{"id":"W2053906918","doi":"10.1207/s15324818ame1703_2","title":"Performance of SIBTEST When the Percentage of DIF Items is Large","year":2004,"lang":"en","type":"article","venue":"Applied Measurement in Education","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":39,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Alberta","funders":"","keywords":"Differential item functioning; Statistics; Psychology; Item response theory; Test (biology); Mathematics; Psychometrics","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.008470052,0.00009000178,0.0002022913,0.0002767493,0.00006880549,0.0000233831,0.000544811,0.0000426277,0.0001138011],"category_scores_gemma":[0.004016101,0.00005734235,0.00004555983,0.0009164349,0.00005582198,0.00008708298,0.0000515973,0.000103167,0.00001811303],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000113517,"about_ca_system_score_gemma":0.0002257586,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001225965,"about_ca_topic_score_gemma":0.00002395923,"domain_scores_codex":[0.9974545,0.00008611271,0.0006589605,0.0002464346,0.001372667,0.000181364],"domain_scores_gemma":[0.9977362,0.0009921847,0.0004216503,0.0005337654,0.0002843705,0.00003189881],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0000604955,0.001300807,0.6753676,0.00006746031,0.00001909356,5.382973e-8,0.01170131,0.0003834654,0.04315415,0.005806043,0.002977504,0.259162],"study_design_scores_gemma":[0.0006949356,0.00008189143,0.9369794,0.00009999202,0.00001356206,0.000001068024,0.01124099,0.0001086081,0.02165337,0.02301773,0.005974601,0.0001338866],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9838824,0.0003156606,0.001091426,0.0005940866,0.0005073806,0.0003318789,0.000002686569,0.000007670061,0.01326678],"genre_scores_gemma":[0.9955154,0.00005325194,0.004076695,0.0002081125,0.0000530021,0.00003584725,0.000001453847,0.00000529286,0.0000509333],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2616118,"threshold_uncertainty_score":0.4807938,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3321836598049474,"score_gpt":0.4109344435671419,"score_spread":0.07875078376219452,"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."}}