.0006344239,0.0003222238,0.000325607,0.0003399829,0.000004799842],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003683697,"about_ca_system_score_gemma":0.0001569696,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001905362,"about_ca_topic_score_gemma":0.01828412,"domain_scores_codex":[0.9978556,0.0002315049,0.0004398239,0.0003319576,0.0005921518,0.00054896],"domain_scores_gemma":[0.9989382,0.0004792823,0.0001597371,0.0002010874,0.00009252957,0.0001291441],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"qualitative","study_design_scores_codex":[0.000005623615,0.0001455537,0.5478673,0.00003936946,0.00003510079,0.00003390112,0.4467611,0.000006284921,0.0003610139,0.004169225,0.0003469616,0.0002285759],"study_design_scores_gemma":[0.001042191,0.00004412037,0.2873902,0.00006271413,0.00001531911,6.906183e-7,0.7103823,0.00002939225,0.00003476147,0.0006744898,0.000172709,0.0001511045],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9863673,0.00474632,0.00004191389,0.0004094843,0.0003111734,0.0004164432,0.0000496051,0.00004042405,0.007617312],"genre_scores_gemma":[0.9959855,0.0006659263,0.002341677,0.0001463452,0.0001869852,0.00009009223,0.00001760981,0.00001535267,0.0005505265],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2636212,"threshold_uncertainty_score":0.9996296,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.119758425185111,"score_gpt":0.46502313377722,"score_spread":0.3452647085921091,"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."}}