{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02477504,0.001641143,0.001771437,0.002318656,0.0007834369,0.002083973,0.001217757,0.001533337,0.004656591],"category_scores_gemma":[0.09194066,0.000654955,0.001489243,0.001507591,0.0008673647,0.002091367,0.001612572,0.001309941,0.002804873],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005502166,"about_ca_system_score_gemma":0.001273194,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00294394,"about_ca_topic_score_gemma":0.003944629,"domain_scores_codex":[0.9811476,0.01004727,0.001889201,0.002806783,0.003459046,0.0006501811],"domain_scores_gemma":[0.9234045,0.05459549,0.006087605,0.005880704,0.008133998,0.001897748],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008829908,0.001311754,0.5652688,0.001166707,0.001398823,0.001030175,0.004090509,0.01498537,0.02622302,0.002973807,0.01401599,0.3587052],"study_design_scores_gemma":[0.0005640925,0.006697451,0.5920144,0.0006295079,0.0009421539,0.004017978,0.002910102,0.2875109,0.07258778,0.008994949,0.02258573,0.000544895],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8525929,0.001204675,0.1232886,0.0005530255,0.0004545785,0.0007577099,0.002012131,0.003553737,0.01558253],"genre_scores_gemma":[0.9213829,0.0002593753,0.07272808,0.0002804529,0.00003970414,0.0007300063,0.001404329,0.0004083549,0.002766829],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02477504,"threshold_uncertainty_score":0.1310245,"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."}}